• OpenAI ships multimodal updates • EU AI Act compliance dates clarified • Anthropic releases new safety evals • NVIDIA earnings beat expectations • New open-source LLM hits SOTA on MMLU
Anthropic AWS deal

Anthropic Just Bet $100B on Amazon — And There’s No Turning Back

Dario Amodei’s company finally stopped hedging on compute — and in doing so, handed its biggest cloud partner a decade-long grip on the future of Claude.

Key Numbers at a Glance

$100B

AWS commitment over 10 years

5 GW

Compute capacity secured

$380B

Anthropic valuation

A Strategic Pivot: From Discipline to Scale

Anthropic blinked first. After months of publicly defending spending discipline in the compute wars — months of watching OpenAI boast about its silicon advantages — Dario Amodei’s company walked into Monday morning with a $100 billion commitment to Amazon Web Services and a press release that reads more like a surrender note than a strategy memo. The company that once told investors there is “no hedge on earth” against overbuying compute just bought five gigawatts of it.

Five Gigawatts: The Scale of the Bet

That is not a typo. Five gigawatts. For reference, a large nuclear reactor generates roughly one gigawatt. Anthropic just signed up for the computational equivalent of five nuclear power plants’ worth of silicon, running continuously, for the next decade — all of it sitting on Amazon’s Trainium chips, a custom silicon line that Amazon has spent years building precisely to displace Nvidia on workloads like these.

Amazon’s Playbook: Back Both Sides

Amazon puts up $5 billion in fresh equity now, with a runway for up to $20 billion more tied to commercial milestones. That brings Amazon’s total stake in Anthropic to roughly $33 billion since its first check in 2023. The arrangement mirrors almost exactly what Amazon struck with OpenAI just two months ago — a $50 billion investment paired with a $100 billion cloud commitment — confirming what many suspected: Amazon runs the same playbook with both top labs, positioning AWS as the indispensable backbone of the AI era regardless of which company ultimately wins the model race.

“Our users tell us Claude is increasingly essential to how they work, and we need to build the infrastructure to keep pace with rapidly growing demand.”
Dario Amodei, CEO & co-founder, Anthropic

The Breaking Point: Infrastructure Under Strain

The timing matters enormously. Anthropic admits in its own announcement that “inevitable strain” on its infrastructure has already hurt reliability and performance for free, Pro, Max, and Team users — especially during peak hours. The company’s run-rate revenue surged from $9 billion at the end of 2025 to over $30 billion today, a trajectory so steep it snapped the pipes. Claude Code, the AI coding assistant that became unexpectedly viral among developers, drove a demand spike that the company’s infrastructure could not absorb cleanly. Anthropic responded by raising enterprise prices for heavy users, a move that telegraphed the crunch long before today’s announcement made it official.

Pressure from OpenAI

OpenAI watched all of this and piled on. Executives sent a letter to investors last week calling Anthropic’s compute position a “strategic misstep” and describing Claude’s operator as running on a “meaningfully smaller curve.” The message was pointed: our GPUs are bigger, our reliability is better, sign with us. Whether or not that narrative holds water technically, it lands politically — and Anthropic’s deal today reads partly as a rebuttal written in dollars rather than words.

Beyond Compute: Deep Integration with AWS

The arrangement extends well beyond raw chip access. Anthropic commits to purchasing capacity across Trainium2, Trainium3, and future Trainium4 chips, giving Amazon a decade-long anchor tenant for its custom silicon roadmap. Nearly one gigawatt of Trainium2 and Trainium3 capacity arrives by year-end 2026. In return, the full Claude Platform — not just API access through Bedrock — integrates directly into AWS accounts with shared billing and security controls. More than 100,000 enterprise customers already run Claude on Bedrock. That number will grow faster now that the checkout friction drops to near zero.

The New AI Power Structure

The deal signals something beyond a vendor relationship. It signals a posture. Anthropic, Microsoft, Google, and Amazon now occupy a peculiar quadrant of mutual dependency — Microsoft backs both Anthropic and OpenAI, Google holds equity in Anthropic while competing directly with it, and Amazon now writes nine-figure checks to both major frontier AI labs while building its own AI products. The hyperscalers no longer bet on winners. They buy the whole tournament.

Timeline of the Compute Wars

2023
Amazon writes its first check to Anthropic. Total committed: ~$4B.

Nov 2025
Microsoft invests up to $5B in Anthropic; Anthropic commits $30B in Azure compute.

Feb 2026
Amazon strikes a $50B deal with OpenAI. Amazon’s own capex guidance: $200B for the year.

Early Apr 2026
Anthropic expands Google and Broadcom partnerships for “multiple gigawatts” of capacity.

Apr 21, 2026
$100B AWS commitment + up to $25B from Amazon. Five gigawatts of Trainium capacity secured.

The Hidden Risk: Scarcity Always Returns

The tension hiding inside Monday’s announcement is structural. Amazon sells AWS compute to Anthropic. Amazon also builds its own AI products on AWS. When Anthropic’s demand spikes and Amazon’s own teams compete for the same clusters, who wins the allocation fight? The relationship works beautifully when capacity is abundant — it gets complicated the moment scarcity returns. And in AI, scarcity always returns.

No Middle Ground Anymore

Amodei once said he would rather lose customers short-term than overbuy compute and destroy margins. That calculus just changed, dramatically and publicly. The compute wars have a new combatant, and unlike the versions of Anthropic that hedged carefully and defended discipline, this one signs hundred-billion-dollar infrastructure commitments and bets on a chip vendor’s roadmap through 2034. Whether that turn proves visionary or desperate depends entirely on whether Claude’s revenue trajectory holds its current slope — or corrects. At $380 billion and climbing toward a rumored IPO valuation north of $800 billion, there is no comfortable middle outcome anymore. Anthropic just raised its own stakes to match its ambitions.

Related: Anthropic Just Declared War on Figma — and the Entire Design Industry Is Shaking

Tags: