Nvidia Restarts H200 Production for China: Jensen Huang Confirms Orders After 10-Month Freeze

Key Takeaways

Nvidia CEO Jensen Huang announced on March 17, 2026, at the GTC conference in San Jose that the company is restarting production of its H200 AI processors for shipment to China. The announcement ends a roughly 10-month freeze on advanced chip supplies to the world's second-largest economy. Huang confirmed that Nvidia has received both US and Chinese regulatory approvals and already has purchase orders from multiple Chinese customers.

This development marks a significant shift in the geopolitics of AI hardware, reopening a critical supply line that had been closed since mid-2025 due to escalating US export restrictions.

Feature Overview

1. The Production Restart: What Changed

Speaking to reporters at Nvidia's GTC 2026 event, Jensen Huang stated: "We've been licensed for many customers in China for H200. We have received purchase orders from many customers, and we're in the process of restarting our manufacturing." Huang noted that this outlook was dramatically different from just weeks earlier, when the company had only one license for a small number of units.

The breakthrough came after months of negotiations with both the US Commerce Department and Chinese authorities. In late February, Nvidia disclosed that small amounts of H200 products for China-based customers had been approved by the US government. By mid-March, the approvals expanded to cover multiple customers at significant volume.

2. Regulatory Framework and Conditions

The H200 sales to China come with substantial regulatory conditions that reflect the ongoing tension between commercial interests and national security concerns:

These conditions represent a new model for controlled technology trade, allowing Nvidia to access the Chinese market while maintaining export control oversight. The 25% revenue share is particularly notable, effectively creating a tax on AI chip exports that funds US government programs.

3. Market Impact and Revenue Implications

China has historically been one of Nvidia's largest markets, representing approximately 20-25% of data center revenue before export restrictions tightened. The H200, based on the Hopper architecture with enhanced HBM3e memory, is Nvidia's most capable chip currently approved for Chinese export.

The financial impact could be substantial. At estimated pricing of $25,000-$35,000 per H200 unit, and with potential volume of up to 1 million processors, the China restart could represent $25-35 billion in additional revenue opportunity. Even after the 25% US government revenue share, this represents a significant growth vector for Nvidia.

Analysts have noted that while the H200 is not Nvidia's latest architecture, it remains highly capable for AI training and inference workloads. Chinese customers, starved of advanced AI hardware for nearly a year, are expected to purchase aggressively.

4. Geopolitical Context: The US-China AI Hardware Chess Match

The H200 production restart must be understood within the broader context of US-China technology competition. The US has progressively tightened AI chip export restrictions since October 2022, with each round of restrictions prompting both adaptation by Chinese companies and lobbying by US chipmakers.

Nvidia has been at the center of this tension. The company created specialized China-market chips (the A800 and H800) to comply with earlier restrictions, only to see those chips restricted as well. The H200 approval represents a pragmatic middle ground: advanced enough to be commercially valuable to Chinese customers, but with sufficient conditions to address national security concerns.

The 10-month freeze had accelerated China's domestic AI chip development efforts, with companies like Huawei's HiSilicon making progress on competing products. However, industry analysts note that Chinese alternatives still lag Nvidia by approximately 2-3 generations in performance, making the H200 restart commercially significant for both parties.

Usability Analysis

For Chinese AI companies and cloud providers, the H200 restart provides access to hardware that significantly outperforms domestically available alternatives. The H200's HBM3e memory delivers nearly double the memory bandwidth of the previous-generation H100, which translates directly into faster AI model training and more efficient inference.

However, the regulatory conditions create practical challenges. The per-customer cap of 75,000 chips means large Chinese cloud providers may need to carefully allocate resources across projects. The third-party verification requirement adds procurement complexity and potential delays. And the 25% revenue share effectively increases the cost to Chinese buyers, who ultimately bear the economic impact through higher chip pricing.

For the global AI ecosystem, the restart signals that complete decoupling of US-China AI hardware supply chains is neither practical nor desirable for either side. The controlled trade model being tested here could become the template for technology exports in other sensitive domains.

Pros

1. Reopens the world's second-largest AI hardware market for Nvidia after a 10-month freeze, creating a significant new revenue stream
2. Dual US-China regulatory approval establishes a viable framework for controlled advanced technology trade
3. H200 with HBM3e memory provides Chinese customers with hardware significantly superior to domestically available alternatives
4. Volume potential of up to 1 million processors suggests substantial commercial impact for both Nvidia and Chinese AI development
5. Third-party verification model could serve as a template for future technology export arrangements

Limitations

1. 25% US government revenue share effectively increases costs for Chinese buyers and reduces Nvidia's net margin on China sales
2. Per-customer cap of 75,000 chips constrains large-scale deployments by major Chinese cloud providers
3. Regulatory conditions could change with shifting political dynamics, creating uncertainty for long-term planning
4. H200 is not Nvidia's latest architecture as the company has already moved to Blackwell and Vera Rubin platforms for other markets

Outlook

The H200 production restart for China establishes a new paradigm in controlled technology trade. Rather than a binary choice between unrestricted exports and complete embargoes, the US has opted for a revenue-sharing model with volume caps and verification requirements. This approach allows Nvidia to maintain its commercial relationship with Chinese customers while giving the US government both financial returns and oversight capabilities.

The key question going forward is whether this model is sustainable. If US-China relations deteriorate further, the approvals could be revoked. If they improve, the conditions could be relaxed. For Nvidia, the restart provides an immediate revenue boost and demonstrates to investors that the China market is not permanently lost.

For the broader AI industry, the H200 China restart suggests that despite geopolitical tensions, the economic incentives for continued AI hardware trade remain powerful. The controlled trade model being tested here may well define the future of technology commerce between the world's two largest economies.

Conclusion

Nvidia's H200 production restart for China, announced by Jensen Huang at GTC 2026, marks the end of a 10-month hardware supply freeze and the beginning of a new era in controlled AI technology trade. With dual regulatory approvals, purchase orders in hand, and a framework of conditions including a 25% revenue share and volume caps, this development is significant for Nvidia's financials, China's AI ecosystem, and the evolving model of US-China technology relations. For investors, AI developers, and policymakers, the H200 restart is a pivotal signal that commercial pragmatism is finding a way to coexist with national security imperatives.