[Digital Today reporter Seung-ah Yoo] An analysis says Jensen Huang (젠슨 황), Nvidia’s former chief executive, stepping down could work in Anthropic’s favor as it heads toward an initial public offering. As momentum weakens for Nvidia’s strategy of building an ecosystem for low-cost, open artificial intelligence models, Anthropic may face less competitive pressure as it sells closed, high-priced models.
Business Insider reported on Nov. 17 that Huang, while leading Nvidia, pushed large investments to develop inexpensive, open AI models. He also moved to build an ecosystem by supporting related startups and broadly distributing models. That strategy directly conflicted with Anthropic’s business model of developing its own high-performance closed models and charging usage fees.
The CEO of a major AI startup assessed that a “cold war was unfolding ahead of a massive clash” between Nvidia and AI research companies. He explained that Nvidia sought to challenge not only Anthropic but also OpenAI through in-house model development, investments in rival startups and ecosystem building.
At the center of Nvidia’s strategy were “open-weight” models. Unlike closed models such as Anthropic’s Fable and Opus and OpenAI’s major GPT models, open-weight models can be downloaded, modified and run independently by users. Companies can build AI systems tailored to their environments without paying fees to AI research firms each time.
Nvidia developed its own set of open models called Nemotron and invested in Reflection AI, Thinking Machines Lab and Poolside. It also pursued an acquisition of Hugging Face, a platform for sharing open-weight models. Amjad Masad (암자드 마사드), CEO of Replit, described Huang’s resignation at a Silicon Valley event as an “attack on open models” and said Huang played a key role in helping U.S. open-model companies raise funds.
Huang also stressed the importance of open models at a Fireworks AI event in San Francisco on Nov. 3. He argued that companies could build more powerful AI systems by combining their own domain knowledge and specialized tools with open models instead of relying on closed models.
The strategy also had business advantages for Nvidia. The more open models that run on its chips, the more it can expand chip demand and potential customers. Anthropic, meanwhile, has developed technology to reduce its dependence on Nvidia. According to its IPO filing, Anthropic spent years developing software that can move AI workloads across different chips from Nvidia, Google, Amazon and AMD. The aim is to cut reliance on any single supplier by choosing the chip that offers better cost and performance for each workload.
Michael Cembalest (마이클 셈벌레스트), an analyst at JPMorgan Asset Management, estimated on Nov. 14 that Nvidia’s share of future computing commitments by major AI research firms has fallen to 39 percent. That contrasts with Nvidia’s share of more than 80 percent in AI chip sales in recent years.
The two sides also clashed over U.S. restrictions on semiconductor exports to China. Dario Amodei (다리오 아모데이), CEO of Anthropic, has argued the United States should strongly limit exports of advanced chips to China. Huang opposed the regulations, saying excluding China could spur local companies to develop substitute technologies. The U.S. government recently announced new rules to block a workaround in which Chinese firms lease Nvidia graphics processing units at overseas data centers. That is interpreted as a step that lends weight to Amodei’s calls to tighten semiconductor restrictions on China.
Even so, the possibility remains that the management team serving as interim CEO after Huang will continue the existing strategy. Still, the analysis says the competitive landscape could change because a key figure who led the expansion of the open AI ecosystem has stepped down as Anthropic heads toward an IPO that could value it at about $1.4 trillion, or about 1,860 trillion won.