17.09.2026, The USA.
Nvidia’s success is not guaranteed. In the world of artificial intelligence, a scheme has taken shape in which Nvidia sells high-performance chips to AI developers, earning the bulk of the industry’s profits. OpenAI, Google, Anthropic, and other IT giants have lined up for the ever-more-expensive, scarce “hardware” for data centers. But the situation began to change not in Nvidia’s favor, and so it decided to break the established rules. The race for leadership in AI is increasingly beginning to resemble a war.
Nvidia did not disclose its steps to change strategy, but information leaked to the media. The Wall Street Journal reported on August 22 that Nvidia had struck an agreement with the AI-developing startup Poolside. Together they intend to create one of the world’s most powerful open-source models. Such a model can be installed on one’s own equipment without using third-party servers. It can also be further trained for specific tasks. The chipmaker intends to invest $6 billion in the project, and more than 100 Poolside employees, including engineers, will move to Nvidia and work there on AI development.
Less than a week later, on August 26, the online publication The Information reported that Nvidia had agreed to acquire Hugging Face for $12.9 billion. This is the world’s largest repository of open AI models. The platform hosts around 250,000 digital models. The site’s annual revenue is $150 million, which is incomparably small compared to the purchase price. So this is not about money. Having gained control over the “showcase” of models, Nvidia gets the ability to regulate who is there and on what terms.
In both cases, the deals became known from unofficial sources. But the chipmaker did not deny the information. These two pieces of news, if viewed separately, raise questions. Together, they paint a picture that changes all the alignments in the global game called “artificial intelligence.”
Will the Open AI world become closed?
Hugging Face co-founder and CEO Clément Delangue has repeatedly emphasized its neutrality, calling it “Switzerland for AI.” The platform provides opportunities for searching, testing, refining, and deploying machine learning models on equal terms for both IT giants and small startups. Last autumn, Hugging Face refused $500 million in investment from the same Nvidia, explaining this by its unwillingness to have a shareholder capable of influencing decisions. Apparently, the size of the investment matters. It is as if a republic proud of its neutrality suddenly agreed to become a province of an empire.
Nvidia, in turn, is acquiring not a business, but control over an ecosystem. It is buying the ability to cut off oxygen at any moment for those who depend on this platform. If all this comes under Nvidia’s control, the company will be able to promote models that can work efficiently only on its chips, as well as “sink” anything that will work on competitors’ hardware. But for this, it needs to win on two fronts.
War on two fronts
In January 2025, the chipmaker lost about $500 billion in market capitalization overnight after the release of an efficient AI model by the Chinese startup DeepSeek. The costs of developing it were far lower than in the US; moreover, it consumes far less capacity, which means high-performance chips may not be so necessary. Chinese open models such as DeepSeek and Qwen have made a forced march in AI development over the past year.
Moreover, China does not want to depend on American equipment, given the sanctions obstacles Washington is building for them, and is developing its own chips. The Chinese company Dongfang Suanxin in July presented a chip for AI computing based on an alternative architecture that does not require expensive memory and imported components. During testing, the processor outperformed the popular H200 model in one of its indicators.
Competitors are also appearing in the US: major American developers of closed models, who are Nvidia’s main clients, have begun creating their own chips. On August 25, OpenAI published the first test results of its first chip; it is claimed that the processor outperformed the world leader’s chips in speed and energy efficiency. Google, Anthropic, and others are also developing their own solutions.
Nvidia leads the chip market, accounting for up to 80% of the market, but it is not the only manufacturer. Moreover, the company has competitors ready to fight for first place in the industry. If Nvidia succeeds in establishing its format of AI accelerators worldwide, it will have a chance to become a monopolist. Betting on the open sector could create an army of consumers of specific chips from startups, universities, and independent developers—all those who want AI models independent of third-party organizations.
However, the execution of these plans is not guaranteed. The deal may not pass through the antitrust filter of the US Federal Trade Commission (FTC). The American regulator has long been watching how tech giants expand. In January 2025, the commission launched a series of investigations and sent requests to Microsoft, Alphabet, Amazon, and OpenAI, demanding disclosure of details of their investments and partnerships in AI. And in February 2022, under pressure from antitrust authorities, Nvidia abandoned the purchase of Arm, one of the largest developers of mobile processors.
However, Nvidia has a trump card. The company can explain the deals by national interests, stating that this will help prevent China from achieving leadership in AI. Given how US President Donald Trump feels about this, such an argument could outweigh concerns about monopolization.
A repeat of history
As an example, one can recall the “format wars” for video cassettes at the end of the 20th century. Sony, with its Betamax, lost to VHS by betting on quality but closing the technology to other manufacturers. And JVC sold licenses for its format. In the 2000s, Sony, already taught by bitter experience, won the Blu-ray war against HD-DVD by creating a powerful ecosystem around the format. The company attracted content producers, game consoles, and consumers.
Nvidia’s actions show the same logic as Sony’s in the Blu-ray era. It is not just selling “hardware“—it is implementing a standard. And a standard wins not when it is technically better, but when a broad and loyal ecosystem is built around it. The acquisition of Hugging Face and the investment in Poolside are Nvidia’s attempt to secure a repository of models and a flagship open AI model, making it the “reference” that sets the standards for everyone else.
If Nvidia manages to combine these assets, it will create a situation in which any open model, to be competitive, must be optimized for Nvidia chips. And any developer wishing to have access to the best open models will be forced to buy equipment from the company. This is not a monopoly on production—it is a monopoly on compatibility, which provides more sustainable control than simply dominating the chip market.
What should Russia do?
Russia in this new reality is in the position of an observer who participates in the creation of neither chips nor advanced models, but acutely needs both. And Russia cannot tie itself to the Nvidia ecosystem, even if chips from this manufacturer dominate, as this is a sure path to trailing behind. The optimal path seems to be developing its own software that works equally efficiently on different standards. This will allow switching between suppliers depending on prices, sanctions, and the political situation.
Also, Russia cannot create advanced AI models of the GPT-5 level from scratch. That is too expensive and requires huge data centers. But it can become a center for adapting and fine-tuning open models for narrow tasks: industry, logistics, the public sector, defense, medicine. This does not require superchips, but it does require qualified personnel and quality equipment. Here one can occupy a niche that no one has occupied.
Instead of trying to catch up with Nvidia in chip production, Russia should focus on alternative architectures with an emphasis on a domestic component base. It is impossible to catch up in process technology, but one can win in computing efficiency through non-standard logic. While Nvidia tries to be “everything for everyone,” Russia can bet not on universality, but on energy efficiency.
It should also be understood that Nvidia’s success is not guaranteed. Chinese companies and American startups have already proven that it is possible to create efficient models with lower costs and on an alternative architecture. At the same time, American companies do not have successful open AI models; leadership in this area belongs to China. It is still unclear how, apart from gigantic investments, Nvidia intends to achieve a breakthrough. It may happen that Nvidia drags out the integration of assets, or the deals get bogged down due to the FTC, or the AI developer community “rebels” and creates a decentralized analog of Hugging Face. Then the plans to seize a monopoly will collapse.
Source: Rossa Primavera News Agency
