Anthropic has hired Google chip veteran Amir Salek as the artificial intelligence company steps up efforts to build its own custom silicon capabilities, according to a Bloomberg report. Salek is joining Anthropic after spending years at Google, where he worked on the company’s custom AI chips, marking a significant addition to Anthropic’s hardware push as demand for computing power continues to surge.
The move comes as Anthropic seeks greater control over the infrastructure required to train and operate its Claude AI models. The company currently relies on chips from several suppliers, including Nvidia, Google and Amazon, but has recently started hiring for an in-house silicon business. Custom chips could potentially help Anthropic manage supply constraints, reduce dependence on external chipmakers and design processors specifically around the computing requirements of its AI models.
Anthropic Builds An In-House Silicon Team
Anthropic’s decision to recruit a senior chip engineer from Google represents a shift from simply purchasing computing capacity toward having greater influence over the hardware underneath its AI systems.
The company has begun hiring and listing positions related to an in-house silicon effort, according to the Bloomberg report. The initiative comes as AI companies increasingly recognize that access to advanced processors has become as strategically important as model development itself.
Anthropic currently purchases chips from Nvidia, Google and Amazon. Building its own silicon would not necessarily eliminate those relationships, but it could give the company another source of computing capacity and allow it to optimize hardware around Claude’s workloads.
Key Details Of Anthropic’s Hardware Push
| Particular | Details |
|---|---|
| Company | Anthropic |
| Hardware initiative | In-house silicon business |
| Key hire | Amir Salek |
| Previous employer | |
| Previous expertise | Custom AI chips |
| Major chip suppliers currently used | Nvidia, Google, Amazon |
| Main objective | Develop hardware tailored to Anthropic’s needs |
| Reported motivation | Supply constraints and AI infrastructure demand |
| Current stage | Hiring and team building |
Anthropic has not publicly disclosed a commercial chip launch date or detailed specifications for a future processor.
Who Is Amir Salek?
Amir Salek is a veteran of Google’s custom-chip efforts and brings experience from one of the world’s largest AI infrastructure operations.
Google has spent years developing its Tensor Processing Units, or TPUs, which are specialized processors designed to accelerate machine-learning workloads. That experience is particularly relevant to Anthropic as the company explores whether it can build or customize silicon for its own AI systems.
Hiring talent with direct experience in custom AI-chip development can shorten the learning curve for an AI company entering semiconductor design.
Why Salek’s Experience Matters
Google Custom Chip Experience
│
▼
AI Accelerator Design
│
▼
Hardware Optimization
│
▼
Anthropic's Claude Workloads
│
▼
Potential Custom Silicon
The appointment does not by itself mean Anthropic will manufacture chips independently. Semiconductor development typically involves multiple partners for design, fabrication, packaging and memory.
Why AI Companies Want Custom Chips
The explosive growth of generative AI has created an enormous demand for specialized computing hardware.
AI models require large numbers of accelerators for both training and inference. Companies that operate these models at scale therefore face substantial costs and potential supply constraints when relying entirely on third-party processors.
Custom silicon can provide a way to optimize hardware for specific workloads.
Potential Benefits Of Custom AI Chips
| Benefit | Potential Impact |
|---|---|
| Greater supply control | Reduces dependence on external availability |
| Workload optimization | Hardware can be designed for specific AI operations |
| Energy efficiency | Specialized processors may reduce power consumption |
| Lower long-term costs | Potentially improves economics at large scale |
| Greater hardware flexibility | Allows tighter integration with AI software |
| Supply-chain diversification | Adds another source of computing capacity |
The economics become particularly important when an AI company operates models at massive scale, because even small efficiency improvements can translate into significant savings across thousands of processors.
Anthropic Still Relies On Multiple Chip Suppliers
Anthropic’s hardware initiative comes despite its extensive relationships with existing chip providers.
The company has been expanding its use of Google Cloud TPUs and has previously announced plans involving as many as one million Google TPUs. Anthropic said that expansion would be worth tens of billions of dollars and was expected to bring more than one gigawatt of computing capacity online in 2026.
This means Anthropic’s custom-silicon initiative should be viewed as part of a broader infrastructure strategy rather than an immediate replacement for Google, Nvidia or Amazon hardware.
Anthropic’s Computing Strategy
ANTHROPIC
│
┌─────────────┼─────────────┐
▼ ▼ ▼
Nvidia Google Amazon
GPUs TPUs Chips
│ │ │
└─────────────┼─────────────┘
▼
Current Compute
│
▼
Custom Silicon Effort
│
▼
Greater Hardware Control
Maintaining several suppliers can help Anthropic avoid becoming overly dependent on a single hardware platform while its own silicon effort develops.
Google Has Become An Important Anthropic Chip Partner
Google’s role is particularly interesting because Anthropic is now hiring from the same organization that developed Google’s TPU technology.
Anthropic announced in October 2025 that it planned to expand its use of Google Cloud technologies, including up to one million TPUs. The company said the expanded infrastructure would support AI research and product development.
Google’s TPUs provide Anthropic with an alternative to Nvidia’s dominant GPU ecosystem.
For Anthropic, access to Google’s custom accelerators can diversify its compute supply while giving the company experience with non-Nvidia hardware architectures.
Nvidia Remains A Major Part Of The AI Supply Chain
Anthropic’s move into silicon also reflects the strategic importance of Nvidia in the AI hardware market.
Nvidia’s GPUs remain widely used for training and inference because of their performance and software ecosystem. Anthropic continues to use Nvidia hardware alongside chips from Google and Amazon.
However, depending heavily on external accelerator suppliers can create challenges when demand rises faster than available capacity.
The AI Hardware Supply Chain
AI Model Demand
│
▼
More Compute Required
│
▼
AI Accelerators
│
┌────┼────┬────┐
▼ ▼ ▼ ▼
Nvidia Google Amazon Custom
GPU TPU Chips Silicon
│
▼
AI Data Centers
│
▼
Claude & Other AI Services
Anthropic’s hardware strategy is therefore part of a much broader industry movement toward greater control over the AI computing stack.
AI Chip Development Is Becoming A Competitive Advantage
The AI industry is increasingly moving toward vertically integrated infrastructure.
Google designs TPUs. Amazon develops its own Trainium and Inferentia chips. Microsoft has developed Maia AI accelerators. Meta has also been working on custom silicon, while OpenAI has reportedly explored its own hardware and chip initiatives.
The trend reflects a fundamental change in the AI business.
Companies are discovering that model performance is influenced not only by algorithms and training data but also by the underlying computing architecture.
Major AI Companies And Custom Hardware
| Company | Hardware Strategy |
|---|---|
| TPU development | |
| Amazon | Trainium and Inferentia |
| Microsoft | Maia AI accelerators |
| Meta | Custom AI accelerator development |
| Nvidia | Full-stack GPU and AI infrastructure |
| Anthropic | Building an in-house silicon effort |
Anthropic’s move therefore places the company within a broader industry race to develop specialized AI infrastructure.
Custom Silicon Could Improve Claude’s Economics
One potential reason for Anthropic’s hardware push is the economics of inference.
Training frontier AI models requires enormous computing resources, but inference — generating responses for users — can also become expensive as usage grows.
If Anthropic can design hardware optimized for Claude’s specific workloads, it could potentially improve the amount of AI computation delivered per dollar or per watt.
This could become increasingly important as AI moves from simple chatbot interactions toward coding, research, enterprise automation and other workloads involving long-running model usage.
Energy Efficiency Is Another Major Consideration
AI data centers consume significant amounts of electricity, making energy efficiency increasingly important.
A specialized accelerator can potentially eliminate unnecessary hardware functions and optimize the processor around particular AI workloads.
Even small improvements in performance per watt can become meaningful when thousands of accelerators operate continuously inside large data centers.
Why Efficiency Matters
| Factor | Impact On AI Infrastructure |
|---|---|
| Compute demand | More processors required |
| Electricity consumption | Higher operating costs |
| Cooling requirements | Greater data-center complexity |
| Chip efficiency | Can reduce power per workload |
| Hardware utilization | Determines infrastructure productivity |
| Customization | Can optimize specific AI workloads |
Anthropic’s hardware team could therefore be focused not only on raw performance but also on cost and energy efficiency.
Supply Constraints Are Driving The Strategy
The Bloomberg report says Anthropic is scrambling to secure enough data-center infrastructure to support its ambitions. Custom chips could potentially help the company cope with supply shortages.
The issue has become increasingly important as AI companies compete for GPUs, advanced memory and data-center capacity.
When several major companies simultaneously expand AI infrastructure, securing enough processors can become a strategic problem rather than simply a procurement exercise.
Building custom silicon does not solve the problem immediately because chip development can take years. But establishing an internal team gives Anthropic a longer-term option for controlling part of its hardware supply.
The Move Could Deepen Anthropic’s Vertical Integration
Anthropic began as an AI model company, but its infrastructure strategy is becoming increasingly complex.
The company now works with multiple cloud and chip providers, while also exploring its own silicon.
This creates the possibility of a more vertically integrated AI platform in which Anthropic controls more of the relationship between model software and computing hardware.
The company would still need outside partners for manufacturing and other parts of the semiconductor supply chain, but internal chip design could give its engineers greater control over the architecture.
Custom Chips Will Not Replace External Suppliers Overnight
Despite the strategic benefits, developing a competitive AI accelerator is extremely difficult.
Chip design requires specialized engineering talent, large amounts of capital and access to advanced semiconductor manufacturing. AI processors also need software ecosystems, compilers, networking and memory systems to function effectively.
Anthropic is therefore likely to continue relying heavily on Nvidia, Google and Amazon while its hardware effort develops.
The immediate significance of Salek’s hiring is less about an imminent Anthropic chip and more about the company’s long-term intention to build internal expertise.
The Bigger Picture
Anthropic’s hiring of Google chip veteran Amir Salek signals that the AI company increasingly sees hardware as a strategic part of its future. The company is already buying computing capacity from Nvidia, Google and Amazon, but its decision to build an in-house silicon team could eventually give it greater control over supply, performance, energy efficiency and the cost of running Claude.
The move also reflects a broader transformation across the AI industry. As model companies scale, access to computing is becoming too important to treat purely as an external procurement issue. Google, Amazon, Microsoft and Meta are all developing custom accelerators or related silicon capabilities, and Anthropic’s latest hiring suggests it wants to participate directly in that race.
Looking Ahead
Anthropic’s next challenge will be turning its newly assembled hardware expertise into a viable silicon program. The company has not publicly announced specifications or a launch schedule for a custom accelerator, so the immediate focus will likely remain on recruiting engineers and developing the underlying architecture. Existing relationships with Google, Nvidia and Amazon are also likely to remain important as Anthropic continues expanding its computing capacity.
Over the longer term, successful custom silicon could give Anthropic another lever for controlling the economics of Claude. If its processors can deliver better performance, lower energy consumption or improved cost efficiency for Anthropic’s workloads, the investment could become strategically valuable. The move also suggests that competition among frontier AI companies is increasingly expanding beyond models and applications into the chips and infrastructure that power them.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
