OpenAI's Custom Chip and China's $7 Billion AI Bet Signal a New Phase in the Compute Race
OpenAI unveiled its first custom-built inference chip on June 24, developed with Broadcom and known internally as "Jalapeño," marking a shift in the United States-China technology contest away from software development and toward control of the underlying computing stack. The announcement came as Chinese computing firm DeepSeek sought roughly $7 billion in new investment, a move analysts and national security observers say reflects Beijing's determination to build frontier AI capabilities independent of American technology.
OpenAI unveiled its first custom-built inference chip on June 24, developed with Broadcom and known internally as "Jalapeño," marking a shift in the United States-China technology contest away from software development and toward control of the underlying computing stack. The announcement came as Chinese computing firm DeepSeek sought roughly $7 billion in new investment, a move analysts and national security observers say reflects Beijing's determination to build frontier AI capabilities independent of American technology.
Infrastructure Becomes the Contested Terrain
The two developments, taken together, signal what national security analysts describe as a second phase in the AI competition — one defined not by which country writes better models but by who controls the full chain of chips, energy, data centers, and cloud systems that power advanced AI. Huawei has continued expanding its domestic semiconductor ecosystem targeting AI applications, while Xi Jinping has directed the People's Liberation Army to develop what he calls "new quality combat capabilities," a phrase Chinese military doctrine explicitly ties to machine-intelligence-enabled warfare. Beijing has pursued a comprehensive national strategy combining state-backed financing, civil-military fusion, and domestic semiconductor development aimed at building a sovereign computing ecosystem.
White House Warns of Industrial-Scale Capability Theft
Washington's concern extends beyond chip counts. White House science and technology director Michael Kratsios warned in an April memorandum that Chinese entities are conducting "deliberate, industrial-scale campaigns" using tens of thousands of proxy accounts to extract functional capabilities from leading American AI models. The technique, which the White House formally terms "adversarial distillation," involves mass queries, coordinated probing, and jailbreaking rather than traditional source-code theft — allowing competitors to train rival systems from the outputs of American models without breaching a network perimeter. The White House has formally accused Chinese entities of running such operations at scale.
Supply Chain Contradictions Undercut the Buildout
The infrastructure boom carries its own vulnerability. Much of the electrical hardware required to support large-scale AI data centers — including transformers, switchgear, and power management equipment — remains dependent on foreign manufacturing, with significant exposure to Chinese suppliers. China also retains dominant global supply positions in the critical minerals and rare-earth elements those systems require. Washington has moved to tighten semiconductor export controls and increase Pentagon investment in autonomous and decision-support systems, but the supply chain dependence remains unresolved, creating what observers describe as a strategic contradiction: competing with China on AI infrastructure while relying on Chinese-linked supply chains to build it.
Policy Response Seen as Lagging the Threat
Recent federal actions — expanded chip export controls, new attention to model security, increased defense investment — indicate Washington recognizes the problem. Whether the pace of response matches the pace of Chinese investment remains the central question for policymakers and markets tracking the sector.