πŸ”₯ Breaking
Big Tech Doubles Down: Hyperscalers Lock In Close To $700B Of 2026 AI Capex

On a single Wednesday evening on 29 April, Alphabet, Amazon, Meta and Microsoft all topped Wall Street revenue forecasts and used the same earnings calls to push their 2026 AI capital spending plans even higher. Combined capex across the four hyperscalers is now on track to approach $700B this year, up from roughly $200B in 2024 and the largest single-year infrastructure surge in tech history.

The results made the AI tailwind impossible to ignore. Alphabet posted Q1 net income of $62.6B and 63% year-over-year growth in Google Cloud. Amazon delivered $181.5B in net sales and EPS of $2.78 against a $1.62 estimate. Meta revenue hit $56.3B with EPS of $10.44 versus a $6.67 consensus. Microsoft fiscal Q3 revenue reached $82.9B, with the company saying its AI business is now running at a $37B annualised rate, up 123% year over year.

But spending is the real headline. Amazon is on track for around $200B of capex in 2026 (most for data centres), Alphabet $175 to $185B, Meta $115 to $135B, Microsoft $120B+ and Oracle $50B. Investors did not all cheer: Meta dropped 6% after-hours on its raised capex guidance and Microsoft also slipped on AI buildout costs, while Alphabet and Amazon rose on cloud strength. Power, cooling and interconnect, executives said, are now the bottleneck this $700B is trying to break.

Why This MattersEditor’s Analysis

This is the week the AI build-out went from “big” to “openly historic”. Combined hyperscaler capex tripled in 24 months, but the gap between capex and AI revenue keeps widening. Microsoft is the only player whose disclosed AI run rate ($37B, up 123%) is even close to consistent with its $120B+ capex. Meta and Amazon still have to back-fit AI revenue into their numbers. The market read the gap on the night: Meta and Microsoft sold off after-hours, Alphabet and Amazon rallied. The four-name AI trade is splitting in real time.

For enterprises, the practical implication is that compute is now the binding constraint on every AI strategy you read about. Microsoft openly admits Azure AI is capacity-bound. OpenAI hit its 10GW Stargate target three years ahead of schedule and is already evaluating new sites. SoftBank is preparing a $100B IPO for Roze, a venture that exists to let robots build data centres faster. Capex commitments by 2030 will only make sense if AI revenue compounds another order of magnitude from here.

The bottom line: $700B is the real story of this earnings week, and it is now the lens through which to read every other 2026 AI announcement. Watch Q2 prints in late July. If Azure, Google Cloud and AWS all keep growing 30 to 60%, the spend looks rational. If not, expect louder questions about overbuild and a sharper Wall Street split between the four names.

Also Major This WeekRunners Up
πŸ“Š
Key Statistics & Insights
The numbers that defined AI this week
Apr 27 – May 3, 2026
The Week’s Defining TrendIntelligence Brief

The capex curve and the layoff curve have officially become the same curve. Every flagship story this week (hyperscaler capex, Microsoft’s $37B AI run rate, OpenAI hitting its 10GW Stargate target three years early, SoftBank’s $100B Roze IPO plan, Microsoft 365 E7 going GA on 1 May) is the same story told from a different angle. The four largest software companies in the world, plus a handful of would-be peers, are committing capital at a pace that only makes sense if AI revenue grows another order of magnitude from here.

Meanwhile, real workforces are being reshaped underneath, with 80,000 cuts in Q1 and almost half explicitly blamed on AI, and Chinese labs are responding with price wars rather than spend wars. DeepSeek’s 75% V4-Pro discount through 5 May is not just a launch promo. It is a deliberate signal that the open-source frontier intends to compete on cost, not on capex. The next 12 months will tell us which curve breaks first: Western capex commitments, or the labour pyramid underneath them.

πŸ”
This Week’s Spotlight
Deep-dive on the stories every AI professional needs to understand
πŸ€–
IBM Bob: From AI-Assisted Coding To AI-Assisted Delivery
On 28 April, IBM announced the global general availability of Bob, an AI development partner pitched not as another coding copilot but as an end-to-end software delivery platform. Bob covers planning, coding, testing, deployment and modernisation across the full SDLC, with the governance, security and audit controls enterprise CIOs keep insisting their existing AI coding tools are missing. Under the hood, Bob is a multi-model orchestrator: it dynamically routes each task to the best-fit model based on accuracy, performance and cost, drawing on a mix of frontier models including Anthropic’s Claude and Mistral open-source models alongside IBM’s own Granite family. 80,000+ IBM engineers have already been using Bob internally, with surveyed teams reporting an average 45% productivity gain. Pricing starts at $20/month for Pro (40 Bobcoins) and runs to $200/month for Ultra (500 Bobcoins).
Read on IBM Newsroom β†’
⚑
OpenAI Stargate Hits 10GW Years Ahead Of Schedule
On 29 April, OpenAI said its Stargate compute build-out had already passed the 10GW US capacity target it set in January 2025, three years ahead of the original 2029 deadline. Roughly 3GW were added in just the last 90 days, with five new data centre sites announced in partnership with Nvidia, Oracle and SoftBank. The original price tag was around $500B; OpenAI is now openly evaluating new sites for a follow-on phase. The milestone re-frames the whole AI infrastructure debate: 10GW is roughly the output of ten nuclear power stations, and OpenAI managed to lock it down before its own GPT-5.5 super-app strategy fully landed. Sam Altman framed it as a community-focused initiative as much as a tech one.
Read on OpenAI β†’
Spotlight: AI-Assisted Delivery Just Became A CategorySector Read

Cursor, GitHub Copilot, Claude Code and Replit dominate the inner loop where developers actually write code. But every CIO will tell you the inner loop is the cheap part of software. The expensive parts (release management, environment provisioning, governance, audit, mainframe modernisation) sit further out, in the SDLC, and almost no AI tool has tried to claim them as a single workflow. Bob is the first major attempt to move the value pool out to “AI-assisted delivery”, and the framing matters because it changes who buys the tool: not the developer, but procurement, security and platform teams.

The strategic read is that this is the same playbook IBM ran with Red Hat and watsonx: build for the regulated enterprise that the cool startups are still figuring out how to sell into. Watch closely for Cursor, GitHub and Anthropic to add governance and delivery framing to their roadmaps inside the quarter. The Cursor SDK on 29 April was a first move in that direction. Expect more.

Read IBM Bob β†’
πŸ’‘
ChatGPT & OpenAI News
All things OpenAI: models, products, business, and funding
β˜€οΈ
Claude & Anthropic News
Everything from Anthropic: Claude updates, research, and business moves
✨
Google Gemini News
Gemini models, Google AI products, and DeepMind research
πŸ’Ό
Corporate AI Developments
Big Tech AI investments, product launches, and strategic moves
πŸš€
AI Innovations
New models, open-source releases, and technical breakthroughs
πŸ“ˆ
AI in Business Applications
Real-world AI deployment across industries
βš–οΈ
Responsible AI
Policy, regulation, safety, ethics, and governance
🧠
Model Tracker
All frontier AI models: status, specs, and benchmark highlights
14 models tracked