---
title: "The Silicon Industrialists - Part 2: Tech’s Great Escape"

author: "Lida Liberopoulou"

date: "2026-06-19"

canonical: "https://threadbaire.com/blog/posts/the-silicon-industrialists-part-2.html"

license: "CC BY-SA 4.0"

description: "Every major infrastructure technology before software followed the same path: massive capital, consolidation, capture. The personal computer, independent software, and the open internet broke that pattern because specific institutional decisions opened the substrate before anyone could close it. Part 2 traces the escape and asks what held the ladder in place."
---



*The comparison most often used to reassure us about AI is the dot-com crash. The preceding boom was extraordinary, its crash was painful, but the technology was solid and what followed was one of the largest expansions of wealth creation in modern history. The implication is that AI's excesses will also self-correct, and what will emerge on the other side will be worth the ride.*

*But that reassurance rests on a misunderstanding of what made the recovery possible. In reality the software economy was an engineered escape from a pattern that had captured every major infrastructure technology before it.*

*Railroads, electrical grids, and the telephone system each followed the same sequence: massive capital deployed before demand, consolidation into fewer hands, and the owners of the foundational layer reaching for capture because their capital structures left no other direction. The personal computer, independent software, and the open internet broke that pattern for the first time putting everything needed to build a product on a desk, reducing the cost of each additional copy to nearly nothing, and opening distribution to anyone with a server and a connection.*

*The escape from this cycle didn't come from the technology becoming smaller, faster, or cheaper. This was the product of specific institutional decisions like a consent decree that opened monopolised science, an unbundling that separated software from hardware, and a conditioned privatisation that built the open network before anyone could close it. This article traces the pattern through three industries, follows the escape through three openings, and asks what held the ladder in place. The conditions that made the software economy possible were not permanent and the pattern they interrupted is assembling again, this time much faster, and aimed at the economy's centre.*

------

[Part 1](https://threadbaire.com/blog/posts/the-silicon-industrialists-part-1.html) of this series described two forces operating simultaneously on the AI economy.

The first is its industrial economics. AI's cost structure looks nothing like software but resembles more things like railroads, electrical grids, and oil fields. Like them it requires hundreds of billions in physical infrastructure, in the form of dedicated power connections, specialised chips, all of it built and paid for before the revenue arrives. The second is value dissolution. The better AI works, the more it undermines the pricing mechanisms of the customers who are supposed to pay for it. User seats mean less when agents do the work and implementation fees shrink when it becomes automated. And code stops being a durable asset and starts looking like a disposable output of intent.

Where those two forces collide, they produce a double bind. Industrial economics gates who can build. Only a handful of players have the money and setup to handle the massive hardware, software and specialized manpower requirements to build and maintain the fundamental elements of AI.  Value dissolution gates who can charge and what can be charged for it as more and more services previously handled by software platforms and the humans operating them are starting to be handled directly by AI. And together they push the most transformative technology in a generation toward regression and toward older, more extractive methods of capturing value, because the current capital structure leaves no other direction.

The double bind was traced through four domains. Infrastructure owners locked in circular dependencies with the labs they fund, each needing the other to keep growing, neither able to leave. A venture machine that raises capital with the language of civilisational transformation and deploys it into automating the same back-office workflows that have existed for decades. A startup ecosystem where frontier ambition resolves into familiar enterprise products because familiar is what the next round of investors know how to price. And an open layer whose independence is structurally impossible to sustain, because the industrial economics make every open project either collapse, get absorbed, pivot to enterprise, or become a demand generator for the companies it was supposed to be the alternative to.

[Part 1](https://threadbaire.com/blog/posts/the-silicon-industrialists-part-1.html) ended with a question: whether this was inevitable. Whether the only possible path for a groundbreaking technology was to be bound to 19th century economics, existing capital structures, and legacy capture mechanisms.

But before asking whether AI can escape the double bind, we first need to establish how the bind appears.

The research points to a recurring sequence of seven markers. A technology begins with shared infrastructure that no individual user can build alone. That infrastructure must be financed before the demand exists to pay for it. Scale then pushes the industry toward consolidation. The new technology dissolves the older economic arrangements around it. The old economy breaks faster than the new one can mature. The owners of the infrastructure reach for capture: tolls, exclusive access, financial engineering, managed transitions, and control over which new businesses are allowed to develop. And the capture, once established, suppresses the innovation that doesn't serve the debt structure. Any outside technologies, competing approaches, and new uses of the infrastructure are blocked, slowed, or starved.

The sequence does not describe every new technology. But it is unmistakably there in three of the industries that built the modern American economy: railroads, electricity, and telephony. Each passed through the same seven markers by a different route. Each created real progress but also produced a period in which the companies that owned the underlying infrastructure gained the power to shape, delay, or extract from the economy forming above it.

But something unusual began in the middle of the twentieth century. The conditions that would eventually break the pattern were being put in place.

The personal computer and the internet created a period in which a small team could have its own setup and build a product without financing a national infrastructure system. It could also copy that product at almost no cost, and distribute it to a global audience without asking a gatekeeper for permission. That escape created the software economy and one of the largest expansions of wealth creation in modern history.

This article asks how that escape happened, what institutional conditions made it possible, and why it was not simply the natural result of technology becoming smaller, faster, or cheaper. It also returns to the comparison now used most often to reassure us that AI’s distortions are merely the growing pains of another great technology boom: the dot-com crash. That crash destroyed capital, companies, and careers. But it did not recreate the older infrastructure trap, and understanding why is central to understanding what is different about AI now.

## The old bind

The double bind is not new. Every major infrastructure technology in recent history has passed through the same sequence. The mechanisms might differ each time but the structure does not. It has seven consecutive markers.

1. **The technology requires shared physical infrastructure**. It could be tracks, wires, grids, networks. Something no individual can build or own alone.
2. **The infrastructure must be financed before the demand exists to pay for it.** Capital needs to go in first. Revenue follows later, if it follows at all.
3. **Scale economics drive consolidation.** The large operator outcompetes the small one. The capital markets favour the established over the local. The industry moves from many providers to few.
4. **The technology dissolves the older economy it replaces.** The previous arrangements, jobs, and pricing structures break. This is produced by the technology's success not its failure. 
5. **The old economy breaks faster than the new one matures.** The debt is due but the new revenue is not ready. The new economy arrives first for the customers who are already advantaged, widening the gap between the served and the unserved.
6. **The owner of the infrastructure reaches for capture.** Tolls, rates, exclusive access, financial engineering, managed transitions. The capital structure requires controlled revenue, and controlled revenue requires captive customers.
7. **The capture suppresses innovation that doesn't serve the capital structure.** Outside technologies, competing institutional forms, and new uses of the infrastructure are blocked, slowed, or starved. The infrastructure's owner decides which futures are allowed to develop.

There are at least three historical examples where the pattern appeared: in railroads, in electricity, and in telephony. Not every infrastructure technology followed this path. For example the highways and water systems were largely publicly financed and never passed through the same private financing trap. The pattern appears specifically in technologies that were privately financed, commercially operated, and that served as foundational layers on which other economic activity depended.

### The Railroads

------

The first railroads were short and local. A line connecting a port to an inland market or a route carrying coal from a mine to a river. They were financed locally with things like town bonds, state charters, small stock subscriptions from merchants who expected to benefit from the traffic. But the track has to be laid across land that must be acquired, over rivers that must be bridged, through terrain that must be graded. All of it has to be built before the first train runs. And that investment was still too big for the farmers and merchants that used to build it on their own. From its first mile, the railroad was shared physical infrastructure that was useful only if someone other than the end user financed and constructed it, and valuable only if it connected to something beyond what any single user needed.

Then the Civil War changed its scale. In 1862, Lincoln signed the Pacific Railway Act, granting millions of acres of public land and government-backed bonds to companies willing to build from the Missouri River to the Pacific. Two years later Congress doubled the land grants because the first round was not enough. Private capital piled on top of the government's foundation while the war effort and later the expansion of industrialisation compounded the demand. But like with the small local railroads the track, the terminals, the bridges, the rolling stock had to be built before the traffic existed to pay for it. But this time the scale was massive and  the railroad bonds became the largest class of securities in the American economy. By 1880, total permanent investment had reached $5.2 billion, and the building was barely half done. Every mile of track carried interest that accrued whether the trains ran full or empty.

The financiers who sold those bonds could not let the railroads fail without destroying their own bondholders. Jay Cooke became the exclusive bond agent for the Northern Pacific and ended up owning seventy-five percent of the company because the bonds would not sell fast enough. Morgan took board seats in the railroads he financed because he could not afford to let his investments collapse. The bank needed the railroad to survive so the bonds would pay. The railroad needed the bank to keep extending credit so construction could continue. The entanglement pulled the industry upward toward fewer, larger systems controlled by fewer, larger financial interests. By 1892, only forty-four percent of railroad stocks paid any return at all. The small local line that a town could finance was being absorbed into continental systems whose economics required traffic volumes that no single region could generate alone.

At the same time the railroads dissolved every local pricing structure in the country. The cost of shipping freight by rail fell more than ninety-five percent. Canal traffic on the Erie system peaked in the early 1870s and declined for two decades even as the national economy grew. Canal towns lost their bottleneck value while the local merchants lost their distance advantage. The farmer could reach distant markets but so could every other farmer, and the price his crop could command fell as geographic isolation stopped protecting it. The railroad companies laid track first, sold the land along the routes, brought the settlers, moved the grain. They also got to decide which town became a stop and which got passed by essentially sealing the fate of every community in the path of their lines. The technology was doing exactly what its builders had promised but at the same time it was also destroying the older arrangements faster than the new national market could absorb the displaced.

And the customers who were supposed to generate the traffic that would pay for the infrastructure were being squeezed by it. The farmer needed the railroad to reach the market and the railroad needed the farmer's freight to service its debt. But the farmer could only ship through the company that owned the track and at rates the company set. By the mid-1880s, twenty competitive routes connected St. Louis and Atlanta. But the competition instead of producing  fair prices produced rate wars that bankrupted the weaker lines and discriminatory pricing that favoured large shippers over small ones. The new economy was forming, but it was not forming fast enough. The debt from the construction was due now while the revenue from a mature national market was years away. The gap between what the infrastructure cost and what it earned was the space where things began to collapse.

So the railroad owners reached for what was available, the infrastructure itself. Discriminatory rates for captive routes where no competition existed. Secret rebates for large shippers who could guarantee volume. Terminal control that decided which businesses could access the railhead and which could not. Pooling arrangements were set between supposed competitors to keep rates high. Every mechanism pointed in the same direction: captive traffic at controlled rates. The financiers behind the railroads did not want open competition because rate wars would bankrupt their investments. And the railroads did not want independent access because independence meant bargaining power for the customer. The pressure of their massive debt required capture that prevented the open market from developing naturally. And the absence of an open market prevented revenue from catching up to the debt. 

### Electricity

In the 1880s, electricity was a local affair. A factory owner might install a generator to power arc lamps on the floor, a hotel might wire some lights for its lobby or a town might build a small municipal plant and run lines to the main street. All this needed modest capital. The town could fund its own plant and the factory owner could afford a generator. In many places, the electrical system was something the community built, owned, and operated for its own residents. By the early 1900s there were thousands of these small municipal and private utilities across the country, each serving a small area. But the power these small units produced could not cover a territory. And even this early it was shared infrastructure that no single user could build alone.

The split between lighting and streetcar generators made the problem even more pressing. Lighting peaked in the evening while the streetcar traffic peaked during the morning and evening commute. As a result in the middle of the day the generators sat partly idle. But unlike individual generators central stations with modern turbines could serve both loads continuously, producing electricity at a fraction of the cost per kilowatt-hour. And alternating current which was the output of these turbines meant power could travel long distances without crippling losses. A generating station could serve a city or even an entire region. But a central station costs far more than a local generator and the transmission lines crossing a region cost more than wires running down a street. Every component of the network, the substations, the distribution networks, the backup systems, the maintenance, all of it scaled with the territory. And all of it had to be built and paid for before the full customer base existed to fund it.

Holding companies began acquiring controlling stakes in operating utilities, often with as little as ten to twenty percent of the stock. They issued their own securities backed by the earnings of the utilities they controlled. The proceeds funded the acquisition of more utilities, more utilities meant more earnings and more earnings backed more securities. The structure grew upward, layer upon layer, each controlling the one below, each issuing paper to fund the next acquisition. By the 1920s the pyramids were accumulating debt at multiples the underlying earnings could never sustain. Samuel Insull's empire controlled over $500 million in utility assets with $27 million in actual equity. By 1932, the consolidation was nearly total. Private utilities generated ninety-five percent of American power. Eight holding company groups controlled roughly seventy-three percent of the business, with three empires dominating the rest. It was even marked on their names, Associated Gas and Electric Company, Standard Gas and Electric. These were all financial structures that had bought both sides of the transition, stacking the gas industry that electricity was replacing and the electrical industry replacing it inside the same pyramid.

Electric light was brighter, safer, cleaner, more controllable than gas. The gas lighting industry, a mature urban infrastructure system with its own workers, supply chains, and capital base, lost its core revenue in less than a generation. The ice trade that began as natural ice harvested and delivered to homes, hotels, hospitals, breweries, and railroad cars, had already been weakened by manufactured plant ice and essentially finished by electric refrigeration. By 1950, ninety percent of American city homes had a refrigerator. The plants, the delivery routes, the supply relationships, none of them survived. Water power had anchored manufacturing to geography for centuries. It was the reason why mill towns of Lowell and Paterson were built around falls and rapids. Because of it a factory's competitive position depended on its proximity to a river or a falls. Electric transmission cut the anchor. A factory could now locate where land was cheap and labour available, anywhere a power line could reach. And the thousands of municipal utilities, where the town owned the generator, set the rates, and decided where to wire, were absorbed into private systems. By 1931, more than a third had been bought out or driven out. The town had lost control over the terms of its own energy supply.

And the grid expanded where the returns were highest. By the mid-1930 urbans, industrial and suburban customers were wired. But only about eleven percent of American farms had electricity. Every dollar spent wiring a remote farm was a dollar that could instead be spent acquiring another urban utility, issuing another round of securities, and adding another layer to the pyramid. The old economy's arrangements of gas infrastructure, ice distribution, water-powered manufacturing geography, municipal energy self-governance, had already been broken. But the new electrical economy had not yet reached the majority of the country while debt was compounding and the customer base that was supposed to fund it was still not fully there.

So the holding companies reached for extraction through the machinery above it. Management fees siphoned earnings upward through layers that no state authority could see. Construction contracts awarded to affiliated subsidiaries at inflated prices. Equipment and fuel routed through affiliated companies at above-market rates. Hundreds of thousands of ordinary people bought securities they believed were backed by the solid earnings of their local power company. Technically they were but after going through six to ten layers of financial engineering that had turned those earnings into something else entirely. And the transition from gas to electricity was managed at whatever pace maximised dual extraction from both revenue streams. The debt required growth, the growth required acquisition, the acquisition required securities, and the securities required investors who believed the paper was backed by something solid. Unfortunately, it wasn't.

### The Telephone

The telephone began as a local service. Alexander Graham Bell patented it in 1876. The first exchanges appeared in towns and cities as a switchboard in a central office, a few copper wires running to nearby subscribers and operators plugging cables into a board to connect one caller to another. Through them a shop owner could reach a supplier a few streets away and a doctor could receive calls at home. The capital requirements for setting this up were modest, a local businessman could start an exchange and a rural cooperative could string wire between farms. But the telephone had a structural feature that distinguished it from every other utility. A railroad carrying grain from a farm to a mill provides value on its own. A generator powering a few hundred lamps is useful whether or not it connects to a larger grid. But a telephone exchange that cannot connect to other exchanges is fundamentally limited. The value of the telephone is in the network and the network becomes more valuable with every additional subscriber it can reach.

Bell's original patents expired in 1893 and 1894. When they did, thousands of independent telephone companies formed. There were over six thousand by 1904, many serving small towns and rural areas the Bell system had not yet reached. The landscape was fragmented and local and could be covered by small investments. But when long-distance came about the capital requirements exploded.  It needed copper trunk lines strung across hundreds of miles, repeaters and amplifiers developed and installed at intervals to maintain signal quality and switching systems that could route a call from a local exchange through the trunk network to another exchange in another state. AT&T (formed in 1885 as Bell's long-distance subsidiary) built the only national long-distance network. That network had to be financed and constructed across the continent before the call volume existed to justify it. And that network was the asset that made everything else possible. 

When Theodore Vail became president of AT&T in 1907, he articulated the vision that would define American telecommunications for seven decades: "One Policy, One System, Universal Service." It was also a consolidation strategy because AT&T refused to interconnect with independents. If you were on an independent system, you were cut off from the largest pool of subscribers in the country. For a business that needed to reach Bell customers, this was often intolerable. AT&T acquired independents steadily, each acquisition adding subscribers that made the network more valuable and the next independent's isolation more painful. Long-distance access could be granted or withheld and an independent network without it was crippled. And AT&T could raise money on Wall Street while a rural cooperative running a local exchange from a small office could not. The Kingsbury Commitment of 1913 slowed the most aggressive acquisition phase, but it did not reverse the structural advantages. By mid-century, the Bell System controlled the vast majority of American telephone service. It owned the local networks in most major markets, it had the only national long-distance network, the manufacturing operation that supplied all the equipment, and the research institution that produced the science and the patents. 

The telegraph was the first casualty. Western Union had built the first instantaneous long-distance communication network. It had its own infrastructure, trained operators, offices in every town, an economy built around the rhythm of coded messages sent through intermediaries. The telephone made it unnecessary for most purposes. A voice call was faster, required no special training, and did not require the sender to travel to an office or wait for a messenger. The telegraph survived only in domains where a written record was essential. Western Union eventually pivoted to money transfer but nevertheless its core business was hollowed out over a generation. Messenger services also contracted. The phone call replaced the boy sent across the city with a note. And the telephone quietly eroded the value of physical proximity in commerce. The merchant could call the supplier instead of walking to the warehouse and the manufacturer could check prices by phone instead of sending someone to the market. The intermediary whose value was primarily in being reachable  (the local broker, the agent who bridged a communication gap) found that function available independently through the wire. The dissolution was spread across many small reductions, quieter and more diffuse, but cumulative.

The dissolution was also uneven. Like the urban centers with electrical power coverage before them urban businesses with telephone subscriptions gained the advantages first. Rural communities, small towns, and businesses that could not afford a subscription were left operating at the old speed while their competitors in the cities moved faster. The telephone gave their counterparts capabilities they did not yet have. And AT&T's "Universal Service" commitment competed inside the same corporate structure with the financial logic that favoured serving the customers who generated the returns. Stringing copper wire across miles of farmland to reach a handful of subscribers was expensive while the urban service was profitable. 

AT&T's capture had built a closed loop. Bell Labs conducted the research. The research produced patents. The patents were assigned to AT&T and Western Electric. Western Electric manufactured every telephone, every switch, every cable used in the Bell System with no outside manufacturer able to sell equipment into the network. The Bell Operating Companies bought everything from Western Electric at prices set internally, not by market competition. Those inflated equipment costs flowed into the rate base and were paid by every telephone customer. The FCC's Special Telephone Investigation in the 1930s documented the distortion in detail. And nothing could connect to the network without AT&T's permission. No answering machine, no modem, no fax machine, no device that an outside inventor might build could gain access to their network. The revenue from the telephone service funded Bell Labs. And Bell Labs in turn produced more research, more patents, which fed back into Western Electric, which fed back into the network. The loop was closed at every point. 

### Three binds, three corrections

In every case, the dissolution did not stop at the boundary of the older economy. It circled back. The railroad dissolved the geographic protections that had sustained its own customers' crop prices. The farmer needed the railroad, but the railroad's own reach had made the farmer's goods worth less. The electrical holding companies overcharged the customers whose bills funded the grid, while the technology dissolved the economic foundations of the communities those customers lived in. AT&T's closed loop inflated every subscriber's bill while blocking the innovations that would have made the telephone more valuable to the people paying for it. The infrastructure's own customers were caught inside the bind. They needed the technology, paid for it while, at the same time, they were being squeezed by its effects and its owners' capture simultaneously.

And in every case, the capture did not only extract from the economy that existed but also suppressed the economy that could have formed. The railroads controlled terminal access and set discriminatory rates that decided which businesses could reach the market. This produced economic gatekeeping that determined which enterprises were permitted to develop on top of the infrastructure. The electrical holding companies managed the pace of the gas-to-electricity transition to maximise dual extraction from both revenue streams, and absorbed or drove out the municipal utilities that represented a competing institutional form: local democratic control over energy supply. AT&T drew a hard corporate boundary around the network and its science, blocking any outside device, manufacturer, or idea about what the telephone could become from developing. Three different mechanisms with one function: the infrastructure's owner deciding which futures were allowed.

The railroads were hit first. The Panic of 1893 was the largest economic crisis the country had experienced. By the time it passed, a quarter of all railroad mileage in America was in receivership. The companies that had built a continent's circulatory system could not service the debt they had taken on to build it. But before the panic arrived, the decades of capture had already exacted a different cost. The infrastructure owners had been deciding which towns thrived and which withered.  They set which shippers could access the market at competitive rates and which were squeezed by captive routes with no alternative carrier. The open freight economy that the railroad's own technology made possible could not form while the companies that owned the tracks controlled which businesses were allowed to develop on top of them.

Morgan and a handful of other financiers reorganised the wreckage, consolidating the bankrupt lines into fewer, larger systems under tighter financial control. The Interstate Commerce Act of 1887 had created the first federal body with authority over railroad pricing and access, but without the power to set binding terms. The Hepburn Act of 1906 provided that power in the form of the authority to set maximum rates, examine the books, and define the conditions of access. The industry that had operated as a private toll system on public land was, two decades and two financial panics later, placed under conditions. The railroads had demonstrated what happened when no one set the terms.

Once the terms were set, the continental economy the railroads had promised could finally take shape. Finally manufacturing at scale, agriculture reaching national markets, industrial supply chains spanning regions could develop without every new enterprise needing permission from the company that owned the route underneath it. The railroads kept running and the toll system above them changed.

The electrical holding companies collapsed a generation later. The rules built after the railroad crashes had been sector-specific. The ICC controlled  transport rates and railroad access but had no jurisdiction over electrical utilities. State commissions could examine the local power company within their borders, but the holding companies operated across state lines, and the extraction happened in the gap between jurisdictions no single institution could see. The same structural pattern ran unchecked in a different sector through a different mechanism. When credit tightened after 1929, the pyramids that had looked solid revealed what they had always been. They were essentially structures that depended on the next issuance, the next acquisition, the next turn of the cycle. Fifty-three major holding companies went bankrupt between 1929 and 1936. Insull fled to Greece, was extradited, stood trial, and was acquitted but his investors were not made whole. Hundreds of thousands of ordinary people who had bought securities they believed were backed by the reliable earnings of their local power company discovered that their money had passed through six to ten layers of financial engineering and vanished.

The Public Utility Holding Company Act of 1935 dismantled the pyramid structure, reducing holding companies from approximately two hundred to eighteen over the next two decades. The Rural Electrification Administration, created the same year, extended the grid on terms the private system's economics had never supported. The technology had been ready for years. The capital structure had pointed every incentive elsewhere. When the terms changed, the grid reached the rest of the country within a generation. The infrastructure survived the collapse of the companies that had built it while the financial engineering above it did not. And the universal, affordable electricity that followed reached every farm, every small town and every household. And it became the foundation on which the consumer economy of the next forty years was built.

In both cases, demand did eventually grow. Railroad freight volumes expanded for decades after the panics while electricity consumption rose for a century. The infrastructure that nearly bankrupted its builders became, over the long run, as valuable as the original investors had assumed. But the bind was never a question of demand. The railroads were profitable by the early twentieth century. But only after a quarter of American mileage had passed through receivership and Morgan had reorganised the wreckage into systems run for bondholders. The grid reached every household but only after the pyramids collapsed, hundreds of thousands of ordinary investors were wiped out, and the terms under which the industry operated were rewritten. Demand growth funded the infrastructure's survival. But it did not prevent the bind from operating, the capture from suppressing the economy above it, or the costs from compounding across decades. What resolved the bind was a change in the terms under which the demand was served.

AT&T was the opposite of the railroad speculators and the electricity pyramids. The system interventions developed to prevent the previous crashes ensured that it was conservatively capitalised, rate-controlled, backed by physical assets and generating steady returns. There would be no panic, mass bankruptcy, or retail investors wiped out. But in the decade before the war, the closed loop was reaching beyond telephony. AT&T had moved into broadcasting, claimed monopoly control over the wire connections linking national radio networks, locked Hollywood studios into exclusive sound equipment contracts through Western Electric, and bought the Teletype Corporation to dominate data transmission. 

The Walker Report in 1938 documented the extraction that funded the expansion. Transfer pricing was inflating every American's phone bill through mechanisms state authorities couldn't see. And behind the financial extraction sat something worse: an airtight corporate boundary around foundational science and the physical network itself. No outside equipment could connect to it. No answering machine, no modem, no fax machine, not a single device an independent inventor might build. Also no outside manufacturer could sell into the network and no outside idea about what the telephone system could become was permitted to develop. 

An entire generation of innovation on the network was gated by a single company's judgment about what served its loop. The rules that had prevented the railroad-style crash and the electricity-style pyramid had also preserved a regressive structure. This structure's cost was not financial collapse but something quieter. A controlled future that could only develop at the pace and in the direction that served one company's interests.

The war interrupted its trajectory. At the break of WWII, Conservation Order L-50 froze civilian telephone expansion and diverted Western Electric's manufacturing entirely to military hardware. AT&T's commercial expansion into adjacent markets was severed overnight. But the war also placed the government inside Bell Labs. It was now full of military liaison officers, scientific monitors and a thousand co-developed defence projects. Just  the laboratories alone doubled from 4,600 to over 9,000 personnel. When Bell Labs produced the transistor in December 1947, the government understood immediately from the inside what was locked behind the corporate boundary. And the events that followed began a chain reaction that would eventually produce the largest explosion of wealth creation in human history.

## The escape ladder

Military procurement drove the first wave of computing. Right after the war the early semiconductor firms, computers and systems that processed calculations were funded by defense contracts. But the technology that emerged followed the same trajectory as every infrastructure technology before it. For most of the mid-twentieth century computing required massive capital investment. Machines were expensive, huge, physically demanding, and operated by specialists. They required dedicated power, cooling, maintenance, procurement budgets, and institutional permission. They belonged to governments, universities, laboratories, banks, airlines, defence contractors, and large corporations.

Both the hardware and the people who worked on it lived in special rooms. The user submitted work to the machine, or reached it through a terminal far away from where the actual computing happened. The machine that did the work sat elsewhere, behind glass, managed by people whose job was to make sure ordinary users did not break it.

Software was not yet an independent economic layer. It was bundled with the machine, or treated as part of the service that made the machine useful. The value sat in the hardware, the lease, the service contract, the installed base, and the institutional relationship. Computing was not something a person could put on a desk, take apart, and build a company around. The vendor owned the machine, the software, the service relationship, and the customer's dependency on all three. One company, IBM, dominated the way AT&T dominated telephony through a bundled arrangement where no single layer could be separated from the rest.

Without what came next, computing would have continued on that trajectory. It would have become an industrial infrastructure controlled by a few vendors, accessed on their terms, at their prices, through their systems. The familiar pattern was already becoming visible.

But in the 1960s and 1970s, that tight industrial lock began to break.

### The first opening: hardware

------

In the mid-1970s, machines like the Altair 8800 showed that a computer could be built around a commodity microprocessor and owned by an individual. The processor, the memory, the boards, the peripherals for these machines were components, manufactured by competing firms, sold on the open market, and assembled by anyone with the technical knowledge and the patience to learn. And these highly technical engineering hobbyists and enthusiasts began designing and assembling machines of their own.

By 1977, the Apple II, Commodore PET, and TRS-80 made the category recognisable to a wider public: a keyboard, a screen, local storage, and software running on a machine one person controlled. In the early 1980s, machines like the ZX Spectrum, the Amstrad CPC, and the Commodore 64 brought computing to millions of homes. A teenager could learn to program, small business could do its own accounting and a school could put machines in a classroom. These personal computers reached people who would never have touched a mainframe or a minicomputer.

But each of those machines was a sealed product from a single manufacturer. Sinclair designed the Spectrum's custom chips. Commodore built the 64's sound and video processors in-house. Amstrad controlled its entire hardware stack. You could buy one and use one but you could not build one. The parts were proprietary and designs were closed. These machines made computing personal but they did not make it open.

In 1981, IBM changed this. The IBM PC used off-the-shelf commodity components: an Intel 8088 processor, standard memory, parts available from multiple suppliers. The architecture was published and the only proprietary element was the BIOS. Compaq reverse-engineered even that within two years through clean-room methods. Phoenix and AMI later sold compatible BIOS chips to anyone who wanted to build a clone. IBM had entered the market as the trusted institutional brand but the compatible ecosystem escaped IBM's control.

What followed was the emergence of an ecosystem. Compaq, Dell, HP, Gateway, and hundreds of smaller companies built compatible machines. Component manufacturers competed on price. And the components were designed to fit together. You slot a card into a bus, plug in a cable, screw in a drive. This didn't require any soldering, circuit design or engineering degree. A mildly technical person could assemble a working computer from commodity parts in an afternoon. That had never been possible before. Local shops sprang up building custom machines while small businesses assembled their own. The barrier to manufacturing a computer dropped from a factory floor to a kitchen table.

Competition between hundreds of manufacturers drove prices down year after year. Not slowly, the way industrial costs sometimes decline with scale, but rapidly with each generation of chip being more powerful and less expensive than the last. The computing power that had once required a room, a building, and a procurement department was, within two decades, available for a few hundred dollars.

This was different from every previous infrastructure technology. A railroad could not be assembled from commodity parts by a small team. An electrical grid could not be built in a garage. A broadcasting network required spectrum, transmitters, and a capital structure to sustain it. But a personal computer could be assembled from components that were getting cheaper every year, on a desk, by one person, with no institutional permission required.  And after that personal computer was assembled it could almost immediately produce something that eventually changed our understanding of the nature of products: software.

### The second opening: software

In the mainframe era, software had been part of the machine. You bought the hardware; the software came with it, or was written for it by the vendor's engineers, or was developed in-house by your own programmers working on that specific system. Software was not something you bought separately, compared across vendors, or carried from one machine to another.

IBM's unbundling in 1969 began changing that. IBM had been the dominant mainframe manufacturer, and its decision to price software separately from hardware did not create the software industry by itself. But it made software visible as an independent economic object. It was finally something that could be priced, sold, compared, and improved on its own terms. Once that separation existed, an independent software market could form.

The personal computer widened the opening dramatically. At first, many of these machines did not merely run software but actually invited their owners to write it. Turn on a ZX Spectrum, a Commodore 64, or an Apple II, and the machine often placed you close to a programming prompt. BASIC was not a professional software factory, but it made the computer feel programmable from the first keystroke. A teenager could type in a listing from a magazine, a small business owner could write a crude accounting tool and a hobbyist could make the machine do something no vendor had shipped yet.

The IBM PC changed the shape of that opening. It did not arrive as a complete development environment in the way later programmers would understand the term. It had BASIC but it did not come with any serious development tools. The assemblers, compilers, editors, debuggers, database tools, and later integrated development environments became separate products. And because the IBM-compatible market became large, common, and commercially legible, those tools could be sold to millions of people using broadly compatible machines.

Programming tools had moved out of the institutional computer room and into the personal market. You no longer needed a mainframe contract, a university affiliation, or a corporate budget. You needed a desk, a machine, the right tools, and time.

And at this point, software's first advantage became clear: it could be copied for almost nothing.

A person could write a program on a personal computer, copy it onto floppy disks, and sell it through mail order, bulletin boards, shareware catalogues, or retail shops. The cost of producing each additional copy was the cost of a disk and a label. If the software was useful, the margin between what it cost to produce another copy and what a customer would pay for it could be enormous.

But software had another advantage, and this one made it different from other copy-based industries. Music and films could be copied. A record, a cassette, a VHS tape, a DVD,  all of them had some version of the same replication logic. Make the expensive first object, then manufacture copies at a much lower cost.

Software was different because the copy did actual work. A copied song reproduced an experience and a copied film reproduced a performance. But a copied program reproduced a capability. A spreadsheet could replace hours of bookkeeping, word processors could change how documents were written and revised. And a database could organise records that had once lived in filing cabinets, ledgers, or human memory.

And unlike a song or a film, each software copy could become part of the customer's own operating machinery. It helped the buyer produce, calculate, organise, communicate, sell, or manage. That meant software could be priced against labour, time, errors, and business output.

This combination of near-zero copy cost attached to executable utility was something new in the history of industrial economics. It was something that the railroad, electrical utilities or manufacturing companies couldn't offer. Software economics broke the relationship between revenue and physical production. More importantly, it broke that relationship for products that could themselves increase the productivity of the buyer.

But the true revolution came when software was released from its last material chains and could move through the open internet.

### The third opening: distribution

Before the internet, distributing software required passing through a series of gates, and each gate had a keeper who decided who got through. A software company needed a publishing deal to get its product manufactured, a distribution deal to get it into the supply chain, and retail shelf space to put it in front of customers. Each intermediary took a cut and had veto power. CompUSA, Egghead Software, and the electronics aisle at Sears decided what the customer could see. A product that the retailer declined to stock did not reach the customer.

The pattern was not unique to software. A musician needed a record label to get a CD manufactured, distributed, and placed in stores. A writer needed a publisher and a filmmaker needed a studio willing to finance production and a distributor willing to place the film in theatres. In each case, the creator could not reach the audience without passing through institutions that controlled the route, took a share of the revenue, and reserved the right to say no.

The economics of this system matched the economics of physical goods. Manufacturing discs, printing manuals, shipping boxes, stocking shelves, managing returns. Distribution had real costs, and the costs created a minimum scale. A product had to sell enough units to justify the shelf space, the manufacturing run, the distribution deal. Products below that threshold simply did not exist in the market, regardless of whether anyone wanted them.

But by the early 1990s, the web had become publicly available. Mosaic made it usable for ordinary people in 1993 and in 1995, when the NSFNET backbone was decommissioned and private networks took over what we understand today as the commercial internet could finally reach people's homes.

The internet removed the gates. Distribution moved out of proprietary networks and physical supply chains and into open protocols. HTTP, TCP/IP, email, FTP. And the infrastructure was open, standardised, and not owned by any single company. A builder could write software on a cheap machine and make it available to anyone with an internet connection, anywhere in the world, by uploading a file to a server. The cost of reaching the entire planet dropped from hundreds of thousands of dollars (manufacturing, distribution deals, retail placement) to a few dollars a month for web hosting.

No one could stop you from publishing or reject your software because it competed with something the gatekeeper preferred. No one took 30 or 50 percent of your revenue for the privilege of being reachable. No one decided your product did not deserve shelf space. The route from builder to user ran through an open network that no single company controlled, and the builder needed nothing more than a website and a download link.

The speed changed what was possible. Physical distribution took months, you had to go through manufacturing, shipping, regional rollouts, shelf placement. Internet distribution was immediate. Upload the file and it was available globally the same day. A bug could be fixed and the fix distributed in hours, not quarters. The entire development cycle compressed because distribution was no longer the bottleneck.

Products that could never have survived the old gatekeeper system found audiences. Shareware reached millions of users through download sites and word of mouth. Open-source projects assembled global communities of contributors who had never met. A student's side project could reach more users than a commercial product with a retail distribution deal. The audience decided what was worth using, not the distributor.

The internet did for distribution what the personal computer had done for computation. It removed the industrial bottleneck. A builder no longer needed to finance distribution infrastructure, negotiate with gatekeepers, or reach minimum manufacturing scale before the product could find its audience. The cost of trying dropped close to zero, and the cost of failure dropped with it.

### The escape economics

Together, these openings produced something that had never existed before. Hardware meant the builder owned their setup. A person with a few hundred dollars and technical knowledge had everything needed to create software. The capital requirement had dropped from millions to almost nothing.

Software meant the product could be reproduced at near-zero marginal cost. Every copy after the first was essentially free. The economics of the product were fundamentally different from the economics of any physical good.

Distribution meant the route to the customer was open and approaching free. The builder did not need to finance a delivery network, rent retail space, or negotiate with distributors who controlled access to the market.

Put them together, and for the first time in the history of industrial economies, a small team (sometimes even a single person) could build a product, produce it at near-zero marginal cost, and distribute it globally without owning any of the underlying infrastructure.

The economics that followed were unlike anything the older industries had produced. Gross margins of 75 to 85 percent were normal for a mature software company because the cost structure was genuinely different. The expensive part was building the product. The cheap part was that everything after that (copying, distributing, serving, and supporting additional users) cost a fraction of what the customer paid.

This model powered the largest creation of economic value in modern history. Companies built on software economics (Microsoft, Google, Salesforce, Adobe, Oracle, SAP, and thousands of smaller ones) produced a generation of growth, employment, and wealth that reshaped the global economy. Venture capital, startup culture, the Silicon Valley mythology, the entire ecosystem of technology investment and entrepreneurship, all of it was built on top of this economic model. Build the product once, sell it again and again and keep 80 cents of every dollar.

For roughly the last twenty-five years, this felt like the natural condition of technology. But it was not. Instead it was a historically unusual escape from the infrastructure trap. For the first time a major technology had completely broken out of the pattern that railroads, electrical grids, and broadcasting had all followed. Finally the builder did not need to own or finance the foundational layer before the product could exist and the economics of the product were governed by the cost of the first copy rather than the cost of every copy.

But the escape did not prevent concentration. Within two decades, the open economy produced its own dominant positions.  We got an operating system monopoly, a search engine that handled most of the world's queries, a marketplace that reshaped retail and social networks with billions of users. But the mechanism of concentration was different. These companies did not own the foundational infrastructure and use that ownership to control who could build above it. They achieved dominance through network effects, data accumulation, and platform economics while building on the open substrate, not owning it. The difference here is important because when the conditions underneath began to erode, the concentrated players were in position to reach for the older form of capture. But that is a later chapter.

The next question is how that escape was possible in the first place. The personal computer did not emerge from technology alone but from a market whose shape had been altered by earlier institutional choices. The internet did not organise itself into open protocols through the natural generosity of the companies that owned the existing communications networks. Software did not separate from hardware because vendors spontaneously decided to let go of their customers.

Each of those openings was produced by specific conditions which came from institutional decisions, legal constraints, public investments, and structural choices. And all combined to open computation, separate software from hardware, and build open distribution before the next generation of companies arrived to build on top of them.

---

## The conditions under the ladder

The breaking of the conditions that had chained every other major infrastructure technology was not some weird out of the blue magic. It was something that emerged from a sequence of institutional choices that opened the substrate before companies climbed on top of it. Monopolized science was forced outward, bundled computing was pulled apart and networking was standardized before it was commercialized.

And for each of these to happen there was a long path of institutional change that set the stage long before the first online stores appeared in our browsers.

### Hardware: The science escaping the monopoly

The case against the Bell System had been building for over a decade. The Walker Report in 1938 had documented the financial distortion that the double bind was already beginning to cause. It was tracking how Western Electric overcharged the Bell operating companies, the inflated costs flowing into the rate base, how every American telephone customer was paying more so the closed loop could sustain itself. Holmes Baldridge, an attorney in the Department of Justice, had been working on the structural problem since the late 1930s. The legal thread was about pricing, purchasing, and the corporate boundary that locked independent manufacturers out of the largest equipment market in the country.

But the war had opened a second line of sight. For several years, military and government personnel had worked inside Bell Labs on its radar, fire control, encryption and communications systems. They had seen not just the telephone research but the scope of what the laboratories were producing: semiconductor physics, materials science, information theory, the foundations of computing. When Bell Labs demonstrated the transistor in December 1947, the people who had been inside understood immediately what it meant. It was not just another telephone component but the foundation of a new technological era, locked behind the same corporate boundary that Baldridge had been trying to open on financial grounds.

In 1949, the Department of Justice filed *United States v. Western Electric*, demanding the separation of AT&T from its equipment manufacturer. The legal case rested on the Walker Report's findings — the pricing distortion, the closed purchasing loop. But the stakes now extended far beyond telephone equipment pricing. The government understood, from direct operational experience inside the laboratories, that what was locked behind AT&T's corporate boundary was not just an overpriced equipment supplier. It was foundational science that would define the next technological era.

The military saw the same thing and drew the opposite conclusion. The Department of Defense wanted the integrated Bell System preserved — for missile guidance, for communications infrastructure, for the defence systems that depended on AT&T, Western Electric, and Bell Labs working as one coordinated unit. The Korean War froze the case. The government that had filed the suit was lobbied by another part of the same government to drop it.

The lawyers who negotiated the 1956 consent decree designed a compromise that split the difference. The telephone monopoly was preserved and the military got the integrated structure it wanted. But the terms under which it could exist were changed. AT&T was fenced into common carrier communications. It could run the telephone system, but it could not expand into adjacent markets such as commercial computing. And AT&T had to license its entire existing patent portfolio. Every one of its approximately 7,820 patents, including the transistor patents and the broader semiconductor research had to become available royalty-free to any applicant. Future patents would be licensed on reasonable terms, with the court retaining authority over what "reasonable" meant.

The remedy went further than fixing the pricing distortion the Walker Report had documented. It targeted the science itself because the people shaping the remedy understood, from their years inside the laboratories, that the patents were the bottleneck that made all the difference. Without the wartime technical legibility, the consent decree might have reformed Western Electric's pricing and opened equipment purchasing to outside manufacturers. That would have addressed the financial distortion. It would not have opened the transistor.

The decree did not create the semiconductor industry by itself. Federal research spending, defence procurement, universities, the GI Bill, and post-war industrial capacity all contributed to it. Even AT&T had already begun sharing some transistor knowledge through symposia. But it was always on its own terms, at its own prices, to applicants it selected. The consent decree made the opening mandatory, universal, and durable enough that the industries which grew from it were fully established before the terms changed. Any firm could apply for the license and build without any need to negotiate, get approval or have a commercial relationship with AT&T.

The semiconductor industry that followed made chips cheaper and more powerful every year. It was these cheap chips that made hobbyist computers possible, then clone manufacturers, then local assemblers, then mass-market personal computers. The science that had been locked inside a telephone company's closed loop became the foundation of an industry the telephone company was not allowed to enter.

### Software: The program was pulled out of the machine

In the mainframe era, computing was sold as a bundled institutional arrangement. A customer leased or bought a machine and received operating software, applications, and support as part of the package. The line between hardware, software, and service was not always clean, and it did not need to be. The customer paid for computing as a whole. The vendor supplied computing as a whole.

But the bundle created a distortion. If software lived inside the hardware relationship, it could not become an independent market. It could not be priced on its own. It could not be compared across vendors. A program written by one company could not easily be sold to customers using another company's machines. The value of the software was hidden inside the value of the machine.

Independent software companies existed, but they lived in the cracks. Applied Data Research, Computer Sciences Corporation, Informatics, and a handful of others wrote custom programs for specific customer needs that the hardware vendor's bundled software did not cover. A bank's particular transaction workflow. An airline's specific reservation logic. Specialised tools the vendor had not built yet. The work was real and the companies were real, but their position was structurally precarious. They competed against free — IBM's bundled software was paid for inside the hardware lease, invisible as a separate cost. And IBM could watch what the independents sold successfully, develop its own version, and fold it into the bundle. The independents existed at the pleasure of the gap. The gap could close at any time.

By the mid-1960s the structural problem was becoming visible from multiple directions at once.

The federal government was the world's largest single customer of computing equipment, and by 1965 it was paying for the distortion directly. The Clewlow Report (a study commissioned by the Bureau of the Budget) documented how the government's dependence on proprietary, incompatible systems inflated federal budgets and prevented competitive bidding. Most of those systems were IBM's. So Congress responded with the Brooks Act, which centralised all federal computing acquisitions under the General Services Administration and mandated standardised, competitively bid systems. No federal body had jurisdiction over the computer industry the way the FCC had jurisdiction over telephony. But it could use its purchasing power to force the market open. The Brooks Act did for computing what the Walker Report had done for telephony, it made the distortion visible and documented, and it gave the institutional machinery something clear to work with.

The independent software industry saw it and went directly to the people building the case. Martin Goetz at Applied Data Research had understood the problem from the inside. He had worked at Sperry Rand before joining ADR, and he knew what it meant to sell software against a vendor that gave its own away for free. In 1968, Goetz received what is generally considered the first software patent: a sorting algorithm. It was unremarkable as technology, but it was a legal assertion that software was a distinct creation with its own value, separable from the machine it ran on. If software could be patented, it was a thing. And if it was a thing, then giving it away inside a hardware lease was a competitive choice that could be challenged. But Goetz did more than file patents. In 1967 and 1968, he and ADR's legal counsel Morton Jacobs met directly with attorneys in the Department of Justice, arguing that IBM's free software was an exclusionary practice that destroyed the property value of independent software. The Association of Data Processing Service Organizations drafted position papers and petitioned the courts. The independent software industry was building the evidentiary case alongside the prosecutors.

Inside the Department of Justice, career attorneys had been building toward a formal action since 1964. The investigation began under Robert F. Kennedy's tenure as Attorney General, gained momentum as IBM's System/360 established dominance, and by 1967 had become a dedicated operation led by Raymond Carlson and a specialised team in the Antitrust Division's Computers and Finance Section. Five years of subpoenas, depositions, and market analysis preceded the filing. This was not one person's project. It was an institutional process that outlasted three Attorneys General.

IBM itself recognised the vulnerability before the government acted. In December 1966, Howard Figueroa, IBM's director of policy development, established an internal task force explicitly chartered to determine how to unbundle software and preempt the legal threat. The task force debated patent, trade secret, and copyright strategies for protecting software once it was separately priced. A second task force formed in late 1968 under senior vice president Spike Beitzel, operating with extreme urgency. IBM had been preparing a defensive unbundling for three years before the filing arrived.

On January 17, 1969 in his last full business day as Attorney General, Ramsey Clark filed *United States v. IBM*. The timing was  calculated against two political clocks. The incoming Nixon administration was expected to be less inclined to pursue the case. And Nicholas Katzenbach, the previous Attorney General, was about to become IBM's general counsel and filing after that appointment would have created an impossible conflict. Clark used the last hours of his authority to enter the case into the judicial record, where the incoming administration would have to actively drop it rather than simply decline to file. The institutional momentum, built by career attorneys over five years, was now locked into the courts.

Five months later, on June 23, 1969, IBM announced it would begin pricing software and services separately from hardware. The company characterised this as a market decision, and market pressures did indeed exist. The independent software companies were growing, customers were beginning to demand choice, and the mainframe market was evolving. But internal documents revealed during the trial showed the timing and scope were driven by the defensive strategy the Figueroa and Beitzel task forces had been preparing since 1966. It was essentially a voluntary unbundling designed to preempt a court-ordered breakup.

Once software had its own price, it could have its own market and its own companies that didn't just survive in the cracks of IBM's bundle, but whose products stood on their own terms. And once the program was no longer treated as part of the machine, it could be improved, compared, bought, copied, licensed, and sold independently. The independents who had been living in the gaps now had a stable foundation. And the space that opened was large enough for an entire industry to form inside it.

------

### Distribution: The route stayed open

AT&T owned the nationwide telecommunications infrastructure. In the late 1970s, Bell Laboratories designed the Advanced Communications Service. It was a nationwide packet-switched data network that would assign customers a digital telephone number and perform protocol translation inside the network itself. It was launched commercially in 1983 as Net 1000, the system was designed to position AT&T as the centralised gatekeeper of both data processing and transmission. Had it worked, computer networking would have become another extension of the telephone system, just another layer inside the closed loop.

But the 1956 decree had already fenced AT&T into common carrier communications. It could operate the telephone system but it could not enter the computing and data processing industries. When computers and telecommunications began to merge in the late 1960s, the FCC drew the line sharper. Its Computer Inquiries drew a structural boundary between basic transmission (a neutral pipeline) and enhanced services that involved data processing. Because of the consent decree, AT&T could not bundle its physical transmission lines with proprietary routing and processing software. The legal separation of the physical pipe from the logical layer above it created a protected space in which independent packet-switching networks could grow using leased, neutral circuits. The decree that had been designed to open Bell Labs' patents had an unintended second effect: it prevented the owner of the wires from owning what ran over them.

The people who built the first network understood the structural problem from the beginning. In 1962, J.C.R. Licklider (a psychologist and acoustic engineer from MIT) became director of ARPA's IPTO (Information Processing Techniques Office) and began sketching out what a network connecting incompatible computers across institutions would look like. He recognised that AT&T's telephone network, built for continuous voice traffic over dedicated circuits, was structurally hostile to the bursty, irregular patterns of computing. What he envisioned demanded a different architecture, one where diverse machines could share resources without any single vendor controlling the path between them. He wrote it up in a series of memos to colleagues in 1963 under the half-serious heading "Intergalactic Computer Network." 

The funding and the engineering for this became available in 1965 when IPTO's next director Bob Taylor decided to connect his Pentagon office terminal with the lab systems at Santa Monica, Berkeley and MIT. Larry Roberts was appointed as program manager and he turned the vision and the funding into architectural decisions and attached conditions. ARPA required every contractor and every participating university to share source code, disclose interfaces, and refrain from claiming exclusive intellectual property over protocols developed with public funding. A contractor that refused faced termination. Openness was an explicit procurement condition, enforced the same way the consent decree had been enforced by attaching structural requirements to access.

The protocol work that ran on top of the physical network was structured the same way. In early 1969, as the first hardware delivery deadline approached, a group of graduate students from the initial ARPANET universities realised they needed to document the host-to-host software they were designing. Steve Crocker of UCLA wrote the first document on April 7, 1969, and designated it "Request for Comments". It was a deliberately humble title chosen because the students expected professional protocol designers to eventually arrive and take over but no one did. The RFC process became the internet's standardisation mechanism. Anyone could read and write an RFC. No corporate sponsorship was required and no licensing fees were charged. The protocols that would become the foundation of the global internet were published as public knowledge, freely implementable by any developer, any vendor, any operating system. 

The protocol that emerged from this process was designed to resist capture at the architectural level. In 1974, Vint Cerf and Bob Kahn published their proposal for what became TCP/IP, the foundational suite of communication rules that governs how devices connect and exchange data across the internet. The Internet Protocol assumed nothing about the physical medium carrying its packets. It could be anything, telephone lines, satellite links, coaxial cable, fibre, the hardware underneath it didn't matter. This meant the protocol decoupled the application layer from the physical infrastructure. A telecommunications carrier that owned the wires could not force proprietary gateways onto the traffic. The architecture meant that ownership of the pipe no longer automatically conferred the right to decide which applications could exist above it.

On January 1, 1983, the Defence Communications Agency decommissioned the old protocol and every host on ARPANET that had not implemented TCP/IP was disconnected from the network. The following year, the Department of Defense designated TCP/IP as the official standard for all military packet-switched networks. Because the US military was the world's largest single buyer of computing equipment, that procurement mandate created a guaranteed market for TCP/IP-compliant hardware and software. Vendors that wanted to sell to the federal government had to build open-protocol support into their systems. The sovereign market sustained open networking against the proprietary alternatives the way defence procurement had sustained the early semiconductor industry after the consent decree opened the patents.

And the experience of the researchers that had worked inside the military's labs was carried in the academic world. When Stephen Wolff arrived at the NSF in 1986 to direct networking and communications research infrastructure he had already spent fourteen years as a communications researcher in the Army's Ballistic Research Laboratory of Aberdeen Proving Ground. Wolff had witnessed firsthand how proprietary computer networks became stagnant and expensive and  how vendor lock-in choked innovation and inflated costs. He understood what happened when infrastructure was handed to private operators without conditions from operating inside it.

So he designed the NSFNET  (the national academic backbone that connected university networks into a continental system) to carry structural conditions at every level. The backbone required TCP/IP. Its Acceptable Use Policy restricted commercial traffic, because the federally subsidised backbone would not become free infrastructure for private business before the open substrate had matured. Wolff interpreted the policy liberally permitting commercial email interconnections, allowing mailing list posts about software releases. This created space for commercial demand to develop alongside the academic network without letting it capture the backbone. In 1992, Congress validated his approach, officially authorising the NSF to carry commercial traffic that supported the network's research mission.

But the conditioned privatisation was Wolff's most consequential design. The NSFNET backbone had been operated by a consortium of Merit, IBM, and MCI, which formed a non-profit called Advanced Network and Services. When ANS created a commercial subsidiary that began monopolising traffic and refusing to interconnect with competing providers unless they paid transit fees, the crisis Wolff had anticipated arrived. Independent ISPs (PSINet, UUNET, CERFnet) formed the Commercial Internet Exchange to bypass the bottleneck, raising fears that the unified network would splinter.

In May 1993, Wolff and the NSF issued Solicitation 93-52. This was the blueprint for how the public network would become private. Instead of auctioning the backbone to the largest carrier, the solicitation mandated a decentralised architecture. There would be four neutral Network Access Points, each in a different city and each operated by a different company, where competing backbone providers would connect and pass traffic between them. A routing arbiter to maintain stable path advertisement across the competing networks. A separate high-speed backbone to preserve the scientific mission after the public network was handed off. And a four-year phased subsidy for regional networks, declining from ninety-five percent in year one to zero in year five, forcing the academic networks to buy bandwidth from competing commercial carriers on commercial terms.

On April 30, 1995, the NSF decommissioned the NSFNET backbone. The public network disappeared and the commercial internet that replaced it emerged within the conditioned space. It had competing providers interconnecting at neutral points, open protocols governing the traffic and no single carrier controlling the route. The person who had built the public network also designed the terms under which it became private.

The internet happened because of three decades of institutional decisions. It was formed by a consent decree that fenced the telephone monopoly out of data processing and procurement conditions that mandated open interfaces and shared code. It had a protocol process that placed foundational knowledge in the commons, a sovereign mandate that enforced the open standard, and a privatisation designed to prevent capture during the transition. Combined they built the open substrate before the commercial layer arrived.

The consent decree opened science that had been locked inside a monopoly. The IBM unbundling separated a product that had been hidden inside a bundle. Both were corrections applied after the distortion had formed. But the internet's conditions were applied before the distortion could form. The open route was built before any carrier could turn it into a tollbooth. And because the road was open, conditioned, and neutrally interconnected, no single company could own the path between the builder and the user.

### What kept the ladder standing

Each of the three openings cut a different loop. The consent decree cut the loop between research, patents, equipment, and monopoly revenue and fenced the monopoly out of the markets its science would create. The IBM unbundling cut the loop between hardware ownership and software value, giving the program its own price and its own market. The network's open protocols and conditioned privatisation cut the loop before it could form, preventing any carrier or vendor from owning the distribution layer before the commercial internet arrived.

The institutional tools were different each time. A consent decree negotiated against military resistance. A last-day filing that locked a five-year investigation into the courts before the political window closed. Procurement conditions that mandated open interfaces, shared code, and neutral interconnection. But the function was the same: each prevented the owner of a foundational layer from controlling every layer above it.

And in each case, the function was performed by specific people who understood the structural problem they were solving. An attorney who spent a decade building a case from the Walker Report's findings. Career prosecutors who ran a five-year investigation across three administrations. A developer at an independent software firm who used a patent to assert that the thing IBM gave away for free was a distinct creation. An Attorney General who filed the case that would reshape computing's economics on his last day in office. A psychologist who recognised that the telephone network was the wrong architecture for computing. A program director who spent fourteen years inside the military watching proprietary networks choke, and designed a public backbone and its privatisation to prevent the same thing from happening to the internet.

All these were acts of institutional engineering that were as deliberate, as skilled, and as consequential as anything that was built in Michael Dell's dorm room and Steve Jobs's garage. None of this required public institutions to replace private enterprise. In every case the purpose was narrower: prevent the owner of a foundational layer from turning private control of the route into control over the entire economy above it. The result was not less market activity. It was more. More firms could build, compete, and fail on their own terms, because access to the underlying infrastructure no longer depended on the permission of the company that owned it.

## The escape

The clearest demonstration of how different the escape economy was from the industries that preceded it came when it crashed.

Between 1995 and 2000, the commercial internet produced a speculative mania that rivalled anything the railroad era had generated. Capital poured into companies built on internet economics  (online retail, web portals, content platforms, delivery services, social networks) before the term existed. Valuations detached from revenue and companies that had never produced a profit were priced as though the entire economy would flow through their servers within a decade. By March 2000, the Nasdaq had climbed above 5,000. By October 2002, it had fallen below 1,200. Trillions of dollars in market value disappeared. Pets.com, Webvan, eToys, Kozmo, Boo.com, and hundreds of others shut down entirely.

It looked, on the surface, like the same story running again. Capital deployed ahead of demand, then speculative excess, then the inevitable collapse. The comparison to the railroad panics and the utility pyramid crashes was made at the time and has been made ever since.

But the seven markers that defined the old bind did not apply to the application-layer boom that people usually mean when they invoke the dot-com crash.

Pets.com, Webvan, and eToys were not building shared physical infrastructure. They were renting it. The fibre, the routers, the backbone networks, the interconnection points were being built and expanded by carriers operating within the conditioned architecture that preceded the commercial era. The application-layer companies operated on top of an open substrate they did not own or finance. When they failed, the substrate remained available to the next builder.

The speculative capital was financing business operations (warehouses, marketing, customer acquisition, delivery fleets). These could be shut down when the money ran out. A warehouse can be closed or repurposed but a railroad cannot be unbuilt. The capital destruction was severe but it did not leave behind physical infrastructure carrying debt that had to be serviced whether the customers came or not.

The consolidation that followed the crash ran in the opposite direction from the old pattern. In the railroad era, the crash drove consolidation *during* the crisis ( e.g. Morgan reorganising bankrupt lines into fewer, larger systems). In the dot-com aftermath, the open substrate remained open and the barrier to building on it remained near zero. A small team in 2003 could still rent a server, write an application in open-source languages, and reach every internet user on the planet for a few hundred dollars a month. Google was built on commodity hardware running Linux while Facebook launched from a dormitory. The crash did not raise the barrier to entry because the conditions that had been set held it down.

But the dot-com crash was not the only financial crisis of that era. Underneath the application-layer companies that rented infrastructure, a parallel disaster was unfolding among the companies that built it. Between 1996 and 2001, telecommunications carriers issued over $500 billion in new bonds to lay fibre optic networks across the country and under the oceans. Capital expenditure grew at 28 percent annually while revenue grew at 10 percent. By the peak of the buildout, 95 percent of the fibre that had been laid was dark, all installed in the ground but never lit. WorldCom collapsed under $30 billion in debt. Global Crossing filed for bankruptcy with $12.4 billion. Winstar's $4 billion network was sold for $42.5 million. This was the old bind's pattern, running in real time: massive physical infrastructure financed ahead of demand, debt that could not be serviced, and builders destroyed by the gap between what they spent and what the market would pay.

But the infrastructure survived its builders. Fibre in the ground has a useful life of fifty years, and the civil engineering that put it there (the trenching, the conduit, the rights of way) represents up to eighty percent of deployment cost. The bankrupt carriers' networks were acquired at extreme discounts by successor companies who could operate them without the debt. The dark fibre was gradually lit as demand eventually caught up. And the backbone remained open. Standardised protocols governed the traffic and neutral exchange points kept the interconnections decentralised. The fibre glut itself forced carriers to lease capacity on open, non-discriminatory terms because there was too much of it to hoard. The conditions that had been built into the network's architecture held through the crash of the companies that built the physical layer beneath it.

The infrastructure survived its builders but that alone was not new. Railroad track had survived the Panic of 1893. The electrical grid had survived the holding company collapses. What was different was that the bankruptcy wrote down the debt without handing any surviving carrier durable control over the application layer above. The open protocols and neutral interconnection points held and the route stayed open.

And unlike the railroads, the utilities, or the telephone system, the internet's dissolution was directed outward. It was pointed at the intermediaries and industries it displaced, not at the revenue base funding the internet itself. Newspapers lost advertising revenue to Google, but newspapers were not Google's paying customers, they were just the previous holders of the position Google took. Retail chains were displaced by Amazon, but Amazon's actual customers got lower prices and wider selection. Travel agents were replaced by online booking, but travellers using the service got better deals. The dissolution was notable and the damage to the displaced industries was severe. But the internet's own pricing mechanisms worked because the technology's success reinforced its customers' ability to pay rather than undermining it. Advertisers could afford more advertising because Google made it more effective. Shoppers could buy more because Amazon made buying cheaper. There was no reflexive double bind in which the better the internet worked, the more it destroyed the pricing mechanisms of the people funding it.

And the owners of the physical infrastructure (the backbone providers, the carriers, the telcos) did not capture the application layer because they could not. The architecture had been designed to prevent it. Open protocols governed the traffic and neutral interconnection points prevented any single carrier from controlling the route. The conditioned privatisation had done exactly what Stephen Wolff had designed it to do. The pipe owner could not control what moved through the pipe.

The dot-com crash was a speculative mania built on top of an open substrate.  And the substrate survived because it was architectural. You cannot put TCP/IP into receivership or bankrupt an open protocol. The conditions that allowed the PC/Internet revolution to happen (the consent decree, the unbundling, the conditioned privatisation) were structural. They held through the worst financial crash the technology industry had experienced.

But those conditions were not permanent. Consent decrees can be lifted, procurement rules can be rewritten and institutional boundaries can be redrawn. The machinery that prevents capture can be allowed to rust through neglect, ideology, or the patient lobbying of the companies it once fenced in. And when the pattern returns, it returns faster and aimed at a different target. The railroad bind took decades to form. The electrical pyramids built over a generation. AT&T's closed loop tightened across half a century. But AI is assembling the same configuration in just a few years. And the older technologies disrupted specific sectors of the economy (agricultural logistics, energy supply, communication intermediaries)  while AI cuts across all of them, attacking the implementation layer that runs through most of the economy's output.

AI's double bind described in [Part 1](https://threadbaire.com/blog/posts/the-silicon-industrialists-part-1.html) is the same structural mechanism, compressed in time and aimed at the economy's centre. But in every previous case, the infrastructure's value grew because its customers' need for it grew, they asked for more freight, more power, more calls, every year. The question that separates AI from every case in this article is whether its customers will need more of what the infrastructure provides as the technology succeeds, or less. But how that question arrived and how the conditions that had answered it in every previous case were dismantled is the subject of Part 3.