Nvidia’s H200 AI chip remains caught in a familiar but increasingly consequential bind: approved in principle, constrained in practice, and emblematic of how U.S.-China technology relations now operate in slow motion rather than absolutes.

Nearly two months after U.S. President Donald Trump gave the green light for exports, sales of the H200 to China have yet to resume in any meaningful way. The delay, as reported by the Financial Times, is not a technical issue nor a question of demand. It is the product of a layered national security review that has exposed fault lines within the U.S. government itself and reinforced uncertainty for Chinese buyers already wary of sudden policy reversals.

At the center of the impasse is the licensing process imposed in January, when the Commerce Department eased export curbs on the H200 but required applications to be reviewed not just internally, but also by the departments of State, Defense, and Energy. That structure reflects how AI chips are no longer treated as ordinary commercial goods, but as strategic assets with implications for military capability, intelligence gathering, and long-term economic power.

According to people familiar with the discussions, Commerce has completed its assessment, suggesting that from a technical export-control standpoint, the H200 can be sold under defined conditions. The sticking point appears to be the State Department, which has argued for tighter restrictions to prevent China from deploying the chips in ways that could undermine U.S. national security. That includes concerns around large-scale AI model training, dual-use applications, and the potential for civilian infrastructure to be repurposed for state or military ends.

This internal pushback matters because it signals that even when export rules are formally relaxed, enforcement and interpretation can remain fluid. For Nvidia, that creates a grey zone where executive assurances do not immediately translate into purchase orders.

Chinese customers, according to the FT, are holding off on placing H200 orders until it becomes clear not only whether licenses will be granted, but also what strings may be attached. Those conditions could include limits on volumes, end uses, data center configurations, or post-sale compliance obligations.

Nvidia CEO Jensen Huang has publicly expressed hope that sales will proceed, saying last week that the license is being finalized. His comments point to confidence that a pathway exists, but they also underscore how dependent the company has become on regulatory discretion rather than straightforward market access.

Over the past two years, Nvidia has repeatedly redesigned and repositioned chips to fit within U.S. rules, turning export compliance into a core part of product strategy.

Reuters reported last month that China approved its first batch of H200 chips for import, a move seen as a pragmatic shift by Beijing. China’s leadership faces its own balancing act: sustaining rapid AI development in the near term while accelerating domestic chip capabilities to reduce reliance on U.S. suppliers. Allowing limited imports of the H200 fits that approach, buying time for local players even as Washington seeks to cap how much advanced capacity China can access.

The uncertainty reinforces a trend already underway for Chinese firms. Cloud providers, AI startups, and research institutions are diversifying supply chains, testing domestic alternatives, and adapting software stacks to work across multiple hardware platforms. Even if Nvidia ultimately secures licenses, the stop-start nature of access weakens its long-term position by encouraging customers to plan for a future where U.S. chips cannot be assumed to be available.

From Washington’s perspective, the drawn-out review reflects a deeper debate about the effectiveness of export controls. Tight restrictions risk pushing China to innovate faster at home, potentially eroding U.S. leverage over time. Looser controls, meanwhile, raise fears of enabling technological advances that could narrow the strategic gap in areas Washington considers sensitive.

The H200 sits squarely in that tension: powerful enough to matter, but already a step behind Nvidia’s most advanced offerings reserved for unrestricted markets.

The episode also illustrates how inter-agency dynamics now shape global tech markets. Decisions are no longer binary approvals or bans, but negotiated outcomes influenced by competing priorities across departments. That means longer timelines, higher compliance costs, and greater revenue volatility for companies like Nvidia.

Until the national security review concludes and license conditions are spelled out, Nvidia’s H200 is expected to remain in a holding pattern. The delay may eventually be resolved, but the broader message, which is: access to advanced AI hardware is now a matter of statecraft as much as commerce, has already been sent.

Shares of Advanced Micro Devices fell sharply in early premarket trading on Wednesday after the chipmaker’s first-quarter revenue outlook failed to live up to the market’s loftiest expectations.

The sharp pullback in early Wednesday trading underscored a growing reality in the AI-driven chip rally: strong earnings are no longer sufficient when expectations are stretched to extremes.

The stock slid about 9% in premarket trading after the company’s first-quarter outlook failed to fully satisfy investors who had been positioned for an even more aggressive forecast, given the scale of global spending on artificial intelligence infrastructure. The selloff came despite AMD delivering a solid fourth quarter that beat Wall Street estimates and reinforced its status as one of the most important challengers to Nvidia in the AI chip market.

AMD reported fourth-quarter revenue of $10.27 billion, topping LSEG consensus estimates of $9.67 billion. The result capped a year in which the company benefited from surging demand for data-center processors, particularly graphics and accelerator chips used in AI training and inference.

For the first quarter, AMD projected revenue of $9.8 billion, plus or minus $300 million. While the midpoint was still above the broader market estimate of around $9.38 billion, some analysts had expected a more forceful signal that AI-related demand would drive a steeper sequential ramp.

That disconnect between expectations and guidance proved costly for the shares.

“Expectations were pretty sky high,” said Chris Rolland, a semiconductor analyst at Susquehanna, in comments on CNBC.

He added that AMD’s disclosure of China-related revenue shipments in the quarter, which were not fully reflected in analysts’ models, made the headline beat appear stronger than it otherwise would have been.

“When you account for that, the beat was far less substantial than we would’ve thought,” Rolland said.

China exposure and regulatory risk

The reference to China highlights a sensitive area for U.S. chipmakers. Export controls imposed by Washington have limited the types of advanced AI chips that can be sold into the Chinese market, forcing companies like AMD and Nvidia to redesign products to comply with restrictions.

Any revenue tied to China is closely watched by investors, both for sustainability and for the risk of further regulatory tightening. AMD did not provide extensive detail on how much of its recent growth was linked to China-specific products, but the mere presence of that revenue added complexity to the market’s assessment of underlying demand.

Despite the near-term disappointment, there was little indication that AMD’s longer-term AI story has weakened. Demand for its data-center products remains strong, and analysts say the company continues to signal large-scale deployments ahead.

Rolland noted that AMD has hinted at multi-gigawatt AI contracts, a scale that underscores how rapidly computing requirements are expanding as companies race to deploy and monetize AI systems.

That trajectory is reinforced by AMD’s recent strategic partnerships. In October, the company announced a landmark agreement with OpenAI, under which the ChatGPT developer could take up to a 10% equity stake in AMD. As part of the deal, OpenAI plans to deploy 6 gigawatts of AMD Instinct GPUs over several years, starting with an initial 1-gigawatt rollout in the second half of 2026.

The partnership positions AMD as a core supplier in one of the world’s most visible AI ecosystems and marks a significant endorsement of its hardware roadmap.

In addition, Oracle said it will deploy 50,000 AMD AI chips beginning later this year as it expands cloud capacity to meet rising demand for AI workloads from enterprise customers.

Valuation pressure in an AI market

AMD’s stock has more than doubled over the past year, fueled by optimism that it can capture meaningful share in a market long dominated by Nvidia. That rally, however, has also raised the bar for performance.

Investors are increasingly demanding not just growth, but clear evidence that AI-related revenue will scale rapidly enough to justify current valuations. Any hint of moderation — even in the context of a beat-and-raise quarter — risks triggering sharp reactions.

The response to AMD’s guidance mirrors a broader pattern across AI-linked stocks, where earnings season has become less about whether companies are benefiting from AI, and more about how quickly that benefit is accelerating.

Looking ahead, the focus will be on how quickly AMD can convert its growing list of AI partnerships into sustained revenue growth, particularly in its data-center segment. Investors will also watch for clearer disclosure around the mix of training versus inference workloads, competition with Nvidia’s next-generation chips, and the impact of export controls on international sales.

For now, Wednesday’s selloff suggests that the AI boom has entered a more demanding phase. For AMD, the long-term opportunity remains intact, but the market is signaling that optimism alone is no longer enough.