Nvidia | Caught in the tech cold war

By the time Nvidia disclosed in a regulatory filing that the U.S. government had imposed new licensing requirements on the sale of H20 chips to China, including Hong Kong and Macau, the chipmaker was already reeling. The decision, delivered by the Trump administration, would abruptly end shipments of its China-focused AI chips and saddle the company with a $5.5 billion hit to revenue due to unsellable inventory.

For the world’s most valuable semiconductor company and the undisputed leader in artificial intelligence chipmaking, this was not just a commercial disruption, it was a geopolitical collision.

This moment represents a dramatic escalation in the long-running tussle between the U.S. and China over control of critical technologies. And Nvidia, caught in the middle, finds itself facing the limits of doing business in both worlds.

The AI engine

Founded in 1993, Nvidia was initially known for its graphics processing units (GPUs), which powered video games with increasingly lifelike visuals. But by the 2010s, its chips found new utility far beyond gaming. Researchers discovered that Nvidia’s parallel processing architecture was ideal for training large language models (LLMs), a breakthrough that transformed Nvidia from a gaming company into the backbone of the AI revolution.

The company’s H100 and A100 chips have become industry standards for training LLMs and running generative AI tasks. With a market share estimated to be over 80% in AI data centre chips, Nvidia has fuelled the rise of companies such as OpenAI, Anthropic, and Google DeepMind.

Nvidia’s chips have also been used by Chinese firms such as Tencent, Alibaba, and ByteDance to build their own AI systems. For years, China accounted for a significant portion of Nvidia’s business. About a fifth of its revenue in 2023, or $17 billion, came from chip sales to companies in China. That business has became more fraught as tensions between Washington and Beijing grew.

In 2022, the Biden administration began tightening export controls on advanced semiconductors, citing national security concerns, with a singular goal of preventing American technology from enabling China’s military ambitions, including autonomous weapons systems, surveillance tools, and cyber warfare capabilities.

Nvidia responded by creating “downgraded” versions of its flagship chips by slowing down their performance with hardware and software tweaks, and also specifically building a new one, the H20 chip. These chips, known to be specifically designed for China, allowed Nvidia to keep a foothold in the market while technically adhering to U.S. restrictions.

That workaround now appears to have been closed. The Trump administration’s latest move requires Nvidia to obtain licences to sell even these downgraded AI chips. The new rules also affect Advanced Micro Devices (AMD) and come just weeks after U.S. lawmakers pressed the White House to halt Chinese tech firms’ access to Nvidia products, citing concerns that the chips were being stockpiled by companies with ties to China’s military.

Nearly $16 billion worth of H20 chips are currently pending export to Chinese tech giants, including Alibaba, ByteDance and Tencent. The White House’s decision has hit Nvidia’s stock hard, triggering a sharp selloff. But the longer-term impact may be more significant than a temporary dip in valuation.

The black swan moment

At the heart of the new urgency is a Chinese company that seemingly came out of nowhere: DeepSeek. In January, it released an AI reasoning model called DeepSeek R1 that it claimed was trained for only $6 million, which is a fraction of the cost associated with U.S. equivalents like OpenAI’s o1 reasoning model.

What truly alarmed U.S. officials was how DeepSeek did it. According to the U.S. congressional investigations, DeepSeek reportedly acquired tens of thousands of Nvidia GPUs through intermediaries in places such as Singapore and used techniques like distillation, allegedly including proprietary OpenAI data, to rapidly develop its model. Analysts estimate the company had access to as many as 60,000 chips, some of which should have been restricted under U.S. export laws.

Reports also revealed that several of DeepSeek’s researchers have connections to Chinese military institutions, including defence labs associated with nuclear weapons development and organisations already sanctioned by the U.S. government. These ties deepened the suspicion that DeepSeek’s rapid rise wasn’t just a tech story, but a strategic threat.

The Trump administration responded by not only tightening rules for Nvidia and AMD, but also considering broader penalties against DeepSeek, including blacklisting it from acquiring American technology and potentially restricting access to its services for U.S. citizens and entities.

For Nvidia, the developments are an existential test of its global business model. The company has worked for years to remain compliant with U.S. laws while still selling to China, the world’s second-largest AI market. But as restrictions multiply and political scrutiny intensifies, that balancing act is becoming impossible.

Nvidia now faces a future in which it must pivot aggressively back toward domestic and allied markets. It has pledged to invest $500 billion in U.S.-based AI infrastructure, including opening a server manufacturing facility in Houston and partnering with chip packaging firms in Arizona.

Still, the company’s China withdrawal could have lasting consequences. Analysts warn that if Nvidia pulls back, Chinese firms like Huawei could fill the gap. Already, Huawei is developing its own AI chips and could soon offer a homegrown alternative to Nvidia’s dominance. Losing market share in China not only negatively impacts Nvidia’s financial performance but also poses a potential threat of establishing a global competitor.

The path ahead

In Washington, the crackdown has bipartisan support. The House Select Committee on the Chinese Communist Party has opened its first investigation into Nvidia’s sales across Asia, requesting detailed records on thousands of chip transactions.

Meanwhile, Commerce Secretary Howard Lutnick has pledged to increase export control enforcement, declaring that there will a special, focused type of tariff on semiconductors.

The implications go far beyond Nvidia. ASML, the Dutch maker of advanced chip manufacturing equipment, and AMD have also felt the ripple effects. What’s unfolding is a global reshaping of the semiconductor industry that is driven as much by geopolitics as by silicon.

For now, Nvidia’s future will be determined not only by how well it continues to innovate, but by how it navigates the volatile terrain of U.S.-China competition.

The company that once simply made better graphics cards is now a symbol of the AI arms race, and a test case for what happens when technology, commerce, and geopolitics collide.