← Blog

A Tale of Two Earnings - Microsoft, Apple, and the Circle Between Them

Lida Liberopoulou ·27 July 2026 · CC BY-SA 4.0

Raw .md ↗ Share on LinkedIn ↗ Share on X ↗

Independent research. Available for advisory, research collaboration, and expert consultation. Get in touch

On Sunday, Nvidia offered to guarantee roughly $250 billion in financing so that OpenAI can lease a data center complex it cannot fund on its own. The chip supplier underwrote its own customer's debt. The same week, Google reported its first negative free cash flow quarter in twenty-two years while booking $99 billion in unrealised gains on stakes in companies that buy its cloud compute. Amazon's Anthropic position contributed $12.3 billion to quarterly income from a single price adjustment.

Three companies, three instruments, one mechanism: the infrastructure costs more than the revenue base can fund, so the supplier finances its own demand.

On Wednesday, Microsoft reports. On Thursday, Apple. They sit on opposite sides of this pattern. This article makes six falsifiable predictions about what each company's reporting structure will make visible and what it will leave impossible to calculate. A scorecard with pass/fail conditions is included.

On Sunday, the Wall Street Journal reported that Nvidia is in talks to guarantee roughly $250 billion in financing so that OpenAI can lease a 10-gigawatt data center complex that SoftBank is building in Ohio. The project's total cost is expected to exceed $500 billion. Lenders were unwilling to fund the lease without a creditworthy guarantee behind it. So the company that sells the chips that will fill the data center, Nvidia, offered to provide one.

In effect, the chip supplier is underwriting its own customer's debt so that customer can afford to buy its chips.

Several people have already noted that Lucent did this in the 1990s. It lent telecom startups money to buy Lucent gear, booked the revenue, and collapsed when those customers went bankrupt. But this comparison is incomplete, because it answers the wrong question.

The question isn't whether vendor financing is risky. Everyone who remembers 2001 knows it's risky. What needs to be asked here is why the most valuable semiconductor company in history has to do it at all.

The disturbing answer is that the infrastructure costs more than the revenue base can fund.

OpenAI is running at about $25 billion in annual revenue and expects to burn $27 billion in cash this year. Lenders declined to fund the site without a stronger promise behind the paper. Nvidia raised its hand because if OpenAI can't build the data center, Nvidia can't sell the chips that go inside it.

This is the same mechanism that just showed up in Google's Q2 earnings. Total revenue rose 24%, Cloud revenue rose 82%, and at the same time the company reported the first negative free cash flow in its 22-year history as a public company. Capex doubled year over year to $44.9 billion in a single quarter. But the reported net income of $112 billion included roughly $99 billion in unrealised gains on investments. And these are investments in Anthropic and SpaceX, companies that are themselves Google Cloud customers.

Anthropic's valuation reportedly tripled during the quarter. Alphabet holds a significant stake in it. The $99 billion in total unrealised gains came from both Anthropic and SpaceX, though Alphabet did not separate the two investments' contributions in its filing and this is itself a disclosure choice worth noting.

Strip those gains out and adjusted EPS missed consensus. The revenue is substantial, but the independence of the demand is the unanswered question here. And the reported profit was dominated by non-cash gains on assets embedded in the same investment cycle that drives the spending. Cash is going out faster than it comes in.

Nvidia guarantees its customer's debt so that customer can lease a data center full of Nvidia chips. Google invests in Anthropic, books the rising valuation as profit, and counts Anthropic's compute purchases as Cloud revenue. Amazon has combined an equity investment in Anthropic (carried at roughly $74 billion as of March 2026) with a commercial agreement under which Anthropic buys AWS compute using Amazon's own chips. In a single quarter, the Anthropic position contributed $12.3 billion to Amazon's income from an upward price adjustment. Amazon has a convertible note with an embedded commercial agreement rather than a debt guarantee or a mark-to-market equity stake. So the instrument is different but the function is identical.

Three companies, each using three instruments that serve one mechanism. In each case the supplier is funding, financing, or underwriting the customer that buys its product. But the reason they exist is the same: the infrastructure costs more than the businesses it serves can pay for on their own. So the supplier finances its own demand.

On Wednesday, we get Microsoft's report. On Thursday, Apple's. They make a useful pair, because they sit on opposite sides of this pattern.

Microsoft is inside the same circle as Nvidia, Google, and Amazon. It holds roughly 27% of OpenAI, sells OpenAI compute through Azure, absorbs a share of OpenAI's operating losses into its own income statement, and promotes an adjusted earnings figure that strips the OpenAI effect out entirely. It funds the customer, sells to the customer, and then asks the market to evaluate it as though the customer doesn't exist.

Apple is outside all of this. It has no equity stakes in frontier labs, no unrealised gains or absorbed losses from AI lab investments and no cloud business selling compute back to companies it funds. It rents its AI capability (the Gemini deal with Google) rather than building and financing the infrastructure at hyperscaler scale. Apple Intelligence is running partly on its own Private Cloud Compute and partly on Google's models through the Gemini integration.

That doesn't mean Apple is free of risk. It is exchanging capital exposure for supplier exposure with its AI capability depending on models and infrastructure it doesn't control. But that dependency doesn't show up in the income statement or reshape the balance sheet. It's a different kind of vulnerability on a different timeline.

So here I am not predicting whether either company beats revenue or earnings estimates. But I'm making a small bet on what their reporting structures will make visible and what they will leave impossible to calculate.

Microsoft

The OpenAI line

Microsoft holds roughly 27% of OpenAI. Google marks its Anthropic stake to market and books the change as income. But Microsoft accounts for OpenAI under the equity method. In practice this means Microsoft absorbs a share of OpenAI's operating losses directly into its own income statement, calculated through a liquidation-value formula tied to what Microsoft would receive if OpenAI's assets were distributed today.

Last year that meant OpenAI's losses dragged Microsoft's net income down by roughly $2 billion over nine months. Then OpenAI recapitalised in October 2025. Microsoft's ownership stake was diluted, and under the accounting rules, dilution in a company whose value has risen triggers a gain ($5.9 billion of it) flipping the full-year OpenAI line from a drag into a $4.5 billion contribution to net income.

OpenAI is still burning cash, nothing changed there. What changed was a funding event that the accounting rules translated into a gain. Whether the OpenAI line prints as gain or loss on Wednesday will depend on both OpenAI's allocated operating results and any financing or ownership event during the quarter. The sign alone will not tell us whether the underlying business improved. Watch it, and then ask what the same relationship would look like under Google's treatment, where a different instrument produces a different number on the same page.

The silence

Earlier this year, Microsoft publicly disclosed that OpenAI has committed to purchasing $250 billion of Azure compute over the coming years as part of their restructured partnership announced last October. That number is on the public record. It supports the growth narrative, and the market can see it.

Last quarter's earnings call offered a glimpse of how concentrated that growth already is. Microsoft disclosed that its commercial remaining performance obligations (the total value of contracted future revenue) reached $627 billion, up 99% year over year. Excluding OpenAI commitments, the growth rate was 26%. One customer accounts for the majority of the acceleration in Microsoft's forward order book. That customer is the company Microsoft funds.

What Microsoft has never disclosed, and almost certainly will not disclose on Wednesday, is how much of Azure's quarterly revenue actually comes from OpenAI right now. Azure's AI revenue will be reported as a single growth number. Analysts will celebrate or question that number. But nobody on the call will be told what portion of it was recognised from the company that Microsoft itself funds, whose operating losses Microsoft absorbs into its own income statement, and whose compute purchases flow back to Microsoft as revenue.

The long-term commitment is public and supports the growth narrative. Current revenue concentration remains aggregated. The effect is asymmetric: the future promise is visible, the current dependency is not.

The margin

Two weeks ago, Microsoft voluntarily disclosed a specific profitability figure: a 3% "accountability margin." This is something it had never done in twenty-five years of running Xbox. Since its inception, the gaming division absorbed billions in losses across multiple console cycles without ever breaking out how much it was actually losing.

The timing is worth noting. Xbox is under political and regulatory scrutiny after the Activision acquisition. The 3% margin appeared at the point where continued silence about profitability carried more risk than disclosure.

No equivalent number exists for any AI business line. Azure AI revenue is bundled into Azure's overall segment. Copilot revenue is folded into Microsoft 365. The capital expenditure going into AI infrastructure (which has already reached $80 billion in the first nine months of the fiscal year) is not matched against the AI revenue it generates in any public reporting.

On Wednesday, no analyst on the call will receive an AI margin figure comparable to what Xbox just got. The pattern is consistent: the number appears when disclosure carries less risk than silence. But it does not appear where the spending still needs the growth narrative to justify it.

Apple

The income statement

Apple holds no equity stake in any frontier AI lab. It has no position in OpenAI or Anthropic, no convertible notes, no preferred stock, and no equity derivatives tied to lab valuations. More to the point, Apple's capital allocation history suggests it never will. Apple acquires companies outright (usually small, for talent or intellectual property) or it negotiates commercial agreements from a position of supplier leverage. It does not take minority stakes in companies it cannot control. The Gemini deal follows this pattern exactly: Apple rents the capability rather than buying a piece of the company that provides it.

When Apple reports on Thursday, there will be no AI lab line in the income statement. No gains to celebrate or losses to absorb, no adjusted earnings measure needed to strip either one out. Microsoft promotes a non-GAAP figure that removes OpenAI's effect from its results. Apple has no comparable effect to remove because joining the circle would break its own operating logic. The income statement already reflects the business without adjustment, because the business has never included a paper position in a cash-burning customer and the company's entire capital deployment model is designed to keep it that way.

This produces a different economic test. Microsoft must eventually prove that AI revenue can pay for the infrastructure built to produce it. Apple only has to prove that AI helps preserve the value of its existing products. One company must justify the cost of building. The other only has to justify the cost of renting.

The silence nobody questions

Apple has not disclosed the financial terms of its Gemini deal with Google, and no SEC filing contains a specific dollar figure for Private Cloud Compute capital expenditure. This is the same kind of non-disclosure that Microsoft practises with Azure's OpenAI revenue. But the market response will be entirely different.

The Gemini deal also makes Apple's position more complicated than simple independence. Google already pays Apple an estimated $20 billion or more annually for default search placement. Now Apple buys AI capability from Google. The financial relationship runs in both directions and neither side discloses the terms of either flow in full. Apple sits outside the infrastructure financing circle, but its AI capability is partly downstream of a company that is deep inside it.

Meanwhile, Apple keeps its own AI infrastructure (Private Cloud Compute). This is small enough to disappear inside a business that spent $4.3 billion on total capital expenditure in six months, less than Microsoft spent in two weeks. That scale is a choice: Apple's AI ambitions are currently sized to what it can rent without the rental reshaping the financial structure. That strategy holds only while frontier models remain cheap enough to rent and available enough to switch between. Apple's missing number here is the price and durability of supplier access.

On Thursday, no analyst will press Apple on AI spending because it hasn't reached the threshold where silence becomes suspicious. The same non-disclosure is tolerated at one company and scrutinised at the other. The difference is whether the spending is large enough to threaten the balance sheet.

The cash

In the most recent quarter, Apple returned $21.5 billion to shareholders, including $17.5 billion in buybacks and nearly $4 billion in dividends. In a single quarter, Apple returned more cash to its shareholders than it spent on capital investment in the entire first half of the year. Its board has authorised $163.8 billion in total buyback capacity. Trailing twelve-month free cash flow is approximately $110 billion.

Every dollar of that was generated by selling products and services to customers who paid with their own money. There are no unrealised gains on equity stakes in companies inside the same investment cycle, and no revenue recognised from a customer the company itself funds. The cash comes in from one direction and goes out in another, and the two flows have no structural relationship to each other.

The buyback programme is also a capital allocation statement the market doesn't read as one. If Apple believed it needed hyperscaler-scale AI infrastructure to protect its business, the buyback programme would be the first thing to shrink. It hasn't. Apple's revealed preference is that the best use of $110 billion in annual free cash flow is returning it, not building data centres. Google has moved into negative free cash flow for the first time in twenty-two years. Microsoft has spent over $80 billion on AI infrastructure in nine months with no public measure of what that spending earns. Apple is telling shareholders (through the cash flow statement, not the earnings call) that it does not need to join that race.

And nobody will ask Apple for an AI margin figure on Thursday, because nobody asks to see the circle at the company that isn't in it.

Scorecard

These predictions are filed before the releases. Here is how to check them on Friday.

Microsoft: The OpenAI line

Prediction: Microsoft's OpenAI-related income-statement effect will not provide a clean measure of whether OpenAI's underlying business improved.

Pass: The reported effect is materially shaped by equity-method loss allocation, financing, dilution, recapitalisation, liquidation preferences, or another ownership-accounting mechanism, without a corresponding operational bridge.

Fail: Microsoft provides enough operating information to connect the movement directly to improved or deteriorating OpenAI business performance.

Microsoft: Azure concentration

Prediction: Microsoft will report Azure or Azure AI growth without disclosing how much of this quarter's revenue was recognised from OpenAI.

Pass: No current-period OpenAI revenue contribution appears in the release, filing, or earnings call.

Fail: Microsoft provides a dollar figure or sufficiently narrow percentage range for OpenAI's share of Azure revenue.

Microsoft: AI profitability

Prediction: Microsoft will not disclose an AI-specific operating margin or return-on-invested-capital measure that matches AI revenue against the infrastructure spending required to produce it.

Pass: No such figure appears in the release, filing, or earnings call.

Fail: Microsoft provides a margin, ROIC, or equivalent measure for any AI-specific business line.

Microsoft: Cash conversion

Prediction: Azure and AI-related revenue growth will not produce proportional improvement in free cash flow, because the infrastructure required to fulfil that demand consumes cash faster than the recognised revenue replenishes it.

Pass: Capital expenditure rises faster than operating cash flow, causing free-cash-flow conversion to deteriorate even as revenue grows.

Fail: Microsoft demonstrates that incremental AI revenue is producing improving free-cash-flow conversion after the associated infrastructure spending.

The existing scorecard items test what Microsoft will not disclose. This one tests what the cash flow statement will positively show. Google has already crossed this threshold with its first negative free cash flow quarter in twenty-two years. The prediction is that Wednesday's filing will show the same pressure, visible not in what management says but in what the numbers do.

Apple: The control group

Prediction: Apple's results will contain no material frontier-lab investment gain or loss, no AI-adjusted earnings reconciliation, and no indication that AI infrastructure spending is displacing its capacity to return cash to shareholders.

Pass: Apple reports no material frontier-lab investment effect, continues cash returns at scale, and provides no indication that AI infrastructure spending is materially altering its capital-allocation capacity.

Fail: Apple discloses material AI infrastructure commitments or costs that substantially change the cash-return picture.

Separately, I will note two things that are not pass/fail tests but evidence about what the market considers material enough to interrogate. First, whether analysts demand an AI profitability measure from Microsoft but not from Apple. Second, whether Microsoft frames AI constraints as insufficient capacity (not enough data centres, chips, or power) rather than insufficient returns on the capacity already deployed. The first reveals which side of the circle attracts scrutiny. The second reveals the language companies use while trapped in forced spending.

Amazon: The Anthropic mark

Prediction: Amazon will report a material gain on its Anthropic equity position in Q2, the gain will be folded into headline earnings, and Amazon will not disclose Anthropic's contribution to AWS revenue.

Pass: Amazon reports a multi-billion-dollar Anthropic-related gain without separating Anthropic's share of AWS compute purchases from overall cloud revenue.

Fail: Amazon provides enough detail to evaluate both the investment gain and the commercial revenue from Anthropic independently.

Amazon's Q1 report showed a $12.3 billion income contribution from a single upward price adjustment on the Anthropic position. Anthropic's valuation reportedly tripled during Q2. If that revaluation is marked in this quarter's filing, the Anthropic line could rival operating income in scale. This is the same pattern Google just demonstrated, predicted in advance rather than analysed after.

The pattern is in the gap between what each company discloses and what the market permits it to leave aggregated. That gap tells you which side of the circle your investment sits on.

Primary sources

This analysis builds on the structural framework developed in The Silicon Industrialists series, which traces how AI's industrial cost structure and circular financing emerged. For the historical conditions that made the software economy's escape from this pattern possible (and how they were dismantled) see Part 2: Tech's Great Escape.