Jefferies’ Chris Wood called Japan’s crash and the U.S. housing bubble. He’s now warning of an ‘AI capex arms race’

Christopher Wood, Jefferies Hong Kong’s global head of equity strategy, warned that an “AI capex arms race” began in 2023 after Microsoft’s investment in OpenAI and that most profits so far are accruing to infrastructure providers—semiconductor makers, data-center builders and the “picks and shovels” of AI—raising the risk of an eventual over‑investment bust if capex fails to monetize. He has sold his Nvidia stake after its roughly fivefold run and shifted exposure to China, arguing Chinese firms are more pragmatic—focusing on energy-efficient, application-led deployments and open‑source models—and that U.S. semiconductor export controls since late 2022 have accelerated China’s domestic chip push. The broader implication is a structural shift as hyperscalers move from asset‑light to asset‑heavy models, driven by FOMO and disruption fears, which heightens industry capex intensity and market risk if revenue catch‑up does not materialize.

Analysis

Christopher Wood of Jefferies Hong Kong warned of an "AI capex arms race" that he traces to Microsoft’s 2023 investment in OpenAI, arguing that most returns to date have accrued to infrastructure providers — semiconductors, data-center builders and "picks and shovels" like Nvidia — rather than the application-layer firms. He highlighted that Nvidia has delivered roughly a five-fold gain for holders, a performance he believes largely prices in extraordinary expectations, and has therefore sold his NVDA position while concentrating remaining AI exposure in China.

Wood and panelists contend that U.S. export controls since late 2022 have accelerated China’s domestic chip push and that controlled chips have still reached China via secondary channels, strengthening Chinese competitors such as Huawei; he also emphasized China’s pragmatic, application-focused, open-source approach and its relative advantage on energy. At the same time U.S. hyperscalers are shifting from "asset-light" to "asset-heavy" models by investing in proprietary frontier models, creating heightened capex intensity and the risk of an eventual over-investment bust if spending does not monetize.

For investors the key trade-offs are clear: infrastructure players have captured near-term profits but future returns depend on capex-to-revenue conversion and data-center utilization, while China-exposed and energy-linked players may benefit from a regional divergence in AI strategy and policy-driven supply-chain reconfiguration. Monitor capex growth, margin trends and policy developments closely as potential triggers for rapid repricing of both hardware suppliers and hyperscalers.

Key Decisions for Investors

• Consider trimming or taking profits in richly valued winners such as NVDA after an approximate five-fold run to lock gains and reassess downside risk relative to priced-in expectations
• Maintain selective exposure to AI infrastructure names (semiconductors, data-center builders) but allocate only to companies with clear monetization paths and demonstrated pricing power while monitoring capex-to-revenue conversion and data-center utilization metrics
• Increase research and selective exposure to China-focused AI applications and energy/compute enablers given Beijing-driven semiconductor acceleration and pragmatic deployment, while actively monitoring export-control developments and using hedges for large capex-sensitive positions