Big 5 AI Vendor Roundup: Week of April 27, 2026

Earnings week was supposed to validate the massive AI CapEx commitments. It did, sort of, but not the way the hyperscalers had hoped. Microsoft, Alphabet (Google), and Amazon (AWS) all reported on April 29. All three beat expectations on revenue, all three guided significant 2026 CapEx growth, and all three saw stock reactions ranging from muted to negative. Combined 2026 CapEx now tracks toward roughly $560 billion across the three (AWS leading the charge with $200 billion, with Microsoft and Google close behind), against disclosed annualized AI revenue of about $52 billion (Microsoft $37 billion run rate, AWS $15 billion+ run rate, Google undisclosed but likely in the range of $10 billion-$15 billion based on token throughput and growth disclosures). That’s roughly a 9-to-1 CapEx-to-AI revenue ratio. The math doesn’t work unless 2027 and beyond delivers big compounding growth, which both Microsoft CFO Amy Hood and Alphabet CFO Anat Ashkenazi explicitly warned investors to expect.

It’s also worth flagging that Microsoft’s $37 billion AI run rate includes “all revenue from model builders” running on Azure, which means OpenAI’s own Azure consumption is counted in Microsoft’s AI revenue.

Microsoft invested $13 billion in OpenAI → OpenAI spends a portion of that on Azure → Microsoft books it as AI revenue growth.

The same dynamic plays out at AWS with the $16.8 billion pretax gain on Anthropic this quarter. None of this is illegal or even unusual, but members reading these numbers should at least make a mental note and apply a circular-flow discount.

Outside earnings, OpenAI used the same week to renegotiate its Microsoft partnership and immediately land on AWS. AWS turned that landing into the flagship of its own “What's Next with AWS” livestream and added two more product drops on top. Anthropic quietly extended its enterprise footprint without a flagship release. Here’s the week, vendor by vendor.

Google: Earnings Crown, Cloud Constrained, TPUs for Sale

• Q1 2026 earnings delivered the operational standout of the week, with revenue of $109.9 billion (+22%), Google Cloud at $20.0 billion (+63%), and a Cloud backlog of $460 billion that nearly doubled QoQ. Cloud growth outpaced both Microsoft and AWS, making this Alphabet's best Cloud quarter ever. The catch is that 2026 CapEx was raised to $180 billion-$190 billion, with Ashkenazi flagging that 2027 will “significantly increase” again, against an undisclosed Google AI revenue number. Members signing 2026 commitments should push Google account teams for AI-specific revenue disclosures, not just Cloud growth percentages.

• Pichai admitted on the earnings call that Google Cloud is “compute constrained” and that Cloud revenue would have been higher with available capacity. That’s useful negotiating leverage to remember the next time a Google account team pushes urgency on a Cloud commit. If the answer is “we’re sold out,” the answer is also “we have less reason to discount.”

• Gemini Enterprise’s paid monthly active users grew 40% QoQ, with first-party model APIs now processing more than 16 billion tokens per minute, up from 10 billion the prior quarter, and revenue from products built on Google’s Gen AI models grew nearly 800% YoY. Gemini Enterprise is quickly becoming a full stack platform, not just a roadmap. Members evaluating agentic platforms should run Gemini Enterprise Agent Platform alongside Microsoft Copilot Studio and AWS Bedrock Agents in any 2026 RFP rather than defaulting to the incumbent.

• Alphabet announced it will begin selling TPU chips to select third-party customers for installation in their own data centers. This is the single most strategically significant move from any vendor this week. It positions Google as a direct NVIDIA competitor, or at least a complementary compute option in customer environments, and gives enterprises a credible non-NVIDIA training and inference option for the first time. Watch for NVIDIA’s response and price reaction over the next two quarters.

OpenAI: Cloud Independence and a New Front Door on AWS

• April 27: Microsoft and OpenAI amended their partnership. Microsoft remains primary cloud and gets first-shipping rights, but OpenAI can now serve any product on any cloud. AGI-related contractual triggers were removed. Revenue share continues through 2030 at the same percentage but is now subject to a total cap. Microsoft retains its IP license through 2032. The AGI-clause removal is the underappreciated piece because it eliminates the contractual escape hatch OpenAI had been holding over Microsoft in exchange for the freedom to expand its cloud footprint. Both sides traded leverage for predictability. On the same day, OpenAI released Symphony, an open-source specification for agent orchestration, signaling that OpenAI wants to influence the agent interoperability standard rather than cede it to MCP (Anthropic) or A2A (Google). It’s worth keeping an eye on it but too early to act.

• April 28: GPT-5.5, GPT-5.4, Codex, and Bedrock Managed Agents launched on AWS in limited preview, the day after the Microsoft restructuring. AWS commit holders can now spend on OpenAI models alongside Claude through the same Bedrock APIs they already use. (This is covered in detail in our “What's Next With AWS” note.) With OpenAI sitting inside both Bedrock and Foundry, model exclusivity as a competitive moat is officially dead. Members should build portability and multi-model fallback clauses into 2026 cloud and AI contracts while they still have leverage.

• April 29: In somewhat surprising news, OpenAI announced it had secured 10 Gigawatts of compute, a full three years ahead of the target it set for 2029 in its Stargate initiative in January 2025. To put that in perspective, that represents 4-5 MILLION high-end NVIDA GPUs. Yeah, that’s a lot of money, and an absurd amount of compute power.

• April 30: OpenAI issued an urgent macOS security alert requiring all users of ChatGPT, Codex, and Atlas desktop apps to update before May 8 following compromise of a third-party JavaScript library. The company rotated code-signing certificates as a precaution. That’s a reasonable response to a supply chain incident but a useful reminder that AI vendor desktop apps now sit inside the same software supply chain as everything else. Treat them as such in security reviews.

Anthropic: Quietly Extending the Enterprise Footprint

• Anthropic opened a Sydney office, named Theo Hourmouzis as general manager for Australia and New Zealand, and announced a partnership with NEC to train what is being described as Japan's largest AI engineering workforce. Anthropic is methodically standing up the enterprise sales foundations in regions where Microsoft and Google already have head starts.

• Claude connectors for creative platforms shipped, covering Blender, Autodesk Fusion, Adobe Creative Cloud, Ableton, Splice, and Affinity by Canva, with school trial partnerships at RISD, Ringling, and Goldsmiths. This is a legitimate vertical opening that Microsoft and Google haven’t matched. Members in creative or design-led organizations should evaluate Claude alongside Adobe’s native AI tooling before locking 2026 enterprise contracts.

• Anthropic reinforced its public commitment that Claude products will remain ad-free, explicitly positioning that stance as a competitive differentiator. The strategic read is that while Google leans further into Gemini-powered ads and OpenAI explores commercial integrations, Anthropic is pricing trust as a feature for risk and compliance committees. Use this in negotiations because Anthropic has incentive to discount enterprise deals to defend the trust positioning.

Microsoft: AI Revenue Validates, CapEx Spooks, Seat + Usage Pricing Coming

• Q3 FY26 earnings: Revenue of $82.9 billion (+18%), Microsoft Cloud at $54.5 billion (+29%), Azure +40%, AI run rate of $37 billion (+123% YoY). That looks like the most concrete commercial AI validation we’ve seen, but FY26 CapEx was guided to $190 billion (up 61%), with $25 billion of that attributed to component prices and Microsoft Cloud gross margin compressed to 66%. Margin compression is the line the press release buries. Members negotiating Azure commits in the back half of 2026 will be doing so with a vendor that has every reason to push for longer-term, higher-value commitments to defend that margin.

• In a signal of what’s to come in licensing and pricing models for all things AI, Nadella announced on the earnings call that the GitHub pricing model would change on June 1 to add usage to per-seat pricing. He said: “Any per-user business of ours, whether it’s productivity or coding or security, will become a per-user and usage business. That’s obviously already happening with GitHub Copilot coding with some of the business model changes we made this quarter, but it also speaks to the intensity of usage.” CIOs should expect that all Microsoft AI-related licensing will add usage to per-seat pricing and plan for the impact on their IT budget.

• Microsoft 365 Copilot crossed 20 million paid seats (up from 15 million in January), with the count of customers running more than 50,000 seats quadrupling and a single Accenture deployment now exceeding 740,000 seats. A meaningful share of Copilot’s reported growth depends on a small number of global services partner rollouts. Members evaluating Copilot expansion should ask account teams hard questions about deployment depth, daily active usage versus seat count, and renewal economics on the largest reference accounts.

• The OpenAI partnership amendment (covered above) preserves Microsoft’s IP license through 2032 and a capped revenue share through 2030, which is the structural win Microsoft needed. Loss of OpenAI exclusivity is a real but manageable hit. Microsoft Foundry now competes on platform features rather than model access, which is actually a healthier long-term position.

Amazon: AWS Reaccelerates, “What's Next” Plants Three Flags

• Q1 2026 earnings: Revenue of $181.5 billion (+17%), AWS at $37.6 billion (+28%, fastest growth in 15 quarters), AWS operating margin of 37.7%. Bedrock customer spend grew 170% QoQ, it processed more tokens in Q1 than all prior years combined, and it is now used by over 125,000 customers, including almost 80% of the Fortune 100. The reacceleration is real, but a $16.8 billion pretax gain on the Anthropic investment was included in net income. Strip that out and the numbers are solid rather than spectacular. The Anthropic gain is the best example in this earnings cycle of why AI revenue numbers need careful reading. Amazon invested in Anthropic, Anthropic’s valuation rose, and Amazon booked the unrealized gain. That’s not customer demand.

• “What’s Next with AWS” on April 28 was the strategic event of the week for AWS, with three big announcements: the OpenAI partnership flagship (GPT-5.5/5.4, Codex, and Bedrock Managed Agents on Bedrock), Amazon Quick desktop (the rebranded and repositioned Amazon Q Business, now an OpenClaw-style personal AI assistant available with Free and Plus tiers and personal credentials), and the expansion of Amazon Connect from a contact center product into four agentic AI solutions (Customer, Decisions, Talent, Health). (We cover this in our “What’s Next With AWS” research note.) Quick is a Trojan horse for enterprise adoption and the Connect family is the most aggressive vertical agent move from any hyperscaler this week. Members should evaluate Quick proactively before it shows up as shadow AI.

• Amazon reported CapEx of $44.2 billion in Q1 and full-year tracking toward $200 billion, while its chips business (Graviton, Trainium, Nitro) topped a $20 billion annual run rate. Trainium2 is largely sold out, and Trainium3 is nearly fully subscribed. Like the Google TPU announcement, the custom silicon story is the most differentiated piece of the AWS pitch. The supply story is more compelling than the price story right now. Members planning Trainium-based deployments in 2026 should secure capacity commitments early.

Our Take

If there was a single through line this week, it was that agentic AI is no longer the future of enterprise computing, it’s the present, and the Big 5 are racing to claim the layer of the stack where you’ll spend your 2027 budget. AWS plants Amazon Quick on the desktop and Connect agents in the contact center and supply chain, for talent and healthcare. Google rebrands Vertex AI as the Gemini Enterprise Agent Platform and bets the next decade on agents as the unit of cloud consumption (covered in our Cloud Next 2026 note from last week). Microsoft frames Q3 around the “agentic computing era.” OpenAI ships managed agents through Bedrock the day after winning cloud independence. Anthropic quietly extends Claude into creative agent workflows. The sales pitch from every vendor has shifted from “buy our model” to “run your agent fleet on our platform.” For IT leaders, that means the procurement decision is now an architectural one with a much longer tail.

The other through line is harder to read but worth keeping an eye on. Roughly $560 billion of CapEx against $52 billion of disclosed AI revenue is a bet on demand that hasn’t yet shown up in customer income statements at the scale required. Some of that demand is real, durable, and growing. Some of it is the same dollar circulating among three or four companies that all have an interest in calling it revenue. We’re not in the business of predicting market corrections, but we are in the business of helping members negotiate from a position of clear-eyed realism. Don’t let vendor enthusiasm about agentic transformation push you into commitments that assume the CapEx curve continues uninterrupted. Build optionality into 2026 contracts. Pilot agents in workflows where governance is nonnegotiable. And keep an eye on which of these vendors is still reporting 100%+ AI growth 12 months from now.

Want to Know More?

• AWS Doubles Down on Agentic AI in “What’s Next” Event

• Google Cloud Next 2026: It’s All About the Agents

• Claude Mythos Preview and Project Glasswing: What IT and Security Leaders Need to Know Now