• Commerce Secretary Howard Lutnick emphasizes prioritizing vast American demand for AI chips like Nvidia (NVDA)'s H200 over exports to China, aligning with a January 13, 2026 policy revision.
  • The new rules require exporters to certify sufficient U.S. supply before shipping to China, amid global DRAM shortages persisting through 2026 and Chinese firms ordering over 2 million H200s against Nvidia's limited inventory.
  • Lutnick's push for onshoring includes a goal to boost U.S. leading-edge semiconductor share from 0% to 40%, backed by $1.2 trillion in investments, while critics flag risks from Nvidia's technical aid to China-based AI firms.

In a move that underscores the intensifying U.S.-China tech rivalry, Commerce Secretary Howard Lutnick has made it clear that American demand for advanced AI chips must come first. Speaking at a recent keynote, Lutnick stressed, "We don't want to do anything to stand in their way," referring to domestic buyers like hyperscalers and data centers scrambling for Nvidia's H200 chips with HBM3E memory. This stance aligns with a January 13, 2026, policy revision by the Commerce Department's Bureau of Industry and Security (BIS), which now mandates that exporters obtain an "America First" certification—verifying that U.S. customers face no delays—before shipping such chips to China.

Efforts to restructure export controls have hit a snag, however, as global DRAM shortages loom. According to people familiar with the matter, Chinese tech firms have placed orders for over 2 million H200 units, dwarfing Nvidia's current inventory of roughly 700,000. In a January 14 letter, industry watchers commended the new rules but warned that shortages from major producers like Samsung (SSNLF), SK Hynix (000660.KS), and Micron (MU) could persist through 2026, potentially undermining U.S. prioritization. Micron's CEO has noted that supply is "substantially short" long-term, while Samsung executives warn of broad impacts across the semiconductor ecosystem.

Without a deal to balance supply, the U.S. risks ceding ground in the AI race. Lutnick's CSIS address highlighted an ambitious onshoring agenda: boosting the U.S. share of leading-edge semiconductors from near-zero to 40%, fueled by $1.2 trillion in investments, including Micron's $100 billion HBM fab and plans to produce 2 million Blackwell chips domestically. But this push comes amid political friction, with the House Select Committee on China urging stricter enforcement after Nvidia's technical aid to China-based DeepSeek AI, which has since integrated with the People's Liberation Army (PLA). A January 28 letter criticized these ties, calling for briefings by February 13 to refine H200 export rules.

Market reactions have been muted so far, with Nvidia shares holding steady amid the regulatory scrutiny. The company, led by CEO Jensen Huang, faces a probe over its DeepSeek optimizations, which enabled "frontier" AI performance on deprecated hardware—a workaround that echoes past export control evasions. Meanwhile, international dynamics complicate the picture: a U.S.-Taiwan deal involves $250 billion in investments for tariff relief, but Taiwanese officials deem a 40% capacity shift to the U.S. "impossible," citing logistical hurdles. Section 232 tariffs, set at 25% on advanced logic chips from January 15, 2026, add another layer of protectionism, though critics argue they could exacerbate supply chain bottlenecks.

In brief comments, a Commerce Department spokesperson reiterated that the rules aim to "secure U.S. technological leadership without disrupting domestic innovation." Attempts to reach Nvidia for further input were unsuccessful, but industry insiders note that the company is navigating a tightrope between global demand and regulatory compliance. As DRAM shortages tighten, the short-term outlook suggests China's H200 exports may face further blocks, with briefings expected by late January or early February to tweak the certification process. Long-term, Lutnick's vision hinges on that $1.2 trillion buildout, but experts like those at Samsung predict ongoing supply issues, while Congress eyes potential bans on AI models like DeepSeek.

Correction: An earlier version misstated the timeline for DRAM shortages; they are projected through 2026, not indefinitely. Updates on the U.S.-Taiwan investment deal are pending as negotiations continue.