Anthropic's Bold AI Infrastructure Push: Custom Chips in the Works While CoreWeave Deal Powers Claude Today

In the span of 24 hours this week, two major stories broke that together reveal how serious Anthropic's Anthropic AI infrastructure ambitions have become. On Thursday April 9, Reuters reported that the company is exploring the design of its own custom AI chips. On Friday April 10, Bloomberg reported that Anthropic has signed a multi-year deal with GPU cloud provider CoreWeave to supply the Nvidia-based compute needed to build and run its Claude models right now. Taken separately, each story is significant. Taken together, they paint a clear picture of a company that is simultaneously shoring up its short-term compute supply while beginning the long, expensive process of reducing its dependence on third-party silicon over the long term.

Anthropic's run-rate revenue has surged from roughly $9 billion at the end of 2025 to over $30 billion by early April 2026 — a more than threefold increase in a matter of months, driven by enterprise adoption of Claude and the breakout growth of Claude Code. That kind of acceleration puts enormous pressure on compute infrastructure. The decisions being made now about chips, data centers, and cloud partnerships will determine whether Anthropic can keep up with the demand it is generating.

The Reuters Report: Anthropic Is Exploring Custom AI Chip Design

According to a Reuters exclusive by Max A. Cherney and Deepa Seetharaman, Anthropic is in the early stages of exploring whether to design its own artificial intelligence chips. The report is based on three sources — two with direct knowledge of the matter and one person briefed on the plans. An Anthropic spokesperson declined to comment.

The key qualifier is that these plans remain genuinely preliminary. The company has not committed to a specific chip design. It has not yet assembled a dedicated team to work on the project. It may ultimately decide to continue purchasing chips from external vendors rather than designing its own. This is exploration, not execution — but the fact that it is happening at all is meaningful context for understanding where Anthropic sees its compute strategy heading.

Why the Cost of Custom AI Chip Design Is So High

Designing an advanced AI chip is not a project any company undertakes lightly. Industry sources cited in the Reuters report put the cost at roughly half a billion dollars to design a competitive chip — covering the engineering talent required, the design and verification process, and the cost of ensuring the manufacturing process produces chips without defects. That is before a single chip rolls off a fabrication line.

The investment requires a long-term commitment, because custom chip development timelines typically run several years from initial design to production silicon. A company that begins exploring chip design today would not realistically see working chips for two to three years at the earliest, and longer if the design goes through multiple revisions. This is why the decision to even begin exploring the option is treated as major news — it signals a conviction that the compute shortage is not a temporary problem that will resolve on its own.

Anthropic's Current Chip Mix

Today, Anthropic relies on a combination of chips from multiple suppliers to develop and run Claude. These include Tensor Processing Units (TPUs) designed by Alphabet's Google and chips from Amazon, reflecting the company's deep relationships with both of its major cloud investors. Earlier this week — just days before the Reuters chip story broke — Anthropic signed a long-term deal with Google and Broadcom (which helps design TPUs) for AI chip supply. That agreement builds on Anthropic's previously announced commitment to invest $50 billion in strengthening US computing infrastructure.

The Google-Broadcom TPU deal and the potential custom chip exploration are not contradictory moves. They represent different time horizons: securing supply from established partners for the near term while investigating whether a proprietary path makes sense for the long term.

Who Else Is Going Down This Road

Anthropic is not moving in isolation. The Reuters report explicitly notes that Anthropic's discussions mirror similar efforts at other major technology companies. Meta has been developing its own AI accelerator chips for years. OpenAI has been reported to be exploring custom silicon as well. Google has been designing its own TPUs for over a decade. The pattern is clear: as AI workloads scale to enormous sizes, companies that run them at the frontier start to find that general-purpose chips — even excellent ones from Nvidia — are not optimally matched to their specific model architectures and training approaches. Custom silicon tailored to a company's exact workloads can deliver efficiency gains that third-party chips cannot.

For Anthropic specifically, the incentive is also strategic. Depending on Nvidia's GPU supply — which remains constrained as demand continues to outpace production — creates vulnerability. A company that can supplement externally sourced chips with its own designed silicon gains negotiating leverage and supply resilience that it currently does not have.

The Bloomberg Report: Anthropic Signs Multi-Year CoreWeave Compute Deal

While the chip story is about the future, the CoreWeave deal is about right now. Announced on Friday April 10 via a CoreWeave press release and first reported by Bloomberg, the agreement sees Anthropic tap CoreWeave's GPU cloud infrastructure under a multi-year contract to support both the development and deployment of its Claude AI models.

CoreWeave will provide capacity across multiple Nvidia chip architectures at data centers located in the United States. The rollout is described as a phased infrastructure deployment beginning later in 2026, with the potential to expand over time. Financial terms were not disclosed.

What CoreWeave Brings to the Table

CoreWeave has a distinctive position in the AI infrastructure market. Founded in 2017 as an Ethereum mining operation that bought Nvidia GPUs in bulk, the company pivoted to GPU cloud services in 2019 as crypto margins compressed. That pivot turned out to be extraordinarily well-timed: as AI training and inference workloads exploded, CoreWeave had exactly the hardware and operational expertise the market needed. The company went public on Nasdaq in early 2026 under the ticker CRWV.

CoreWeave's infrastructure is purpose-built for AI workloads in a way that distinguishes it from general-purpose hyperscale cloud providers. The company has earned the top Platinum ranking in both the SemiAnalysis ClusterMAX 1.0 and 2.0 evaluations, which independently measure the performance, efficiency, and reliability of AI cloud platforms. Its MLPerf benchmark results — the industry standard for measuring AI inference performance — have been among the strongest in the field.

With the addition of Anthropic to its customer roster, CoreWeave now counts nine of the world's ten leading AI model providers as platform users. That list alongside Anthropic includes Meta, OpenAI, Mistral, Cohere, IBM, and Nvidia itself. The breadth of that customer base is a strong signal of where serious AI compute demand is being channeled.

The Timing: CoreWeave's Biggest Week

The Anthropic deal landed just 24 hours after CoreWeave disclosed an expanded $21 billion agreement with Meta Platforms for dedicated AI cloud capacity running from 2027 through December 2032. That Meta deal brought the total value of the two companies' infrastructure relationship to approximately $35 billion. CoreWeave also expanded its agreement with OpenAI by up to $6.5 billion earlier in 2026. In under a week, CoreWeave announced partnerships or expansions with three of the four most prominent AI model developers in the world.

Financial markets responded accordingly. CoreWeave stock gained over 4% in premarket trading after the Anthropic announcement and climbed more than 13% by midday Friday. The company generated $5.13 billion in revenue in 2025 and is guiding for more than $12 billion in 2026, backed by a contracted backlog that exceeds $66 billion. Landing Anthropic alongside Meta and OpenAI in a single week is not just a revenue story — it is a validation story that significantly reduces the question of whether CoreWeave's growth trajectory is sustainable.

Why Anthropic Needs Both: The Short and Long Game of Compute Strategy

The juxtaposition of these two announcements in the same 48-hour window is not coincidental. It reflects the dual reality any frontier AI lab faces as it scales rapidly: you need to secure the compute you can get today, while investing in the supply infrastructure that will serve you at whatever scale you reach in three to five years.

Anthropic's revenue growth from $9 billion to $30 billion run-rate in a matter of months is extraordinary, but it also means the company's demand for compute is growing faster than any single supply arrangement can keep pace with. The Google-Broadcom TPU deal, the CoreWeave Nvidia-GPU deal, Amazon chip integrations, and now the potential custom chip exploration all represent pieces of a diversified infrastructure strategy designed to avoid the situation where any single supplier's constraints become Anthropic's constraints.

CoreWeave CEO Michael Intrator captured the current moment in AI infrastructure clearly: "AI is no longer just about infrastructure, it's about the platforms that turn models into real-world impact. We're excited to work with Anthropic at the center of where models are put to work and performance in production shows up." That framing — infrastructure in service of real-world model deployment — is precisely what the CoreWeave deal provides. Anthropic gains flexible, production-scale Nvidia GPU capacity it can ramp up as Claude demand grows, without having to own or operate the underlying data centers.

The custom chip exploration, if it advances, serves a different purpose. Proprietary silicon would give Anthropic chips specifically optimized for Claude's architecture — potentially more efficient and more capable than general-purpose accelerators for Anthropic's specific training and inference patterns. It would also give the company supply independence that no amount of cloud contracts can fully provide.

The Chip Shortage Context: Why Everyone Is Looking at Custom Silicon

It is impossible to understand Anthropic's chip exploration without understanding the broader supply environment. The AI chip market in 2026 is characterized by demand that consistently outpaces Nvidia's ability to manufacture and deliver GPUs. Nvidia's Blackwell architecture and the next-generation Vera Rubin GPUs (unveiled at GTC 2026, with volume shipments expected in the second half of 2026) are allocated far in advance, with hyperscalers and AI labs competing for supply. Meanwhile, a broader memory crisis has pushed GPU prices higher, with GDDR and HBM memory shortages flowing through to finished GPU costs.

In this environment, any company that develops alternative sources of high-quality compute — whether through proprietary chip design, long-term supply agreements with specialists like CoreWeave, or diversified multi-vendor strategies — gains a meaningful competitive advantage. The Reuters report notes that designing a custom chip costs around half a billion dollars. For a company generating $30 billion in annualized revenue and growing at Anthropic's current pace, that is a significant but not unreasonable investment if the economics of reduced per-compute-unit cost pencil out at scale.

What This Means for Claude Users and Enterprises

For individual Claude users and enterprise customers, the practical implication of all this infrastructure activity is straightforward: it is Anthropic investing in the capacity to meet demand that is growing faster than anyone predicted. The CoreWeave deal specifically focuses on deploying compute for production Claude workloads — meaning real users running real queries — rather than purely for model training. As the phased rollout expands through 2026, it should translate into sustained availability and performance for Claude across the developer, startup, and enterprise customers that have driven the revenue growth.

The custom chip exploration, if it progresses, would not affect current Claude users in the near term. But over a multi-year horizon, purpose-built Anthropic silicon could enable more efficient inference — potentially allowing Claude to respond faster or at lower cost per query, which matters enormously at the scale Anthropic is now operating.

The Bigger Picture: Anthropic's Infrastructure Bet

Both stories this week point to the same underlying reality. Anthropic is no longer a research lab that happens to have a product. It is a company generating tens of billions of dollars in annualized revenue, growing at a pace that puts it among the fastest-scaling technology businesses in history, and making the kind of capital-intensive infrastructure bets that only companies with serious long-term conviction make.

The CoreWeave deal is the near-term move: lock in Nvidia GPU capacity at production scale with a proven AI cloud provider that already serves nine of the ten top AI model developers. The custom chip exploration is the long-term hedge: begin the process of understanding whether Anthropic can build silicon as well as software, and whether the economics justify the enormous up-front investment. Neither is a sign of weakness — both are signs of a company that takes its infrastructure needs seriously at a moment when infrastructure is the limiting factor for the entire AI industry.

The question of whether Anthropic ultimately proceeds with custom chip development will likely become clearer over the next twelve to eighteen months. In the meantime, the CoreWeave deal ensures that Claude has the compute it needs to keep growing. That is the infrastructure story of this week — and probably of the year.

Want more coverage of AI industry news, infrastructure developments, and technology strategy? Browse our other posts for the latest.