Is the AI “perpetual-motion machine” facing a challenge? Google partners with Anthropic, preventing NVIDIA and OpenAI from monopolizing the spotlight; AI development looks healthier

After OpenAI announced various forms of cooperation with NVIDIA（輝達）, AMD（超微）, and Broadcom（博通）, on October 23 (U.S. time) Anthropic, the developer of the Claude chatbot, and Google also announced a cloud-services partnership.

This transaction, valued in the tens to hundreds of billions of dollars, could develop into a force that challenges the dominance of NVIDIA and OpenAI, but for the broader rollout of the AI industry it can instead be seen as a very positive factor.

First, the substance of the Google–Anthropic partnership: Anthropic plans to adopt Google Cloud at scale, including procuring 1 million customized TPU (Tensor Processing Unit) chips supplied by Google, and expects to bring more than 1GW of AI compute online in 2026.

Anthropic’s approach highlights the intense demand for compute in large-scale AI model development. Its main competitor, OpenAI, has not only established chip-supply relationships with NVIDIA and AMD in recent months, but has also begun co-developing in-house AI chips with Broadcom, underscoring that securing sufficient compute has become a key element for AI companies to maintain competitiveness.

A look at Anthropic’s basics: founded by former OpenAI staff, its major shareholders include ICONIQ Capital, Fidelity, and Lightspeed. Its Claude models are now primary competitors to the GPT series. After completing a $13 billion Series F round in September this year, the company’s valuation reached $183 billion—triple the $61.5 billion from the 2023 Series E.

On performance, Anthropic’s revenue through October this year has approached $7 billion, with an expected $9 billion by the end of 2025; projected 2026 revenue is $20–26 billion.

Although Anthropic has not disclosed net income figures, revenue growth is evident, and enterprise products such as Claude Code have already contributed nearly $1 billion in annualized revenue.

Anthropic’s revenue figures pose a notable threat to market leader OpenAI. OpenAI’s revenue this year is expected to reach $12.7 billion—triple the $3.7 billion of 2024—yet Anthropic’s growth rate is no less rapid, making it a close and formidable rival.

As for Anthropic’s key investors, ICONIQ Capital was the lead investor in the Series F round; the well-known Silicon Valley firm has invested in technology companies such as Slack and GitLab. Fidelity is one of the world’s largest asset managers, and Lightspeed is an early investor focused on technology and AI.

At the same time, Anthropic has also secured strategic cooperation and financial support from Google Cloud, further strengthening its cloud-computing capacity and market expansion.

Therefore, looking further, what impact could the Anthropic–Google partnership have on OpenAI, the AI market leader?

The challenge to OpenAI—dual-track competition in cloud and technology

In terms of market share, OpenAI currently relies on Microsoft Azure (微軟 Azure). If Anthropic expands rapidly with support from Google Cloud, it could siphon off enterprise clients and developer demand.

On the technology front, Google itself has the Gemini model; with its support for Anthropic on top, OpenAI faces competition on two technical fronts.

Next, what impact might this have on NVIDIA, the AI-chip leader that draws the most attention?

Although NVIDIA remains the dominant player in AI chips, the Google–Anthropic partnership involves Google’s customized AI accelerator, the TPU (Tensor Processing Unit). If deployment succeeds, more AI companies may shift toward non-GPU architectures. The rise of TPU constitutes a modest threat to GPU dominance.

At present, Anthropic uses Amazon/AWS’s Trainium, Google’s TPU, and NVIDIA’s GPU in parallel, indicating that the AI-chip market is moving toward diversification; future chip demand will become more distributed and less concentrated on NVIDIA.

Comparing the progress of in-house chips among major CSPs (cloud service providers), Amazon/AWS and Google are currently the more successful. Moreover, the accelerated progress of Google’s Gemini model benefits from its self-developed AI chips and relatively lower hardware costs—below systems based on NVIDIA and below OpenAI’s costs—thus attracting Anthropic’s orders.

For Google, already a strategic supporter and investor in Anthropic, further drawing Anthropic in as a major customer clearly secures another marquee TPU-compute client. If Gemini and Anthropic form a cooperative posture, it will certainly pose some threat to OpenAI.

However, Anthropic’s challenge to OpenAI’s leadership position and the emergence of Google’s TPU to carve up the GPU market with NVIDIA are overall positive for the development of the AI industry.

Viewed over a longer time horizon, an industry that is dynamic and competitive—neither monopolized by a few giants nor closed to new challengers—moves forward faster through intense competition, a decidedly positive and healthy evolution. The AI industry today is precisely in such a virtuous cycle.

Therefore, from the broader perspective of global AI development, the impact and implications of the Google–Anthropic partnership should not be underestimated and can be analyzed from three angles.

Three angles—cloud competition, model safety, enterprise demand

First is the intensifying competition for cloud resources. Google, Amazon (AWS), and Microsoft (Microsoft Azure) are actively vying for partnerships with AI startups, accelerating the expansion of global AI infrastructure.

Second is model diversification and improved safety. Anthropic emphasizes “AI safety,” forming a contrast with OpenAI’s more open strategy and helping to advance AI ethics and safety technologies.

The final major issue is that while many watch for the emergence of a “killer app,” Anthropic projects $9 billion and $20 billion in revenue for 2025 and 2026, respectively, indicating rapidly growing demand for AI solutions among enterprises and the potential for accelerated enterprise-grade AI adoption.

The Anthropic–Google partnership has also become a weather vane for global AI development. In the contest among models from OpenAI, Anthropic, Google’s Gemini, and Meta’s Llama, OpenAI is accelerating, with Anthropic and Gemini close behind, while Meta’s recent large-scale layoffs suggest it is falling behind.

On October 23, Meta announced layoffs of about 600 employees in its AI division. The cuts affect the AI infrastructure division, Fundamental AI Research (FAIR), and other product-related roles, impacting many hired during Meta’s rapid AI expansion over the past three years. The stated purpose is to streamline layers and improve operational efficiency.

However, Chief Executive Mark Zuckerberg（祖克柏） is seeking to realign teams and recruit more specialized talent to fill key roles. Alexandr Wang, who became Meta’s Chief AI Officer in June, directly oversees the TBD Lab division, which recruited multiple top AI talents this summer and was not included in the layoffs—suggesting Zuckerberg is leaning more heavily on the newly hired, highly paid team to take over AI development.

Finally, fierce competition and surging additional investment in the U.S. AI industry are unequivocally positive signals for Taiwan’s chip and server sectors, and Taiwan’s role will become even more critical.

On chips, whether through OpenAI’s alliances with NVIDIA, AMD, and Broadcom to obtain GPUs or ASICs, or through Google’s provision of TPU chips to Anthropic, most of these chip orders will flow to TSMC（台積電）, the market-share leader. The biggest beneficiary is TSMC, along with related supply-chain companies and back-end packaging and testing firms.

Second, whether for GPU or ASIC servers, Taiwan’s server ODMs are also the biggest winners. After TPU chips are fabricated by TSMC, Foxconn（鴻海）, Quanta（廣達）, and Wistron（緯創） will handle server assembly and system integration, making Taiwan’s broader supply chain the greatest beneficiary.

Therefore, whether OpenAI has set the “perpetual-motion machine” spinning or Anthropic and Google’s cooperative mechanism is now turning as well, these moves are propelling the AI industry forward—and the ultimate biggest beneficiaries are Taiwan’s semiconductor and ICT industries.

That said, within the entire AI ecosystem, the question remains: who is the most vulnerable group? Cheng Shijia（程世嘉）, CEO of Taiwan software company iKala, has raised the notion of a “compute-reseller trap,” arguing that current AI companies are essentially NVIDIA’s compute resellers, compelled to sacrifice profits—or even subsidize purchases—to buy chips.

For example, OpenAI must procure compute from multiple sources—NVIDIA, Microsoft Azure, Oracle, CoreWeave, and Google Cloud—falling into a complex and costly web of dependencies, with spending projected to reach $115 billion by 2029 yet still failing to achieve profitability.

Cheng also contends that the only exception is Google, which, by virtue of its self-developed TPU chips and vertically integrated ecosystem, has escaped this role and become a “compute original manufacturer.” Google’s supply chain designs its own chips and works with Broadcom and TSMC.

When the AI bubble will burst remains uncertain. But the more heated the debate, the higher the vigilance—and paradoxically, the lower the chance of an imminent burst. More important now is to identify who would be most harmed if the bubble bursts; clarifying that point can help investors avoid sharing the same fate.