$650,000,000,000
$650,000,000,000
An open letter to the hyperscalers, delivered after the bell
Gentlemen,
Tonight you report earnings. The four of you — Meta, Microsoft, Alphabet, Amazon — collectively committed somewhere between $600 and $650 billion in AI infrastructure spending for 2026. That is not a typo. That is a number larger than the GDP of Sweden, deployed in a single calendar year into data centers, custom silicon, power infrastructure, and cooling systems, in service of a technology whose monetization path is, to put it gently, still being workshopped.
I want to engage with that seriously for a moment. Not mock it. Not cheer it. Really sit with the arithmetic.
Meta is guiding to $115–$135 billion in capex this year — nearly double 2025's spend. The company laid off roughly 8,000 people to help fund it. Alphabet has telegraphed $175–$185 billion. Amazon's Andy Jassy, in his shareholder letter, wrote that his company plans to spend $200 billion this year — a 50%-plus increase from last year — and did not mention the Iran war, the Strait of Hormuz, or the fact that energy prices are doing something interesting to the cost of running a data center at scale. Microsoft's last quarter alone saw capex hit $29.88 billion. Up 89% year over year. Eighty-nine.
The question the market is asking tonight — politely, anxiously, with the barely concealed desperation of someone who has held these names through a very strange spring — is whether any of this is working.
Here is the problem with that question: it is now almost impossible to answer it cleanly, because the OpenAI situation muddied the water at exactly the wrong moment.
On Tuesday, a Wall Street Journal report landed claiming OpenAI has been missing its own internal revenue and user growth targets. The CFO, Sarah Friar, reportedly flagged to leadership that the company might struggle to cover its computing contracts if the top line doesn't accelerate. Oracle — which has a $300 billion, five-year partnership to supply OpenAI with compute — dropped more than 3% before recovering slightly. The Nasdaq shed 0.9%. The S&P closed at 7,138, down half a percent.
OpenAI is not a public company. It does not report earnings. It cannot be directly interrogated on a conference call. What it can do is function as a proxy for every anxiety the market has about AI monetization — and on Tuesday, it did exactly that. The unspoken question underneath every model demo, every Copilot integration, every Gemini benchmark: if the most visible AI company in the world is struggling to grow into its cost structure, what does that imply about the $650 billion the hyperscalers are committing behind it?
Microsoft is particularly exposed to this framing. Azure growth is the bellwether — the cleanest signal that enterprise AI spend is real and durable. The market is pricing in Azure acceleration. If that doesn't arrive tonight, the argument for the entire capex cycle comes under pressure from an angle that analyst notes cannot fully anticipate.
Let me offer a second variable that is not getting enough oxygen.
Energy costs are not what they were six months ago. Brent crude touched $112.70 a barrel on Tuesday — the highest since late March. WTI briefly crossed $100. The average gallon of regular gasoline in the United States hit $4.176 this week, the highest since August 2022. This matters for data centers in ways that run deeper than the utility line item: it affects DRAM pricing, it affects construction timelines, it affects the operational cost of running inference at scale.
IDC is forecasting DRAM at $9.71 per gigabyte in 2026, compared to $3.71 in prior years. The memory costs alone are restructuring the unit economics of every AI workload. Bridgewater estimates total Big Tech AI investment at roughly $650 billion for 2026, up from $410 billion in 2025. The cost per unit of compute is going up. The spend is going up. The revenue — we're about to find out.
The Iran war and the Hormuz disruption were not in any of these companies' original capital allocation frameworks. They are now an operating condition. None of the CEOs will say so directly tonight, but the cost pressures are baked into every number they'll present.
To be fair — and let's be fair — there are real signals of demand. Alphabet's Google Cloud grew 48% in Q4 2025, accelerating from 34% in Q3. The market is pricing in 58–59% growth for Q1. AWS has reportedly declined to raise prices despite confronting higher costs. Meta's ad machine has been the most consistent story in tech for two years running, with Wall Street modeling roughly $55.5 billion in Q1 revenue on 31% year-over-year growth. If advertising holds and Meta's Andromeda AI tools are genuinely improving ad performance, the stock arguably deserves to be 29% higher than it was a month ago. Arguably.
But here is the thing about a company that has laid off 8,000 people to fund a $125 billion capex year: the efficiency narrative and the ambition narrative are in direct tension. You can run lean, or you can build a new civilization in the Nevada desert. You cannot comfortably do both and then present it on a call as strategic coherence.
Microsoft, meanwhile, is down 12% year-to-date. The company offered voluntary buyouts to employees last month. It did not provide a formal capex guide, but analysts are modeling fiscal-year capex of $107.5 billion — the lowest among the four. That relative restraint should, in theory, make the numbers cleaner. In practice, it means Azure's growth rate tonight carries the entire weight of the Microsoft bull case, with no capex story to compensate if the cloud numbers disappoint.
Positioning heading into tonight is worth noting. Meta is up 29% in one month. Amazon 31%. Alphabet 28%. These are not cheap entry points for earnings that need to be merely good. They need to be definitively good. A beat without raised guidance, in this environment, with these valuations, after the OpenAI report, during an active energy crisis — that might not be enough.
A capex raise without a commensurate revenue raise is the worst outcome. Full stop. The market has already absorbed one story this week about an AI company that is spending faster than it can grow. Four more versions of that, across the companies that are building the infrastructure underneath the whole ecosystem, would not be a contained event.
What would be contained: a capex hold paired with cloud growth acceleration. Clean. Bullish. Narrative-confirming. The market rallies, the AI trade is vindicated, and we all spend another quarter not asking the harder questions.
That outcome is possible. I'd say it's roughly coin-flip likely across the four reports, taken together.
The other outcome — one or more capex raises, patchy cloud growth, cautious guidance language around energy and macro uncertainty — is also coin-flip likely.
$650 billion is a bet on a future that hasn't arrived yet. Tonight we get the first real accounting of whether the trajectory is right. Not the final verdict — it's one quarter — but the first genuine accountability moment for a spending cycle that has been running on consensus and momentum.
The bill is due after the bell. The receipts will tell us something important about what we're actually building — and who's going to pay for it.
Published April 29, 2026