Anthropic reportedly explored a roughly $7 billion acquisition of AI chip startup MatX as the Claude maker accelerates efforts to develop its own custom processors and reduce dependence on Nvidia. According to people familiar with the discussions cited by Reuters, the acquisition talks are no longer active and have shifted toward a potential partnership between the two companies.
The potential deal highlights how Anthropic is moving beyond being primarily an AI-model company and increasingly investing in the hardware needed to train and operate Claude. MatX was founded by former Google Tensor Processing Unit engineers and is developing chips designed for large-scale AI model training. The startup is now reportedly seeking new funding at a valuation of about $4 billion, significantly below the roughly $7 billion acquisition price Anthropic had discussed.
Anthropic Considered $7 Billion MatX Acquisition
Anthropic discussed buying MatX for approximately $7 billion, according to two people briefed on the matter.
The proposed acquisition was intended to accelerate Anthropic’s development of custom AI hardware by bringing MatX’s chip-design expertise and engineering talent inside the company.
The acquisition talks have since evolved into discussions about a potential partnership, although Reuters could not determine why the proposed purchase was abandoned.
Neither Anthropic nor MatX publicly confirmed the reported acquisition discussions. Anthropic declined to comment, while MatX did not respond to Reuters’ request for comment.
Anthropic-MatX Deal At A Glance
| Particular | Reported Details |
|---|---|
| Potential buyer | Anthropic |
| Target | MatX |
| Reported acquisition value | ~$7 billion |
| Current status | Acquisition talks no longer active |
| Possible next step | Partnership |
| MatX current fundraising target | ~$4 billion valuation |
| MatX founders | Former Google TPU engineers |
| Main technology | AI chips |
| Primary focus | AI model training |
| Strategic objective | Accelerate custom-chip development |
The reported transaction would have been one of Anthropic’s largest strategic moves into semiconductor technology.
Why Anthropic Wants Its Own AI Chips
Anthropic’s Claude models require enormous amounts of computing power.
The company currently relies on several external hardware providers, including Nvidia, Google and Amazon.
Building custom chips could allow Anthropic to tailor processors specifically for its models and workloads.
That could potentially improve performance, energy efficiency and long-term computing economics.
Anthropic’s Custom Chip Strategy
Claude AI models
│
▼
Growing compute demand
│
▼
Reliance on external chips
│
├── Nvidia
├── Google
└── Amazon
│
▼
Anthropic builds silicon team
│
▼
Custom AI processors
│
▼
Greater hardware control
Anthropic has said its in-house silicon team will co-design hardware and models so Claude can operate faster and more efficiently at scale.
MatX Could Accelerate Anthropic’s Chip Program
The attraction of MatX appears to be its specialized semiconductor expertise.
MatX was founded by former Google engineers who worked on the company’s TPU program.
The startup has been developing a processor aimed at training large AI models, one of the most computationally demanding stages of AI development.
Why MatX Matters
| MatX Capability | Potential Anthropic Benefit |
|---|---|
| Former Google TPU engineers | Specialized chip expertise |
| AI processor development | Faster hardware program |
| Training-chip focus | Support for model development |
| Startup engineering team | Faster recruitment |
| Custom architecture | Potential workload optimization |
| Existing technology | Could reduce development time |
Designing a new AI processor internally can take years, making an acquisition of an experienced chip team potentially attractive.
MatX Is Seeking A $4 Billion Valuation
The reported $7 billion acquisition discussions are particularly notable because MatX is now seeking new capital at a valuation of approximately $4 billion.
That suggests the startup’s current fundraising ambitions are considerably below the price Anthropic had reportedly considered for an outright acquisition.
However, the two figures represent different circumstances and should not be treated as directly comparable valuations.
MatX Valuation Comparison
| Item | Reported Value |
|---|---|
| Anthropic’s discussed acquisition price | ~$7 billion |
| MatX reported fundraising valuation | ~$4 billion |
| Difference | ~$3 billion |
| Acquisition premium vs. fundraising valuation | ~75% |
The reported figures come from different stages of negotiations and financing discussions, so they do not establish a formal market valuation for MatX.
Anthropic Is Building An In-House Silicon Team
The MatX discussions come shortly after Anthropic publicly confirmed that it was creating an in-house silicon team.
The company said the group would design custom chips for Claude while continuing to use hardware from Nvidia, Google, Amazon and other providers.
This is important because Anthropic is not pursuing an immediate replacement for third-party chips.
Instead, the company appears to be building a multi-chip strategy.
Anthropic’s Hardware Model
| Hardware Source | Role |
|---|---|
| Nvidia | AI accelerators |
| TPU infrastructure | |
| Amazon | Trainium / Inferentia |
| Anthropic | Future custom silicon |
| MatX | Potential technology/partnership |
The strategy could allow Anthropic to diversify its hardware supply while developing processors optimized for Claude.
Anthropic Has Hired Chip Experts
The company has also been expanding its semiconductor talent base.
Reuters reported that Anthropic recently hired Amir Salek, a Google chip veteran, as part of its hardware push.
In June, Anthropic hired former OpenAI chip engineer Clive Chan, who worked on OpenAI’s custom processor effort.
Anthropic’s Chip Talent Expansion
| Executive / Engineer | Previous Connection | Role In Strategy |
|---|---|---|
| Amir Salek | Chip engineering | |
| Clive Chan | OpenAI | Custom AI chip expertise |
| MatX team | Former Google TPU engineers | Potential acquisition/partnership |
The recruitment strategy indicates that Anthropic is attempting to build semiconductor capabilities internally rather than relying exclusively on hardware suppliers.
Custom Chips Could Reduce Nvidia Dependence
Nvidia remains the dominant supplier of AI accelerators, but its processors have become increasingly difficult for AI companies to secure at the scale they need.
Reuters reported that Nvidia expects its processors to remain in short supply through 2027.
Anthropic’s custom-chip ambitions could therefore serve two purposes.
First, the company could tailor processors to Claude’s specific workloads.
Second, it could reduce exposure to shortages and pricing pressures affecting third-party AI accelerators.
Nvidia Dependence Vs Custom Silicon
| Factor | External GPUs | Custom Anthropic Chips |
|---|---|---|
| Availability | Dependent on suppliers | Greater internal control |
| Design | General-purpose AI workloads | Claude-specific |
| Development time | Immediate access | Years of development |
| Upfront investment | Lower | Very high |
| Optimization | Broad | Potentially specialized |
| Supply risk | Higher | Potentially lower |
| Flexibility | High | Depends on architecture |
Custom silicon does not eliminate the need for Nvidia or other providers.
Anthropic has explicitly said it intends to maintain a multi-chip approach.
Anthropic Is Spending Tens Of Billions On Compute
The MatX discussions come as Anthropic commits enormous amounts of capital to AI infrastructure.
The company recently signed a $45 billion agreement with Nscale to rent computing capacity from a planned data center in West Virginia. The arrangement is expected to provide access to 460 megawatts of Nvidia Vera Rubin processors when the facility comes online.
Anthropic has also committed to other major infrastructure arrangements.
Anthropic’s Reported Compute Commitments
| Partner / Source | Reported Commitment |
|---|---|
| Nscale | ~$45 billion over six years |
| Google AI chips | ~$36 billion purchase plan |
| SpaceX | $1.25 billion per month through May 2029 |
| Google broader infrastructure | >$150 billion chip lease commitment reported |
| Custom silicon | In development |
The figures cover different types of agreements and should not simply be added together as direct chip purchases.
SpaceX Deal Adds Massive Computing Capacity
Anthropic has agreed to pay SpaceX $1.25 billion per month through May 2029 for computing capacity across its data-center clusters, according to Reuters.
One facility involved in the arrangement, Colossus 1, houses more than 220,000 Nvidia chips.
This illustrates the scale of Anthropic’s current appetite for computing resources.
Anthropic’s Compute Expansion
Claude demand
│
▼
More AI inference + training
│
▼
Massive compute requirement
│
├── Nvidia
├── Google
├── Amazon
├── Nscale
└── SpaceX
│
▼
Custom silicon development
│
▼
Long-term hardware independence
The custom-chip strategy is therefore being developed alongside, rather than instead of, huge external infrastructure commitments.
Anthropic Could Focus On AI Training Chips
The MatX discussions also provide clues about what type of processor Anthropic might develop.
MatX has been working on chips useful for AI model training.
Training and inference have different requirements.
Training involves processing huge datasets and updating model parameters, while inference involves generating responses after a model has been trained.
AI Training Vs Inference
| Characteristic | Training | Inference |
|---|---|---|
| Purpose | Build model | Run model |
| Workload | Massive parallel computation | Repeated model execution |
| Data | Training datasets | User prompts / inputs |
| Optimization focus | Throughput | Latency + efficiency |
| MatX reported focus | Training | Not primary reported focus |
| Anthropic interest | High | Potential future option |
Anthropic could eventually develop processors for both workloads, but the current MatX connection appears particularly relevant to training.
OpenAI Is Taking A Similar Path
Anthropic’s hardware strategy mirrors a broader movement among AI labs.
OpenAI has also been developing custom chips.
Reuters reported that OpenAI recently highlighted its first custom processor, called Jalapeno, saying it outperformed a comparable Nvidia processor in an inference workload while consuming less energy.
This suggests the leading AI labs increasingly view hardware optimization as a competitive advantage.
AI Labs Building Custom Silicon
| AI Company | Custom Hardware Strategy |
|---|---|
| Anthropic | In-house silicon team |
| OpenAI | Jalapeno custom chip |
| TPU | |
| Amazon | Trainium / Inferentia |
| Meta | Custom AI accelerator efforts |
The shift could gradually reduce the AI industry’s dependence on a single accelerator architecture.
Google And Amazon Already Have Custom AI Chips
Anthropic’s multi-chip approach is partly built around hardware developed by cloud providers.
Google’s TPUs are designed specifically for machine-learning workloads.
Amazon has developed Trainium for training and Inferentia for inference.
Anthropic already uses hardware from these providers alongside Nvidia.
The development of its own silicon would give the company another option.
Custom Chips Can Improve Model-Hardware Co-Design
One major advantage of owning chip design capabilities is the ability to optimize hardware and software together.
A general-purpose accelerator must support a wide range of models and customers.
An Anthropic-designed processor could potentially be built around the specific computational patterns of Claude.
Claude model architecture
│
▼
Identify computational bottlenecks
│
▼
Custom chip architecture
│
▼
Optimize memory + compute
│
▼
Run Claude more efficiently
This type of model-hardware co-design is becoming increasingly important as AI models become more specialized.
Chip Development Is Expensive And Slow
The main disadvantage is time and cost.
Reuters reported that producing a viable new chip can take a year or more, while the design cost for a single generation can reach hundreds of millions of dollars.
That creates a strong incentive to acquire an established team if doing so can accelerate development.
AI Chip Development Economics
| Factor | Reported Reality |
|---|---|
| Development time | 1+ year for viable hardware |
| Single-generation design cost | Hundreds of millions of dollars |
| Engineering expertise | Highly specialized |
| Manufacturing | Requires external foundries |
| Software ecosystem | Critical |
| Scale | Essential for economic benefits |
A $7 billion acquisition would therefore represent a massive investment, but Anthropic could potentially save years of recruitment and development effort.
Anthropic Is Talking To Other Chip Startups
The MatX discussions do not appear to be an isolated event.
Reuters reported that Anthropic has recently met with several AI chip startups as its engineers and executives study different approaches to processor design.
The company has not decided to make an acquisition.
This suggests Anthropic is still evaluating the best way to build its hardware organization.
Anthropic’s Current Hardware Options
Anthropic
│
├── Build internally
│
├── Acquire a chip startup
│
├── Partner with a chip startup
│
├── Work with Broadcom / other designers
│
└── Continue buying external accelerators
The final strategy could combine several of these approaches.
Anthropic’s IPO Adds Strategic Pressure
Anthropic is expected to pursue an IPO in 2026, according to Reuters reporting.
The company is reportedly targeting a valuation of around $2 trillion, based partly on expectations of enormous future revenue.
That makes control over infrastructure costs increasingly important.
Why Custom Chips Matter Before An IPO
| IPO Concern | Hardware Relevance |
|---|---|
| Revenue growth | More compute supports more customers |
| Margins | Efficient chips can reduce costs |
| Supply | Less dependence on scarce accelerators |
| Product performance | Hardware optimization can improve Claude |
| Capital requirements | Compute is a major expense |
| Long-term competitiveness | Proprietary silicon can create differentiation |
Investors may increasingly evaluate AI companies not only on model quality but also on their ability to secure affordable computing capacity.
The AI Chip Market Is Becoming More Competitive
The reported MatX discussions arrive during a broader restructuring of the AI semiconductor market.
Nvidia continues to dominate AI accelerators, but major technology companies are developing alternatives.
Google has TPUs.
Amazon has Trainium and Inferentia.
OpenAI is developing Jalapeno.
Anthropic is building its own silicon team.
This could gradually create a more fragmented AI hardware ecosystem.
The Bigger Picture
Anthropic’s reported exploration of a roughly $7 billion MatX acquisition demonstrates how aggressively AI companies are moving into semiconductor design. The deal was intended to accelerate Anthropic’s custom-chip program by bringing in MatX’s team of former Google TPU engineers and its work on processors designed for large AI model training. The acquisition discussions are no longer active, but Reuters reports that the two sides have shifted toward a possible partnership.
The move comes as Anthropic commits tens of billions of dollars to external computing infrastructure while simultaneously building an internal silicon team. The company has signed a reported $45 billion Nscale computing deal, plans to buy billions of dollars of Google’s AI chips and has agreed to pay SpaceX $1.25 billion per month for computing capacity through May 2029. Custom silicon could eventually give Anthropic greater control over performance, supply and computing economics, although the company says Nvidia, Google and other external providers will remain part of its multi-chip strategy.
Looking Ahead
Anthropic’s next move with MatX will be closely watched because a partnership could allow the AI lab to access specialized chip expertise without committing to the full $7 billion acquisition reportedly considered earlier. MatX is currently seeking new capital at a valuation of about $4 billion, while Anthropic is also evaluating other AI chip startups. The company has not yet committed to another acquisition, suggesting its semiconductor strategy remains under development.
The broader trend is clear: leading AI companies increasingly want control over the hardware beneath their models. Anthropic’s custom silicon effort, OpenAI’s Jalapeno project and the continued expansion of Google and Amazon’s proprietary accelerators could gradually challenge Nvidia’s dominance. For Anthropic, the objective is not necessarily to eliminate external chips, but to build enough proprietary capability to make Claude faster, more efficient and less vulnerable to shortages and rising infrastructure costs.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
