From OpenAI to Anthropic: AI Giants’ Battle for Computing Autonomy Begins, Is In-House Chip Development a Must?
AI Podcast
Anthropic is exploring in-house AI chip development to secure computing power and reduce supplier reliance, driven by significant business growth and exceeding $30 billion in annualized revenue for its Claude model. While currently using third-party chips from Google and Amazon, this preliminary research into custom chips mirrors industry trends from OpenAI and Meta. Developing proprietary chips is seen as a strategic necessity for supply chain autonomy and cost optimization, though the high R&D barriers remain. This pursuit signifies a broader industry race for computational control and specialized hardware.
TradingKey - The current shortage of AI chips has become a common challenge for leading companies. As a representative firm in the generative AI field, Anthropic is also exploring ways to break through. Recent reports suggest the company is evaluating the feasibility of developing its own AI chips to strengthen its control over computing power and reduce reliance on external suppliers.
However, sources familiar with the matter emphasized that the plan is currently only in the preliminary research stage and has not yet moved into a substantive development phase. Anthropic still retains the possibility of abandoning internal development in favor of continuing to purchase third-party chips. One source further revealed that Anthropic has neither finalized a specific chip design direction nor formed a dedicated R&D team for this purpose.
The core driver behind Anthropic’s consideration of in-house chips is the explosive growth of its business. Earlier this week, Anthropic executives publicly disclosed that demand for its AI model, Claude, surged significantly in 2026, with the company's annualized revenue surpassing $30 billion, compared to approximately $9 billion at the end of 2025.
Despite its surging business, Anthropic remains heavily dependent on external vendors for computing support. The company currently utilizes a multi-chip solution, including TPUs designed by Alphabet's Google and Amazon's in-house chips, to support the development and operation of its AI software and the Claude chatbot.
Notably, earlier this week, Anthropic reached a long-term TPU supply agreement with Google and Broadcom. This agreement is based on Anthropic's previously committed $50 billion investment plan for U.S. computing infrastructure.
Anthropic is not the only tech company venturing into custom AI chips; other industry leaders such as OpenAI have also initiated similar explorations.
However, industry insiders pointed out that the barrier to developing advanced AI chips is extremely high. R&D investment alone could exceed $500 million, covering costs for hiring top-tier chip engineers at high salaries, as well as expenditures for chip tape-outs and mass production verification.
Global AI Giants Launch In-House Chip Race
Against the backdrop of a global AI chip supply crunch, Anthropic’s choice is by no means an isolated case. Recent industry reports indicate that Meta ( META) and other leading companies such as OpenAI have also begun exploring the development of their own AI chips.
Although this movement is still in its early stages, it clearly reflects a trend in the AI industry, spanning from Google ( GOOGL ), Amazon ( AMZN) and other tech giants, to Meta and Microsoft ( MSFT) and other all-around players, and down to AI-native companies like OpenAI and Anthropic—a "shadow war" for computational autonomy has fully commenced.
In-house chip development is no longer a risky pursuit of technical gimmicks; rather, it has become a strategic necessity for companies to reduce supply chain dependency, optimize long-term costs, and build core competitiveness.
Current technical roadmaps reveal a clear trend of differentiated competition. Google has cultivated its TPU series for over a decade; its latest TPU v7p is deeply optimized for the multimodal training requirements of the Gemini large model, demonstrating efficient computing support within its own AI ecosystem.
Amazon's Trainium series focuses on AI training scenarios. The upcoming mass-produced v3 version will integrate HBM memory, doubling the bandwidth, with the goal of reducing large model training costs through higher energy efficiency ratios.
Meta's MTIA series targets inference scenarios. The v2 version is already in mass production, and the v3 version is expected in 2026, which will further optimize the operational efficiency of AI features across its social media platforms.
While Microsoft's Maia series has faced delays, it continues to progress with the goal of creating specialized AI computing chips for the Azure cloud platform. OpenAI has chosen to partner with Broadcom and TSMC, planning to deploy its first inference chip using a 3nm process in the second half of 2026; a single chip can support 10 gigawatts of computing power, providing a robust engine for large model inference.
It is clear that both traditional cloud providers and AI-native companies are attempting to break their sole reliance on Nvidia's general-purpose GPUs. By developing customized chips, they aim to achieve energy efficiency ratios better suited to their specific businesses and gain a more controllable computing supply chain, thereby seizing the initiative in the long-term competition of the AI industry.
Anthropic's Compute Strategy
As a U.S. startup dedicated to artificial intelligence safety research, Anthropic was established in San Francisco in 2021. Its flagship products are the Claude series of large language models. Through partnerships with platforms like Amazon AWS and Google Cloud, its AI model integration services now span multiple sectors, including healthcare, finance, and technology.
In September 2025, Anthropic closed a $13 billion Series F funding round, reaching a valuation of $183 billion. Only five months later, a Series G round in February 2026 propelled the company's valuation to $380 billion, highlighting its powerful growth trajectory.
As the commercialization of Claude models accelerates, Anthropic’s demand for computing power is surging explosively. Despite recently securing a 3.5-gigawatt compute agreement with Google and Broadcom, chip supply stability remains a widespread industry challenge. Amid a complex and shifting international landscape, supply chain uncertainties are mounting; any disruption in the supply of critical chips would directly impact the company’s core business operations.
This supply chain risk is compelling tech giants to pursue self-sufficiency. For Anthropic, developing in-house chips does not imply full manufacturing autonomy; instead, by mastering core design capabilities, the company gains technical alternatives and supply chain bargaining power, effectively insulating its business from external supply volatility.
On a deeper level, chips have emerged as the strategic high ground in the AI arms race. Deep synergy between software and hardware is essential to unlocking the potential of algorithms. In-house chip development is more than just a hardware strategy; it is a fundamental pivot for building competitive moats, crafting unique product experiences, and anchoring the future AI ecosystem, ultimately determining long-term corporate competitiveness.
This content was translated using AI and reviewed for clarity. It is for informational purposes only.
Recommended Articles
