Why the Anthropic MatX Acquisition Collapsed: Partnership Over Ownership
The abandoned Anthropic MatX acquisition, a proposed $7 billion deal, says more about the economics of vertical integration than about the deal itself. The two companies discussed a full merger this week before pivoting to a design partnership, leaving the chip startup independent and Anthropic to spread its custom-silicon bets across multiple vendors.
MatX's founders came from Google's TPU engineering group, and its chips target large language model training specifically, a focus that separates the startup from general-purpose accelerator designs. At $7 billion, the offer would have been a large premium over the roughly $4 billion valuation at which MatX is now raising fresh capital, following a $500 million Series B in February 2026. MatX does not expect to ship silicon in volume until 2027.
What Happened
Anthropic entered the MatX talks as part of a broader push on in-house chips and a drive to cut its dependence on Nvidia, which supplies a large share of the compute behind its Claude models. The merger talks stalled, and the two companies have since turned to a partnership model instead. Anthropic has also held meetings with several other chip startups and has not yet committed to a single approach.
Neither company has disclosed why the acquisition collapsed. The most visible pressure point is price: at $7 billion, Anthropic would have paid roughly 75 percent above the $4 billion valuation MatX is seeking from new investors. Anthropic is readying a 2026 IPO that reportedly targets a $2 trillion valuation, a milestone that intensifies scrutiny of premium-priced acquisitions and puts the hardware strategy at the center of the story it tells investors. For an IPO, every line item of infrastructure spending will face analyst questions, and a $7 billion acquisition would have stood out on the balance sheet.
The IPO frame explains why the hardware push is accelerating now. Anthropic views custom silicon as a key step to consolidate competitive barriers ahead of a listing, and a 2026 IPO target leaves little time to build a silicon program from scratch. Partnerships and strategic investments are faster routes to that capability, and they avoid the multi-year cost of an internal program running to hundreds of millions of dollars per generation.
Custom chip development is a slow, expensive path even when it works. A new silicon generation costs hundreds of millions of dollars and takes years to reach production, and MatX's own volume shipments are not due until 2027. For a company spending heavily on compute today, owning a pre-shipment design house delivers little near-term relief, which is why the partnership route keeps the technology on the table without the ownership bill.
Why the Anthropic MatX Acquisition Fell Apart
The price gap explains why the Anthropic MatX acquisition stalled: a $7 billion offer for a company raising at roughly $4 billion, with volume production still two years away. A partnership gives Anthropic access to MatX's training-focused designs and the former Google engineers building them, while MatX keeps the freedom to sell to other customers. That structure preserves the startup's market incentives instead of locking its roadmap to a single buyer.
Control has a price, and this time it was a $3 billion premium plus integration risk. Folding a young engineering team into Anthropic's corporate structure would have pulled MatX's founders away from the independent roadmap that makes the company attractive to a broader market. Acquisitions of engineering-led startups often fail to keep the people they were bought for, and MatX's value sits with a small group of former Google TPU engineers. With an IPO in view, capital spent on compute capacity carries more near-term return than capital spent on an acquisition premium.
The choice between buying and partnering came down to what Anthropic needed most. An outright purchase would have secured exclusive access to MatX's architecture and folded the team in immediately, but at roughly 75 percent above the valuation where the startup is raising new money. The partnership keeps the technology within reach while MatX continues raising capital at its own pace and selling to a wider market. What Anthropic gives up is exclusivity: a partner's roadmap serves many customers, where an owned subsidiary would serve one.
| MatX deal metric | Figure |
|---|---|
| Discussed acquisition price | ~$7 billion |
| Current fundraising valuation | ~$4 billion |
| Premium implied by the $7B offer | ~75% |
| Series B raised (February 2026) | $500 million |
| Planned volume shipments | 2027 |
The Multi-Supplier Silicon Strategy
Walking away from MatX does not mean Anthropic is giving up on custom silicon. It means the bets are being spread. The company is in talks with Samsung about manufacturing custom parts, works with Broadcom as part of its hardware and financing strategy, and has taken a $5 billion investment from AMD. At the same time, Anthropic keeps running workloads on Nvidia, Google, and Amazon hardware as part of an explicit multi-chip approach.
The portfolio is the practical answer to a market where no single vendor offers both the supply and the negotiating leverage Anthropic wants. Nvidia remains the default for large-scale training, Google's TPUs handle specific workloads, and in-house designs are meant to cover future generations. Senior engineers brought in from Google and OpenAI suggest the internal effort continues regardless of the MatX outcome, and the company has stated it will keep Nvidia and Google in its hardware mix while internal designs mature. The approach also hedges against supply constraints: if any single vendor's allocation tightens, workloads can shift to another part of the stack.
The cost reality behind the strategy is steep. Anthropic pays about $1.25 billion per month to SpaceX for compute capacity, a commitment that runs to roughly $15 billion a year, and bills for Google and Amazon cloud infrastructure add billions more. The annual SpaceX outlay alone is more than double the price Anthropic was willing to pay for MatX. Against that backdrop, the $7 billion acquisition would have bought a design house whose chips do not ship in volume until 2027, a down payment on capacity that arrives years after the cash leaves the balance sheet.
The supplier map shows who captures that recurring spend: AMD put $5 billion into Anthropic as an investor and chip partner, Samsung would manufacture Anthropic-designed parts, Broadcom sits inside the hardware and financing strategy, and SpaceX collects $1.25 billion a month for compute. Several of these relationships are deliberately two-sided: AMD's investment aligns its incentives with Anthropic's growth, and Broadcom's financing role gives it a stake beyond chip supply. Each of those players earns margin from Anthropic's expansion, while the model lab keeps optionality rather than ownership.
The alternative of simply buying more Nvidia compute remains the default for much of the industry, but it carries allocation risk and leaves pricing power with the supplier, the two problems Anthropic's multi-chip strategy is designed to reduce. What the MatX partnership must deliver is real supply: the startup's chips target LLM training, the workload where Anthropic's demand is growing fastest, and its 2027 volume shipments would create a second training path alongside Nvidia. The value of the partnership now rests on that timeline and on MatX serving other customers while supplying Anthropic.
Why This Matters
The failed Anthropic MatX acquisition makes visible who captures the margin in the AI compute stack: the infrastructure providers, not the model labs. A $1.25 billion monthly SpaceX commitment and reliance on Nvidia, Google, and Amazon hardware are recurring costs that no partnership structure removes. For decision-makers, the lesson is that even a lab preparing a $2 trillion IPO treats silicon ownership as optionality rather than necessity. Diversification across suppliers is what scales.
Photo by Brecht Corbeel on Unsplash
Related Articles
- Google's $150B Anthropic Chip Financing Machine Beats Nvidia on Borrowing Costs
- Broadcom AI chip financing: $100B debt plan for Anthropic
- Anthropic TPU partnership: Scaling Compute or Just Costs?
✔Human Verified
Researched and cross-referenced against primary sources by the Bytevyte editorial team. This article was generated with the assistance of artificial intelligence and reviewed by the Bytevyte editorial team.