AI Ready
ArchiveDaily Brief
Subscribe
Big Tech is spending $725B and supply is still the bottleneck
Daily Brief

Big Tech is spending $725B and supply is still the bottleneck

Microsoft, Alphabet, Amazon, and Meta all said capacity, not demand, is the constraint. The procurement implications start now.

By Haroon Choudery·May 1, 2026·7 min read

A newsletter by Seeko. Want the 3 workflows your company should automate this quarter, plus the AI teammates built to run them? Our 2-week Readiness Audit gives you both. Visit seeko.so to learn more.

THE WEEK IN AI
THE WEEK IN ONE SENTENCE

Four hyperscalers reported earnings on the same Wednesday and told investors the same thing: AI demand is now ahead of the infrastructure they can build, capex guides went up by tens of billions, and the office software you license every month became agentic by default while the bill was being written. I read the four transcripts back-to-back, and the constraint this week is supply, not demand.

THREE SIGNALS
01 • Compute

Capex got rewritten in a single afternoon

Microsoft, Alphabet, Amazon, and Meta all reported on April 29 and all raised 2026 AI capex. Microsoft is now guiding $190 billion in calendar-year 2026 capex, with $40 billion of that in Q4 alone and roughly $25 billion attributable to higher component pricing. CFO Amy Hood told investors that Azure supply will remain constrained through 2026. Across the four companies, the combined 2026 capex line is now tracking up to $725 billion, versus a prior consensus near $630 billion going into the calls. The supply-side bet and the demand-side print are both real. The gap is the constraint.

For an operator on a multi-year cloud contract written before this week, two questions are now worth asking. One: is your committed capacity guaranteed against the queue, or are you behind net-new bookings if you ask for more. Two: how does your renewal price compare to the spot price on Bedrock, Azure, or Google Cloud today, because the GPU-shortage premium is showing up in customer pricing in ways it wasn't six months ago.

02 • Agents

Your office software became agentic by default

On April 22, Microsoft made agentic capabilities in Word, Excel, and PowerPoint generally available for Microsoft 365 Copilot, Premium, and Personal and Family customers, with the agent features set as default. Six days later, Anthropic released nine new Claude connectors for creative tools, including a system-level Adobe integration that lets Claude orchestrate Photoshop, Premiere, and 50+ other Creative Cloud tools from a single conversation.

The pattern is the same in both cases. The model isn't the news. The default configuration is. Copilot has been able to write Excel formulas for a year, and Claude has been able to describe how to retouch an image for two. What changed is that the agent now executes inside the application, on company files, with whatever permissions the tenant left in place from when the same product was a sidebar. Microsoft also reported that paid Copilot seats now exceed 20 million, up 250% year over year, and that GitHub Copilot Enterprise reached nearly 140,000 organizations, tripling year over year. The footprint of agent-capable software inside enterprises is no longer hypothetical, and most tenants have not refreshed their Copilot governance posture for it.

03 • Distribution

The model wars stopped pretending to be model wars

Three data points this week describe a market that has moved past benchmarks and into distribution. Anthropic reported it crossed $30 billion in annualized run-rate revenue on the back of new compute partnerships with Google and Broadcom. Adobe and Anthropic shipped system-level Claude integration into Creative Cloud. Google used Cloud Next '26 the prior week to launch Workspace Studio, the A2A protocol at 150 organizations, and Project Mariner.

The frontier labs have stopped trying to win on a leaderboard and started trying to win on integrations. Anthropic now ships in Photoshop. OpenAI ships in Word. Google ships in Workspace and Chrome. The bet is that whoever owns the surface where the work happens owns the budget, and the model becomes a feature of the surface rather than the reason to pick it. For procurement, that collapses the multi-vendor decision into a stack decision, and the question worth asking by next Friday is which of these surfaces your employees are already using with or without IT's blessing.

UNDER THE RADAR

The most interesting number Microsoft printed on Wednesday wasn't actually capex. It was a commercial remaining performance obligation of $627 billion, up 99% year over year. Twenty-five percent of that, roughly $157 billion, is expected to be recognized as revenue in the next 12 months, up 39% year over year. That is the contracted backlog of cloud and AI commitments customers have already signed, and it nearly doubled in a year.

Two things sit underneath the headline. First, RPO at this scale is the cleanest indicator that enterprise AI spend is locked in for years, not months. Hood framed the structure of those contracts as a hybrid of seat entitlements and consumption-based usage rather than traditional seat licensing alone, which means the procurement model is changing under enterprise contracts in real time.

Second, a backlog this large means the queue Hood referenced is full of names. For operators with custom training or large-scale inference contracts, the conversation worth having with your account team this month is whether your committed reservations are guaranteed against the next 12 months of bookings or sit behind them. For everyone else billed downstream of that queue through Microsoft 365 or Bedrock, the supply-and-demand story most coverage led with this week understates how much of the demand is already on paper. The constraint isn't whether AI gets adopted. It's the order in which the queue gets served.

QUOTE OF THE WEEK

“Customers want predictability for budgets and procurement, and the seat-based pricing is just entitlement to some consumption. That is the way to think about it."

Satya Nadella, on Microsoft's Q3 2026 earnings call

The translation: the "X seats at Y dollars" line item your CFO is used to is becoming a base fee with a usage meter on top, on Microsoft's biggest products.

WHAT’S ON THE CALENDAR

See you Tuesday with the Ready Memo,
Haroon

Enjoying this issue?
Get the next Daily Brief as it lands.
Free. Four sends a week.
Keep reading

Get every issue, as it lands.