Published: July 27, 2026 | Reading Time: ~12 minutes | Channel: Business
Here's a number that should make you sit up straighter in your chair: $725 billion. That's what Microsoft, Amazon, Alphabet, and Meta plan to spend on capital expenditure this year — nearly double what they spent in 2025.¹ Another number: $35 billion. That's the combined annual revenue of every major pure-play AI company — OpenAI, Anthropic, Cohere, Mistral, Perplexity, and the rest — that this infrastructure is supposedly being built for.²
That's a 20-to-1 ratio of infrastructure investment to end-customer revenue.
If a restaurant spent $20 million on a kitchen that generated $1 million in sales, you'd call it a money-laundering front. When Big Tech does it, Wall Street calls it a "generational opportunity."
Let me be blunt: the AI infrastructure buildout is the most audacious bet in the history of corporate capital allocation. And the people telling you "this time is different" are the same people who'll tell you they saw the dotcom bust coming — after it happened.
Let's lay out exactly what's happening, because scale matters here. This isn't a few billion in R&D. This is industrial-policy-level spending conducted by four private companies.
| Company | 2025 Capex | 2026 Capex (Guided) | YoY Change |
|---|---|---|---|
| Amazon | ~$100B | ~$200B | +100% |
| Microsoft | ~$95B | ~$190B | +100% |
| Alphabet (Google) | ~$85B | $175–185B | +110% |
| Meta | ~$70B | $115–135B | +80% |
| Oracle | ~$21B | ~$50B | +136% |
| Combined | ~$371B | ~$660–725B | +85% |
Sources: Futurum Group, ValueAdd VC, CNBC, Fortune/JPMorgan.¹ ² ³ ⁴
The combined $660–725 billion figure doesn't include the Stargate project — the OpenAI-SoftBank-Oracle-MGX joint venture targeting $500 billion by 2029.⁵ It doesn't include the Middle Eastern sovereign wealth funds pouring billions into AI infrastructure. It doesn't include China's Alibaba ($53 billion over three years), ByteDance ($23 billion in 2026), or the EU's €200 billion AI Continent Action Plan.²
JPMorgan now projects global AI-related capital expenditures will reach $5.5 trillion through 2030 — up from their previous $5.1 trillion estimate — with hyperscaler spending alone hitting $1.1 trillion in 2027.⁴
These aren't conservative forecasts. They're straight-line extrapolations of a spending binge that's barely 18 months old.
Here's where it gets genuinely uncomfortable. Building data centers, buying GPUs at $30,000–$40,000 a pop, and wiring them together with InfiniBand networking doesn't come cheap. These companies are burning through cash at a rate that would make a venture-backed startup blush.
The damage, by company:³
Amazon: Morgan Stanley projects negative free cash flow of $17 billion in 2026. Bank of America says it could be negative $28 billion. Amazon filed with the SEC that it may seek to raise equity and debt — Amazon. The company that pioneered free cash flow as a religion.
Alphabet: Pivotal Research projects FCF will plummet roughly 90% — from $73.3 billion in 2025 to $8.2 billion in 2026. Alphabet already held a $25 billion bond sale in November 2025. Its long-term debt quadrupled last year to $46.5 billion.
Meta: Barclays analysts — who maintained their overweight rating, mind you — wrote that they now model "negative FCF for '27 and '28, which is somewhat shocking to us." Meta CFO Susan Li was asked about buybacks on the earnings call. Her response? The "highest order priority is investing our resources to position ourselves as a leader in AI." Translation: no buybacks for you.
Microsoft: The relative "good news" — FCF down only 28%, with a projected rebound in 2027. But Microsoft disclosed an $80 billion Azure backlog it can't fulfill because of power constraints, not demand issues.³
Let me put this in perspective: in 2024, these four companies generated $237 billion in free cash flow. In 2025, that fell to **$200 billion.**² In 2026, analysts are modeling a number that could approach — or dip below — zero in aggregate.
That's a quarter-trillion-dollar swing in cash generation in 24 months. If this were any other industry, CNBC would have a "MARKETS IN TURMOIL" chyron on permanent rotation.
The industry's defense goes something like this: "Yes, spending is high, but demand is insatiable. Every GPU is sold before it's built. Cloud backlogs are at record levels. This is supply-constrained, not demand-constrained."
All of that is true. And all of it misses the point.
The question isn't whether there's demand for the infrastructure. The question is whether the end users — the enterprises and consumers who are supposed to generate the revenue that justifies all this — will pay enough to make the math work.
Right now, the numbers don't add up:²
| Revenue Source | Annual Run Rate |
|---|---|
| OpenAI | ~$20 billion |
| Anthropic | ~$9 billion |
| Cohere | ~$150 million |
| Mistral | ~$400 million |
| Perplexity | ~$148 million |
| Total Pure-Play AI | < $35 billion |
The entire cohort of companies whose existence depends on this infrastructure generates less than $35 billion in combined annual revenue. Against $700 billion in annual spending, that's a 5% yield.
"But Peter, the hyperscalers also use this infrastructure for their own AI services!"
Fair. AWS hit a $142 billion annualized revenue run rate. Microsoft says its AI business is already larger than some established franchises. Alphabet's cloud backlog surged 55% sequentially to over $240 billion.²
But here's the problem: if the revenue is coming from the hyperscalers' own cloud businesses, and those cloud businesses are selling capacity to AI startups — many of which are funded by the same hyperscalers' venture arms or by debt arranged through the same banks underwriting the data center bonds — then what we're looking at isn't a market. It's a closed loop.
The Atlantic called it a "trillion-dollar ouroboros" — Big Tech advances money to AI startups so those startups can buy cloud services from Big Tech, which uses the revenue to justify more infrastructure spending, which requires more AI startup customers, which need more funding.⁶
It works until it doesn't.
If corporate earnings calls and sell-side research aren't enough to make you nervous, how about the United States Department of the Treasury?
A draft Treasury report — prepared for Secretary Scott Bessent, Fed Chair Kevin Warsh, and other financial regulators — warns that the AI sector is "particularly vulnerable to funding for data centers and other infrastructure projects drying up and sustained growth expectations going unmet."⁵
The career analysts who wrote the report found that these dynamics are "reminiscent of the dotcom crash" — but with a crucial difference: AI firms are "more deeply entrenched in the broader U.S. economy than their dotcom predecessors." A downturn would send shockwaves across stock markets, private credit markets, data center financing, cloud providers, chip manufacturers, and utilities.⁵
The Treasury report stopped short of predicting an immediate crash. But it concluded that investors are taking risks significant enough that "much of the financial system now rests on AI meeting its stated expectations for productivity gains and profitability."⁵
The Bank of England, the International Monetary Fund, and a Federal Reserve survey published in May have all flagged AI-linked equity valuations and debt-funded data center spending as destabilizing risks to the broader financial system.⁵
When the Fed, the Treasury, the IMF, and the Bank of England are all pointing at the same thing and saying "this looks risky," maybe — just maybe — it's worth paying attention.
The standard rebuttal to bubble warnings is: "This isn't like the dotcom era. These are real companies with real revenue and real products."
That's half right. These are real companies with real revenue. Alphabet generated $350 billion in revenue in 2025. Amazon cleared $638 billion. These aren't Pets.com with a sock puppet and a prayer.
But here's what's different — and genuinely more dangerous:
1. Concentration risk is off the charts. The Magnificent Seven — Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia, and Tesla — now account for one-third of the value of the entire S&P 500.⁶ In 1999, the top seven stocks accounted for roughly 17%. The market isn't just tech-heavy; it's seven-stock heavy.
2. The leverage is hidden. Unlike the dotcom era, when retail investors funded the bubble through E-Trade accounts and mutual funds, the AI buildout is financed through corporate bonds, private credit, and structured deals that are "complicated, opaque, and structured so as to be invisible on traditional balance sheets," as Columbia Business School's Stijn Van Nieuwerburgh has documented.⁶ Morningstar describes "buy-side indigestion" as lenders grow reluctant to meet Silicon Valley's demands.
3. It's all connected. AI infrastructure investment is responsible for "essentially all American GDP growth at the moment," according to The Atlantic.⁶ Without it, "we might be in a recession." The economy isn't just riding the AI wave — the AI wave is the economy right now.
4. The revenue is circular. When Nvidia sells GPUs to Microsoft, and Microsoft sells cloud capacity to OpenAI, and OpenAI is funded by Microsoft, and Nvidia's revenue growth justifies its valuation, which supports the entire AI thesis — you're not looking at a value chain. You're looking at a daisy chain.
I'm not telling you to sell everything and hide in a bunker with gold bars and canned beans. But I am telling you that the consensus narrative — "AI spending is fine because demand is infinite" — is intellectually lazy and historically dangerous.
Here's what you should actually do:
1. Audit your tech exposure. If you own an S&P 500 index fund, roughly one-third of your money is in seven stocks whose valuations depend on an AI revenue thesis that hasn't been proven. That's not diversification — that's a sector bet wearing an index fund costume. Consider whether your allocation matches your actual risk tolerance.
2. Watch free cash flow, not revenue growth. Revenue growth can be bought — just spend more on capex. FCF tells you whether the spending is generating returns. If Amazon, Alphabet, and Meta all report negative or near-zero FCF through 2027, the market will eventually notice. Don't be the last one to look at this number.
3. Separate the AI value chain. Not all AI bets are equal. Companies selling picks and shovels (Nvidia, Arista Networks, data center REITs) have real revenue today. Companies promising AI transformation without showing AI revenue are selling hope. Know which you own.
4. Build your own AI ROI case. If you're a business leader, the hyperscalers' spending spree means AI infrastructure costs will keep falling. The cost to run GPT-4 class inference dropped approximately 95% between 2023 and 2025.¹ But cheaper infrastructure doesn't automatically create business value. Before you sign that enterprise AI contract, know exactly what problem you're solving and how you'll measure success.
5. Keep 12-18 months of dry powder. If the Treasury and the IMF are right and the AI spending cycle hits a wall, the correction won't be limited to tech stocks. Credit markets will tighten, M&A will freeze, and the companies with cash will have their pick of distressed assets. The time to prepare for that isn't when it's happening.
Even the most compelling bear case has counterarguments. Here's what could make me wrong:
Jevons Paradox plays out faster than expected. As inference costs fall, AI usage could explode in volume, driving more infrastructure demand rather than less. Satya Nadella has pointed to this dynamic. If enterprises go from experimenting with AI to running it pervasively, the demand might actually justify the buildout.
The hyperscalers have $420 billion in cash. As of their latest quarters, Microsoft, Amazon, Alphabet, and Meta held over $420 billion in combined cash and equivalents.³ They can absorb years of negative FCF without existential risk. The "cash crunch" narrative overstates the immediate danger.
AI could deliver genuine productivity breakthroughs. If AI drives even 1-2% of sustained productivity growth across the economy, the ROI on $700 billion in infrastructure will look cheap in retrospect. The bet is enormous because the potential prize is larger.
Custom silicon changes the economics. Google's TPUs, Amazon's Trainium, Meta's MTIA, and Microsoft's MAIA all aim to reduce dependence on Nvidia's $30,000 GPUs. If these programs succeed, the cost per unit of compute could drop dramatically, making the same infrastructure budget go further.¹
Four companies are spending three-quarters of a trillion dollars this year building infrastructure for a revenue stream that currently generates one-twentieth of that — and Wall Street's response is "looks good to us, here's another trillion for 2027."
The AI buildout will produce genuine technological breakthroughs. It will create enormous wealth for some. But the idea that $700 billion in annual spending is safely underwritten by $35 billion in end-customer AI revenue requires a suspension of arithmetic that should make any serious investor deeply uncomfortable.
The companies that win from AI won't necessarily be the ones spending the most on it today. They'll be the ones whose AI investments generate returns that show up in free cash flow, not just in press releases.
When the music stops, what you want to know isn't who had the biggest GPU cluster. You want to know who had actual paying customers.
ValueAdd VC — "Big Tech's $725B AI Capex in 2026 — Up 77% From 2025." Company-by-company breakdown of hyperscaler AI infrastructure spending plans. https://valueaddvc.com/blog/big-tech-ai-capex-in-2025-microsoft-google-meta-amazon-and-the-spending-race
Futurum Group — "AI Capex 2026: The $690B Infrastructure Sprint." Detailed analysis of hyperscaler capex, pure-play AI vendor revenue, Stargate project, and global AI infrastructure race. https://futurumgroup.com/insights/ai-capex-2026-the-690b-infrastructure-sprint/
CNBC — "Tech AI spending approaches $700 billion in 2026, cash taking big hit." Free cash flow projections for Amazon, Alphabet, Meta, and Microsoft, including Barclays and Morgan Stanley analyst notes. https://www.cnbc.com/2026/02/06/google-microsoft-meta-amazon-ai-cash.html
Fortune — "AI spending boom accelerates as Big Tech pours trillions into infrastructure." JPMorgan's $5.5 trillion global AI capex forecast through 2030, Qualcomm's data center ambitions. https://fortune.com/2026/06/29/ai-spending-boom-accelerates-big-tech-trillion-infrastructure-qualcomm-cfo/
PYMNTS / US Treasury — "Treasury Report Warns AI Bubble Could Trigger Disruption." Draft Treasury report findings: AI sector vulnerability, systemic risk parallels to dotcom era, concentration concerns. https://www.pymnts.com/news/artificial-intelligence/2026/us-treasury-report-warns-ai-bubble-could-trigger-economic-shockwaves/
The Atlantic — "The AI Bubble Is No Ordinary Bubble." Annie Lowrey's analysis of the trillion-dollar ouroboros, concentration risk, hidden leverage, and AI infrastructure's dominance of GDP growth. https://www.theatlantic.com/ideas/2026/07/ai-economy-stock-market/688004/
All claims verified against Gold-tier (Reuters, Bloomberg, SEC, Federal Reserve) and Silver-tier (CNBC, Fortune, The Atlantic, Futurum Group) sources. Each source URL was scraped and confirmed accessible. Last verified: July 27, 2026.
The biggest risk in markets isn't being wrong — it's being wrong and crowded. 🎯
