Skip to content
gentic.news — AI News Intelligence Platform
Connecting to the Living Graph…

Listen to today's AI briefing

Daily podcast — 5 min, AI-narrated summary of top stories

MEMORY BANDWIDTHBEFORE3.35H100AFTER4.8H200 +43.3% deltagentic.news
Auto-generated diagram from article data — Memory bandwidth
Policy & EthicsBreakthroughScore: 88

US Clears Nvidia H200 Sales to 10 China Firms, Reversing Ban

US cleared Nvidia H200 sales to 10 China firms on May 14, 2026, reversing a ban that had reduced Nvidia's China share to zero.

·20h ago·3 min read··1 views·AI-Generated·Report error
Share:
Source: reuters.comvia hn_ai_infraSingle Source
Which Chinese firms can now buy Nvidia H200 chips after the US cleared sales?

The US cleared Nvidia H200 chip sales to 10 Chinese firms on May 14, 2026, reversing a ban that had reduced Nvidia's China market share to zero, as CEO Jensen Huang seeks a market breakthrough.

TL;DR

US approved H200 sales to 10 Chinese firms · Nvidia CEO Jensen Huang seeks China breakthrough · Export controls had reduced Nvidia's China share to zero

The US cleared Nvidia H200 chip sales to 10 Chinese firms on May 14, 2026. The approval reverses a ban that had reduced Nvidia's China market share to zero, per CEO Jensen Huang.

Key facts

  • US cleared H200 sales to 10 China firms on May 14, 2026.
  • Nvidia's China market share was zero per Jensen Huang on May 3.
  • H200 offers 4.8 TB/s memory bandwidth vs H100's 3.35 TB/s.
  • Nvidia's Blackwell successor Vera Rubin ships later this year.
  • H200 lacks Blackwell's transformer engine and FP4 support.

The US government approved Nvidia to sell its H200 chips to 10 Chinese firms on May 14, 2026, according to Reuters. The H200 is a less powerful variant of Nvidia's flagship Blackwell architecture, designed to comply with export controls while still serving China's AI training demand.

How the ban hollowed out Nvidia's China business

The approval marks a sharp pivot. Just 11 days prior, on May 3, 2026, Nvidia CEO Jensen Huang stated that US export controls had reduced the company's China market share to zero [Reuters reports]. Nvidia's China revenue was effectively zero before this clearance. The 10 firms — which Reuters did not name — are likely hyperscalers and AI labs that previously purchased Nvidia's H100 and H800 chips before stricter export curbs took effect.

What the H200 clearance means for Nvidia and competitors

The H200 is a memory-bandwidth upgrade of the H100, not a new architecture. It offers 4.8 TB/s bandwidth versus H100's 3.35 TB/s, but lacks Blackwell's transformer engine and FP4 support. This clearance lets Nvidia recapture some China revenue without exposing its most advanced IP. The move also pressures competitors like Huawei's Ascend 910B and AMD's MI300X, which had gained share in China's AI chip market during Nvidia's absence.

One unique take

The clearance is less a policy shift than a calibrated leak: the US is allowing Nvidia to sell mid-tier silicon to keep the company's China engineering teams funded and prevent Huawei from owning the entire inference stack. The H200 is obsolete by Nvidia's own roadmap — Vera Rubin ships later this year — so selling it to China now poses minimal national security risk while generating billions in revenue.

What to watch

Watch for the Q3 2026 Nvidia earnings call: if China revenue exceeds 5% of total revenue, the clearance is material. Also monitor Huawei's Ascend 910B pricing — a price cut would signal competitive response. The US Commerce Department's next entity list update, expected within 60 days, will reveal if more firms are added or if the H200 clearance expands.


Sources cited in this article

  1. CEO Jensen Huang.
  2. Jensen Huang
Source: gentic.news · · author= · citation.json

AI-assisted reporting. Generated by gentic.news from 4 verified sources, fact-checked against the Living Graph of 4,300+ entities. Edited by Ala SMITH.

Following this story?

Get a weekly digest with AI predictions, trends, and analysis — free.

AI Analysis

This clearance is a calibrated leak, not a policy reversal. Nvidia's H200 is a mid-tier chip — obsolete by the company's own roadmap given Blackwell and Vera Rubin — so selling it to China now generates billions in revenue without exposing cutting-edge IP. The zero-market-share statement from Huang on May 3 was likely a negotiating tactic to pressure the US government, and it worked within 11 days. Competitors like Huawei and AMD face asymmetric pressure. Huawei's Ascend 910B had gained inference share during Nvidia's absence, but H200's CUDA ecosystem and memory bandwidth advantage make it a stronger training option. AMD's MI300X, which lacks the software maturity of CUDA, will struggle to retain China customers who now have a familiar alternative. The structural story is that US export controls are becoming a tax, not a ban: the most advanced silicon stays home, but mid-tier chips flow freely. This creates a two-tier global AI chip market that benefits Nvidia's margins (selling older silicon at high prices) while containing China's access to frontier models.
Compare side-by-side
H200 vs Blackwell
Enjoyed this article?
Share:

AI Toolslive

Five one-click lenses on this article. Cached for 24h.

Pick a tool above to generate an instant lens on this article.

Related Articles

From the lab

The framework underneath this story

Every article on this site sits on top of one engine and one framework — both built by the lab.

More in Policy & Ethics

View all