Nvidia B300 Hits $1M in China; Pentagon Signs AI Deals for Classified Use

Nvidia's B300 hits $1M in China as export enforcement bites, while the Pentagon signs AI deals for classified operations.

🗄️ Nvidia B300 Server Prices Reach $1 Million in China as Chip Smuggling Crackdown Tightens Supply

Decoded: Nvidia's B300 AI servers — which sell for approximately $550,000 in the United States — have reached prices of roughly 7 million yuan, or about $1 million each, in China, according to four industry sources cited by Reuters in an April 30 report. The near-doubling reflects a compression of the black-market supply channel that had been providing Chinese companies with export-restricted Nvidia hardware: stricter enforcement of U.S. chip export controls has made smuggling significantly harder. Chinese AI companies and cloud providers face sustained high demand for AI compute despite Nvidia being formally banned from selling its most advanced chips into China. The B300 is Nvidia's latest Blackwell-architecture GPU system, intended for large-scale AI training and inference. The price surge follows months of increased enforcement by U.S. Customs and Southeast Asian trade authorities, which had previously served as gray-market routing points. (Reuters, April 30, 2026)

Why it matters: A 100% price premium on Nvidia B300 servers in China is a real-time measure of how thoroughly U.S. export enforcement is closing the black-market channel that had partially offset chip restrictions. Chinese AI labs — including Baidu, ByteDance, and Alibaba Cloud — now face the choice of paying $1 million per B300 unit through gray-market channels, scaling on Huawei Ascend hardware, or reducing AI infrastructure ambitions. For Nvidia (NVDA), the premium creates an ironic signal: Chinese demand remains intense enough to sustain double the list price, but Nvidia cannot legally capture that revenue. For U.S. policy, $1 million market prices confirm enforcement is generating scarcity — the intended effect — while simultaneously confirming that Chinese AI investment is not deterred by cost alone.

🏛️ Pentagon Signs AI Deals With OpenAI, Google, and xAI for Classified Operations

Decoded: The U.S. Defense Department confirmed on May 1 that it has signed agreements with OpenAI, Google, and Elon Musk's xAI to deploy their AI models in classified settings, authorized for "any lawful use" of the technology, the New York Times reported. The Pentagon's contract language — allowing deployment for any lawful government purpose — had initially been resisted by Anthropic, which was the first AI company to make its models available on classified government infrastructure. The DoD confirmed Anthropic subsequently joined the classified framework as well. The agreements represent a formalization of the U.S. government's AI procurement strategy, placing frontier commercial models into classified defense and intelligence operations rather than relying on government-developed systems. (New York Times, May 1, 2026)

Why it matters: Pentagon access at the classified level is qualitatively different from standard government cloud contracts: it means OpenAI, Google, and xAI models are authorized to process information at the highest sensitivity tiers in the U.S. national security apparatus. The "any lawful use" language removes restrictions that governed commercial AI in prior government procurement, giving defense and intelligence agencies broad discretion to apply these models to intelligence analysis, logistics planning, and operational support. For frontier AI companies, classified DoD contracts represent sticky, long-duration revenue streams with minimal competitive pressure — the opposite of consumer subscriptions that turn over annually. The inclusion of xAI alongside OpenAI and Google also formalizes Musk's company as a peer competitor in the defense AI market, against an OpenAI organization he is currently suing in civil court. For Palantir (PLTR), which pioneered classified AI infrastructure for U.S. defense agencies and built the AIP platform around military and intelligence use cases, the formalizing of OpenAI, Google, and xAI classified contracts introduces direct competition in the segment Palantir has treated as a structural moat.

Stay decoded. See you tomorrow.

— The Get AI Decoded Team