The Big Risk Building in Big Tech

By AltIndex Research · 12 min read · July 28, 6:55 am

Google, Microsoft, Amazon and Oracle have booked $2.2 trillion in contracted cloud revenue between them. About half of it traces back to two customers that earn $72 billion a year and lose money on it.

Last wekk, Alphabet reported Q2 2026 revenue of $119.8 billion, up 24%, with Google Cloud accelerating to 82% growth and $24.8 billion. Net income came in at $112.2 billion, a 298% increase. Then investors read the footnote. A $99.0 billion gain on equity securities, driven by markups on Alphabet's stakes in Anthropic and SpaceX, added $77.1 billion to net income and $6.26 of the $9.11 in reported EPS. Strip it out and adjusted EPS was $2.85 against a $2.89 consensus, a small miss.

The stock fell. Not because of the accounting, but because of what sat next to it: $44.9 billion of quarterly capital expenditure, full-year guidance raised to $195 billion to $205 billion, and the first negative free cash flow quarter since Alphabet went public in 2004. The largest reported profit in company history and the first negative free cash flow quarter in company history arrived in the same press release.

Two Customers, $2.2 Trillion in Commitments

Every hyperscaler now leads with backlog. Oracle ended fiscal 2026 with remaining performance obligations of $638 billion, up 363% year over year. Microsoft reported $627 billion in commercial RPO at the end of its March quarter, up 99%. Google Cloud's backlog reached $513.9 billion in Q2, up more than $50 billion sequentially. AWS reported $364 billion in Q1 and told investors that figure excluded a separate Anthropic agreement worth more than $100 billion.

Add them together and you get roughly $2.2 trillion of contracted future cloud revenue. The composition is where it gets uncomfortable. The Information reported that Anthropic's five-year, roughly $200 billion commitment to Google Cloud, which begins in 2027 and covers five gigawatts of capacity, accounted for more than 40% of Google's backlog at the time it was disclosed. Oracle's backlog is anchored by a reported $300 billion OpenAI contract tied to Stargate. OpenAI has committed roughly $250 billion of additional Azure spending, and Microsoft's RPO grew 26% year over year once those commitments are excluded rather than 99%. Bloomberg reporting puts OpenAI and Anthropic commitments at about half of the AWS number once the April Anthropic deal is included. Andy Jassy pushed back on that framing directly on the Q1 call.

Contracted cloud backlog by provider, latest reported quarter. Lab-linked shares are reported estimates from The Information, Reuters and Bloomberg, not company disclosures. Only Microsoft breaks out an ex-customer growth figure.

"it is not just one customer or two customers"
Andy Jassy on the AWS backlog, Amazon Q1 2026 earnings call

No hyperscaler discloses customer-level backlog, which is the whole problem. Investors are being asked to underwrite a decade of capital spending against a demand signal whose concentration can only be reconstructed from press reports and analyst notes.

The Money Moves in a Circle

The concentration would be easier to assess if the customers were funding themselves. In April, Reuters reported Google plans to invest up to $40 billion in Anthropic, the same company that has committed roughly $200 billion back to Google Cloud. Google has also agreed to guarantee lease payments at five Anthropic data centers, which allowed Anthropic to secure about $35 billion in loans.

Amazon has put $8 billion into Anthropic and agreed in April to invest up to $25 billion more, alongside an Anthropic commitment to spend over $100 billion on AWS across ten years and take up to five gigawatts of Trainium capacity. Microsoft and Nvidia committed up to $15 billion combined to Anthropic in late 2025, paired with a $30 billion Anthropic commitment to Azure. Nvidia took a $30 billion equity stake in OpenAI in February and is now in talks to guarantee as much as $250 billion of OpenAI data center leasing, on top of more than $540 billion of similar arrangements announced this year.

Alphabet's quarter is the cleanest illustration of where this ends up in the financials. Anthropic's valuation went from $380 billion in February to $965 billion after its $65 billion Series H closed on May 28. Accounting rules require Alphabet to mark that private stake to the new round every quarter, so the revaluation flowed straight into reported net income without a dollar changing hands. The same private company that anchors Google's cloud backlog also produced most of Google's record quarterly profit, and one of those two facts is a paper entry. It also mechanically depressed Alphabet's trailing earnings multiple, which now sits near 16 times earnings against a $3.9 trillion market cap.

The Revenue Underneath the Commitments

Anthropic disclosed a $47 billion annualized run rate alongside its Series H in late May, up from roughly $9 billion at the end of 2025, with about 85% of revenue coming from enterprise and developer customers. OpenAI is at roughly $25 billion annualized, with $5.7 billion of revenue in Q1 2026 and $3.7 billion of cash burn in the same quarter. Both filed confidentially for IPOs in June, Anthropic on June 1 and OpenAI on June 8.

Those are extraordinary growth curves. They are also small relative to what has been contracted against them. Roughly $72 billion of combined annualized revenue currently sits underneath something close to $1 trillion of commitments and a 2026 capital spending program north of $700 billion across Amazon, Microsoft, Alphabet and Meta, itself up around 77% from about $410 billion in 2025.

Contracted backlog and 2026 capex versus the annualized revenue of the two anchor customers, in billions of dollars

The bet embedded in those bars is that both labs grow into the commitments. OpenAI's own projections show losses continuing through 2028 before an inflection, with cumulative burn of roughly $115 billion through 2029 and revenue reaching about $200 billion by 2030. Anthropic projects burn falling to roughly a third of revenue this year and 9% next year. Neither company expects to fund the next three years of compute out of operating cash flow.

Oracle Is the Stress Test

If you want to know how the market prices a backlog anchored to a single AI lab, Oracle has already run the experiment. ORCL closed at $119.90, down 60% from its September 2025 peak of about $302, even though RPO over that period went from roughly $455 billion to $638 billion. Backlog nearly doubled. The stock lost three fifths of its value.

The financing math is why. Oracle spent $55.7 billion on capex in fiscal 2026, guided to roughly $70 billion net for fiscal 2027, and posted negative $23.7 billion of free cash flow for the year. Total debt reached a record $153.1 billion by the February quarter. The company has been explicit that prepayments and customer-supplied GPUs, now totaling $75 billion, reduce the capital it needs to raise. The market repriced Oracle not on demand but on who is funding it and what happens to the asset base if that customer's plans change.

AltIndex AI Score and weekly share price for Oracle, July 2025 to July 2026

Our AI Score for Oracle dropped out of the 50s in the first week of December 2025, when the stock was still near $200. It has stayed in hold or sell territory every week since, and the stock has fallen another 40% over that stretch. The score currently sits at 35, firmly in sell territory, with job postings down 8.4% month over month. Insider activity did not help the bull case either: chairman emeritus Jeffrey Henley sold more than 60,000 shares across several tranches on June 24 at prices between $159 and $166, roughly a third above where the stock trades today.

What the Alternative Data Shows Across the Four

The repricing is not confined to Oracle. Microsoft trades at $389, down about 26% from its late October 2025 high near $524, and carries an AI Score of 36, our lowest reading of the group. Alphabet sits at $326.56, down roughly 17% from its early May peak near $395, with an AI Score of 56. Amazon holds up best at an AI Score of 58 and $231.39, supported by the strongest hiring signal in the group, with job postings up 57.8% month over month heading into its July 30 report. Monthly hiring swings are noisy at this scale, and Google's postings fell 38% over the same window, so treat single prints as direction rather than conclusion.

None of the four carries a buy signal in our data right now, which is unusual for companies compounding revenue at these rates. The scores are being held down by the same thing pressuring the stocks: fundamentals that look excellent on the revenue line and increasingly strained on the cash flow line.

The Case That This Is Fine

The bear framing has real counterweights. Anthropic's revenue grew from $9 billion to $47 billion annualized in under six months with gross margins reported around 70%, and more than 1,000 enterprise customers now spend over $1 million a year with it. Microsoft's backlog still grew 26% excluding OpenAI, which is normal enterprise cloud growth at enormous scale, and its AI business passed a $37 billion annual run rate, up 123%. Google says just over half of its backlog converts to revenue within 24 months, which is a much shorter duration than the headline number implies. Oracle's prepaid and customer-supplied hardware structure genuinely shifts risk onto the customer.

The demand is also visibly real at the usage layer rather than only the contract layer. Alphabet's first-party model APIs processed about 22 billion tokens per minute in Q2, up from 16 billion in Q1. AWS said Bedrock processed more tokens in Q1 2026 than in all prior years combined. If inference volume keeps compounding at that rate, the capacity being built gets absorbed regardless of which two logos signed the anchor contracts.

The Bottom Line

The risk here is not that AI demand is fake. It is that four of the largest companies on the planet have converted a concentrated, privately funded, loss-making customer base into the primary growth story investors are being asked to pay for, and none of them will tell you exactly how concentrated it is.

Two disclosures this week move the estimate. Microsoft reports fiscal Q4 on July 29, where the number that matters is RPO growth excluding OpenAI. Amazon reports July 30, where it is whether the backlog grows on breadth or on another single lab commitment. If the ex-lab growth rates hold up, the concentration is a footnote. If they decelerate while the headline backlog keeps climbing, the market has already shown you what it does with that, and its name is Oracle.

Sources are linked inline throughout. Primary references: AltIndex AI Score, job postings and price data; Alphabet Q2 2026 earnings release and 10-Q; Microsoft FY26 Q3 press release; Amazon Q1 2026 earnings; Oracle Q4 FY2026 earnings release; reporting from The Information, Reuters, Bloomberg, CNBC and Fortune. Lab-linked backlog shares are third-party estimates, not company disclosures. Stock prices are historical and for informational purposes only. This article does not constitute investment advice.

Disclosure: This article is for informational purposes only and does not constitute investment advice. AltIndex aggregates publicly available alternative data signals. Past signal performance does not guarantee future results. Always do your own due diligence before making investment decisions.