The agreement, which expands a partnership the two companies began in 2023, strengthens Amazon’s place in the fast-growing market for advanced computing services.

Amazon said it will invest $5 billion in Anthropic immediately, with a further $20 billion available later if agreed commercial targets are met. That comes on top of the $8 billion Amazon has already invested in the company.

Anthropic, the maker of Claude, said it will use current and future generations of Amazon’s Trainium chips to train and run its models, expecting to secure up to five gigawatts of computing capacity over time.

The company also confirmed that one gigawatt of capacity using Trainium2 and Trainium3 chips should be available by the end of this year.

Giving Anthropic more access to the large-scale infrastructure needed to support high demand, Amazon gets a major long-term customer for its custom-built chips and cloud services.

More than 100,000 customers already use Anthropic’s Claude models on AWS, according to the companies.

Customers will also be able to access Anthropic’s Claude Platform directly through AWS accounts, allowing them to use existing billing, security controls and monitoring tools without separate contracts.

Amazon has invested heavily in expanding data centres and computing power as demand for advanced software tools rises. The company recently said it expects around $200 billion in capital spending this year, with much of that linked to technology infrastructure.

Chief Executive Andy Jassy said Anthropic’s long-term use of Trainium chips showed the progress both companies had made together.

“Our custom AI silicon offers high performance at significantly lower cost for customers, which is why it’s in such hot demand,” he said.

“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, as we continue delivering the technology and infrastructure our customers need to build with generative AI.”

Anthropic Chief Executive and co-founder Dario Amodei said demand for Claude continued to grow quickly.

“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,” he said.

“Our collaboration with Amazon will allow us to continue advancing AI research while delivering Claude to our customers, including the more than 100,000 building on AWS.”

Amazon shares rose about 2.7% in extended trading after the announcement.

The agreement is similar to OpenAI’s move, which recently made its ChatGPT Enterprise product available to participating agencies under the same pricing model.

The offer extends to the executive, legislative and judicial arms of government. While there is no requirement for agencies to adopt Claude, the deal opens the door for lawmakers, judges and federal workers to integrate the chatbot into day-to-day operations for sensitive but unclassified tasks.

“America’s AI leadership requires that our government institutions have access to the most capable, secure AI tools available,” said Anthropic chief executive Dario Amodei. “By offering expanded Claude access across all three branches of government, we’re helping the federal workforce leverage frontier AI capabilities.”

The GSA’s Federal Acquisition Service commissioner, Josh Gruenbaum, said affordability would speed adoption. “The price is going to help uptake from agencies happen that much quicker,” he noted.

Claude has already been deployed in select agencies, including the Department of Defense, and is approved under the FedRAMP High security standard. The government recently added Anthropic’s Claude, OpenAI’s ChatGPT, and Google’s Gemini to its Multiple Award Schedule, making them official suppliers for AI procurement.

Tech giants see little direct profit from such low-cost agreements. However, they gain valuable insight into how AI is applied in high-value government use cases, intelligence that could change future enterprise products or renewals once the current deals expire.

Discussions are ongoing with other providers, including Meta’s Llama, Elon Musk’s xAI’s Grok, and smaller niche AI platforms. Google is also reportedly negotiating a similar offer for its Gemini chatbot.

The approval process comes against a political backdrop. The White House has pledged to block AI systems found to have “partisan bias or ideological agendas” from doing business with the federal government. 

Critics of existing models, including some of former President Donald Trump’s allies, have claimed that tools such as ChatGPT and Gemini lean towards liberal viewpoints.

Government officials have stressed that approval for procurement does not indicate preference for any particular platform and that all approved models remain subject to ongoing bias reviews.

For years, product-market fit has been treated as a reliable signal that a product has found a stable position within its market, often validated through growth, retention, and engagement, yet these indicators were shaped in an environment where building required coordination, access to advanced capability was limited, and replication carried real cost, conditions that have shifted materially through 2024.

Artificial intelligence is no longer an experimental layer applied at the margins of products, but an embedded capability shaping workflows and expectations, while advances in developer tooling continue to compress the distance between idea and execution, reducing what previously required months of effort into cycles that can now be completed in days and, in some cases, hours.

Even the underlying capabilities are becoming interchangeable, with the CEO of Anthropic, Dario Amodei, noting in 2024 that “models are a bit more fungible than cars,” a shift that lowers the barriers that once separated established products from emerging alternatives.

This is reflected in how generative artificial intelligence is embedded into workflows, with developers relying on AI-assisted tools as part of their standard process, accelerating both the creation of new products and the replication of existing ones.

At the same time, capabilities that were once confined to specialised tools are now accessible through general-purpose platforms, enabling individuals and small teams to reproduce outcomes without recreating entire systems, which means that value is no longer tied exclusively to the product itself, but can be assembled through simpler paths.

The result is not just more competition, but a different kind of competition, because it no longer arrives only as a direct alternative offering a better version of the same product, but increasingly emerges as a more efficient way of achieving the same outcome.

This shift is reinforced by a growing preference for flexibility over lock-in, with Mark Zuckerberg arguing in 2024 that organisations “need to control their own destiny and not get locked into a closed vendor,” reflecting a broader move toward open models and architectures that reduce dependency on any single product.

As a result, what appears to be product-market fit is often less about deep alignment with enduring demand and more about timing, because a product may look successful not because it is essential, but because a simpler alternative has not yet fully emerged.

This becomes particularly evident in environments where products are built around organising access to capabilities, because as those capabilities become more accessible through artificial intelligence, the need for the product itself begins to weaken through gradual substitution.

Users do not necessarily abandon products outright, but instead adjust how they use them, relying less on certain features and reconstructing workflows in ways that reduce dependence over time, creating a situation in which usage metrics may remain stable even as underlying relevance declines.

This makes the transition difficult to detect, because traditional indicators of success continue to show positive signals even as the foundation that sustains them begins to shift, creating an illusion of stability at the moment when resilience is quietly eroding.

This dynamic is reinforced by the way artificial intelligence is being embedded into existing platforms, where new capabilities are introduced directly into tools that users already depend on, reducing the need for standalone products and shifting value toward those who control distribution and workflow integration.

Entire categories of product value are already being compressed into internal capabilities, with the CMO of Klarna, David Sandström, observing in 2024 that “essentially, we have removed the need for stock imagery,” illustrating how quickly previously monetised workflows can disappear.

The implication is that software is no longer competing as complete products, but as pathways to outcomes, and in such an environment, the simplest path to achieving a result becomes the most competitive one.

For product builders, this introduces a more fundamental question than how to achieve product-market fit, which is whether the value being delivered is deeply embedded in a way that can withstand simplification, or whether it exists within a structure that can be bypassed as more efficient alternatives emerge.

The risk, as evidenced throughout 2024, is not that products fail because they are poorly designed or executed, but that they become less necessary as the market discovers better ways to achieve the same outcomes.

What many teams interpret as product-market fit may, in practice, be product-timing fit, a reflection of when a product entered the market rather than how strongly it is anchored within it.

This strategic partnership aims to further advance the development of safer generative AI technologies. Focused on expediting the development of Anthropic’s future foundation models and extending their accessibility to Amazon Web Services (AWS) customers, the agreement encompasses several key components.

Anthropic will harness AWS Trainium and Inferentia chips to construct, train, and deploy its upcoming foundation models. This move leverages the price, performance, scale, and security advantages of AWS. Furthermore, the two companies will collaborate on the evolution of Trainium and Inferentia technology.

AWS will become the primary cloud provider for Anthropic, particularly for mission-critical tasks such as safety research and foundation model development. Anthropic plans to migrate the majority of its workloads to AWS, thereby benefiting from the cutting-edge technology offered by the world’s leading cloud provider.

Anthropic commits to providing AWS customers worldwide with access to future iterations of its foundation models through Amazon Bedrock, AWS’s fully managed service that offers secure access to top-tier foundation models. Additionally, Anthropic will grant AWS customers early access to unique features for model customization and fine-tuning.

The partnership heralds exciting possibilities for both companies. Amazon developers and engineers will gain access to Anthropic models through Amazon Bedrock, enabling them to incorporate generative AI capabilities into their projects. This move is poised to enhance existing applications and cultivate entirely new customer experiences across Amazon’s diverse business ventures.

Amazon CEO Andy Jassy expressed his optimism about the collaboration, emphasizing the potential of Amazon Bedrock and AWS Trainium to deliver enhanced value to customers. He stated, “Customers are quite excited about Amazon Bedrock, AWS’s new managed service that enables companies to use various foundation models to build generative AI applications on top of, and our collaboration with Anthropic should help customers get even more value from these two capabilities.”

Dario Amodei, Co-founder and CEO of Anthropic, echoed this sentiment, highlighting the significance of using AWS’s Trainium chips to develop future foundation models. The partnership expansion builds upon Anthropic’s earlier support of Amazon Bedrock and aims to unlock new possibilities for organizations of all sizes.

Anthropic has rapidly evolved into a global leader in foundation model provision and a staunch advocate for the responsible deployment of generative AI. Their foundation model, Claude, excels in a wide array of tasks while maintaining reliability and predictability. Claude’s impressive capabilities span industries from manufacturing and aerospace to finance, legal, and healthcare. Customers have reported a higher degree of control and reliability with Claude compared to other foundation models.

The partnership’s resources will assist customers in harnessing the potential of Claude and Claude 2 on Amazon Bedrock. These AI models are already transforming industries such as travel, asset management, and legal services. With early access to model customization features and fine-tuning capabilities, customers can further tailor these models to their specific needs.

The collaboration between AWS and Anthropic marks another significant step in AWS’s generative AI offerings. Customers can access a wide range of foundation models, customize them to protect their data privacy, and seamlessly integrate them into their AWS workloads through Amazon Bedrock. The partnership strengthens AWS’s position across all layers of the generative AI stack, including hardware, foundation models, and applications and services.

As part of this collaboration, AWS and Anthropic are dedicating substantial resources to help customers leverage Claude and Claude 2 on Amazon Bedrock. The AWS Generative AI Innovation Center will provide support for organizations of all sizes, enabling them to develop innovative AI-powered applications and drive transformation within their businesses.

Both Amazon and Anthropic are active advocates for the responsible development and deployment of AI technologies. They participate in numerous organizations and initiatives, including the OECD AI working groups, GPAI, the Partnership on AI, ISO, NIST, and the Responsible AI Institute. Their commitment to fostering AI’s safe, secure, and responsible advancement is further exemplified by their voluntary commitment to support ethical AI development.

This partnership signifies Amazon’s dedication to enhancing its AI capabilities, making AI accessible and responsible. The future of generative AI holds exciting prospects for AWS customers and beyond.