Anthropic Locks In 5 Gigawatts of AWS Compute for a Decade

The Amazon Anthropic investment announced Monday is the largest single equity injection into the AI lab to date and locks in the longest cloud-infrastructure commitment yet seen between a hyperscaler and a frontier AI developer. Amazon will deploy US$5 billion now and up to US$20 billion more if Anthropic hits commercial milestones, bringing the maximum potential to US$33 billion against Anthropic’s reported US$380 billion valuation.

The Rio Times, the Latin American financial news outlet, reports that Anthropic’s reciprocal commitment is the headline figure for the cloud-infrastructure side: more than US$100 billion in spending on Amazon Web Services over the next decade, securing up to 5 gigawatts of compute capacity. That capacity covers current and future generations of AWS’s custom-designed Trainium and Graviton chips, with significant Trainium2 capacity coming online in the second quarter and nearly 1 gigawatt of combined Trainium2 and Trainium3 capacity by year-end 2026.

The structure mirrors Amazon’s February deal with OpenAI, in which the company joined a US$110 billion funding round with a US$50 billion contribution and a parallel US$100 billion cloud commitment. Amazon CEO Andy Jassy framed Monday’s announcement around custom-silicon performance: “Anthropic’s commitment to run its large language models on AWS Trainium for the next decade reflects the progress we’ve made together on custom silicon.”

Why the Amazon Anthropic Investment Lands Here

Anthropic CEO Dario Amodei said in the joint announcement that Claude usage growth has forced the infrastructure scale-up. The company’s annualized revenue run rate hit roughly US$30 billion in 2026, more than tripling from US$9 billion at year-end 2025. More than 100,000 enterprise customers now run Claude models on Amazon Bedrock.

The deal also restructures access architecture. AWS customers will be able to use the full Anthropic-native Claude console directly inside AWS, with their existing AWS contract and credentials — a tighter integration than Amazon Bedrock’s previous marketplace-style model. The move responds directly to Microsoft’s November 2025 US$5 billion Anthropic investment paired with a US$30 billion Azure commitment, which had created a competitive vector Amazon needed to neutralize.

As Rio Times reporting on the Project Rainier collaboration documented in late 2024, Amazon and Anthropic had already built one of the world’s largest compute clusters using Trn2 UltraServers with 64 Trainium2 chips each. Anthropic now uses more than 1 million Trainium2 chips to train and serve Claude.

The Mexico Connection

For Latin American readers, the most consequential dimension is infrastructure. AWS announced a US$5 billion Querétaro data center cluster in February 2024, spread over 15 years, designed to be Amazon’s second Latin American region after São Paulo. The cluster opened in 2025 and is now a strategic node in the global Claude inference network described in Monday’s announcement.

Anthropic’s deal expands Claude inference capacity across Asia and Europe to meet rising international demand — capacity that AWS’s Latin American region serves directly into Mexican, Brazilian, Argentine and Colombian enterprise customers. Mexican stock exchange BIVA, airline Aeroméxico, cinema chain Cinépolis and the local Banco Santander unit are among the existing AWS customers who can now access Claude through their existing contracts.

The hardware story is even more direct. Mexico’s computer and AI-server exports surged 145% in 2025 to overtake China as Foxconn, Inventec, Pegatron, Wiwynn and Quanta concentrated production in Chihuahua and Jalisco. The Trainium chips that anchor the Anthropic deal are designed in Austin and Israel but assembled increasingly inside the Mexican nearshoring corridor that has become the AI-hardware factory floor for North America.

Brazil and the Sovereign-Cloud Question

Brazil sits in a different position. The São Paulo AWS region has been operational since 2011 and hosts customers including Itau Unibanco, Magalu, B3 and the federal Serpro infrastructure. The Anthropic-AWS deepening means Claude becomes natively available to Brazilian government and enterprise users without separate procurement — a meaningful change for sectors negotiating LGPD data-residency requirements.

The Lula government’s sovereign-AI agenda, including the Itamaraty’s push for a regional Latin American AI policy framework discussed at recent Mercosul-EU consultations, faces a structural tension here. The deeper Claude becomes inside AWS, the harder it is for Brazilian regulators to argue that domestic alternatives could displace US AI services without crippling enterprise productivity.

CADE has not commented on the latest deal. The agency cleared the original Amazon-Anthropic partnership in 2024 alongside the UK Competition and Markets Authority, but the scale of the new commitment and the integration depth may trigger fresh review under Brazilian merger-control thresholds applied retroactively.

The Amazon AI Doctrine Is Now Visible

As Rio Times reporting on Anthropic’s funding trajectory noted, the company moved from a US$40 billion valuation target in 2024 to US$380 billion now. The arithmetic of running both OpenAI and Anthropic as primary AWS workloads has clarified Amazon’s strategic posture: rather than picking a single AI lab, Jassy is funding both leading frontier developers and locking in a decade of cloud revenue from each.

The two-lab strategy is expensive. Amazon has now committed up to US$75 billion in equity (US$50 billion to OpenAI, US$25 billion to Anthropic) and US$200 billion in cloud-revenue commitments — numbers that approximate the entire AWS annual revenue base. The bet is that the AI compute spend that flows back through the Trainium pipeline produces durable infrastructure economics that justify the equity exposure.

As Rio Times analysis of the original UK and EU regulatory review documented, the antitrust case against treating these arrangements as full mergers turns on whether the AI labs retain operational independence. The new console-integration provision — Claude appearing natively inside AWS billing — tightens that question considerably.

What to Watch Next

Three near-term variables matter for Latin American markets. First is the Anthropic IPO timing — both Anthropic and OpenAI are now structuring long-term commitments precisely to satisfy the public-market demands an IPO will impose, and a 2026 or 2027 listing would be one of the largest tech debuts in history.

Second is the Mexico nearshoring read. Trainium chip volume scaling means Foxconn and the other Taiwanese ODMs operating in Chihuahua have a multi-year demand floor independent of consumer-electronics cycles. That helps lock in the export surge already documented in 2025 trade data.

Third is the regional AI-policy consequence. With Microsoft, Amazon and Google all now running deep equity-plus-cloud structures with Anthropic and OpenAI, Latin American governments seeking to build sovereign or independent AI capacity face escalating costs every month. As detailed in Rio Times reporting on the original 2023 Amazon-Anthropic partnership, the structural concentration was visible from the start — the new deal makes it permanent.