The AI infrastructure buildout has entered a new phase — one where capital allocation and supply chain sovereignty matter as much as raw computational power. But as billion-dollar contracts signal that monetization is finally arriving, a set of geopolitical fault lines are widening beneath the sector's feet.
Nvidia, the undisputed backbone of the AI arms race, has watched its China revenue deteriorate sharply under successive waves of U.S. export controls. Washington's restrictions on advanced chip exports — targeting H100 and A100-class GPUs — have effectively severed Nvidia from one of its largest historical markets. China once represented a significant revenue contributor; now that stream has been choked, forcing Nvidia to redirect supply toward U.S. hyperscalers and sovereign AI programs while leaving a gap that domestic Chinese chipmakers like Huawei's Ascend division are racing to fill.
The implications extend beyond Nvidia's quarterly filings. The export control regime has created a bifurcated global semiconductor market — one that structurally disadvantages Western chipmakers seeking scale while accelerating China's push toward indigenous alternatives. For investors holding positions in Nvidia, AMD, or the broader fabless semiconductor space, this is not a temporary headwind; it is a permanent reconfiguration of the addressable market.
Rare Earths: The Invisible Chokepoint
Less visible but potentially more destabilizing is the rare earth element (REE) dependency embedded throughout AI infrastructure. Semiconductors, data center power systems, and networking hardware all rely on REEs — neodymium, dysprosium, terbium — for magnets, phosphors, and specialized alloys. China controls roughly 60% of global rare earth mining and over 85% of processing capacity.
Recent U.S. legislative scrutiny around REE procurement bans reflects a growing recognition that supply chain sovereignty is an unresolved vulnerability. Any escalation in U.S.-China trade tensions — tariff expansion, retaliatory export controls on processed REEs — could introduce meaningful cost shocks into semiconductor manufacturing and data center construction timelines. The $1.17 billion CoreWeave infrastructure contract and SoftBank's pursuit of Marvell Technology are the kinds of capital-intensive commitments that assume stable input costs. That assumption deserves scrutiny.
Valuations Priced for Perfection
Haim Israel, among others, has noted that investors cannot ignore the risks of bottomless AI spending against sky-high valuations. The current infrastructure wave — CoreWeave, Marvell, data center REITs, power equipment makers — is priced for a scenario where supply chains hold, geopolitics remain manageable, and enterprise AI adoption accelerates linearly. None of those conditions are guaranteed.
Even enterprise software players like Teradata, guiding for modest 2–4% ARR growth in full-year 2026, reflect a more measured reality: AI adoption is real but uneven, and capital efficiency is increasingly the metric that separates durable business models from overhyped ones.
What Investors Should Watch
The immediate watchlist for semiconductor and infrastructure investors includes U.S. REE policy developments, any Chinese retaliatory measures on mineral exports, and Nvidia's ability to sustain margin expansion as China revenue remains constrained. Advances in reasoning efficiency — such as FGO chain-of-thought compression techniques that reduce computational overhead — may partially offset hardware demand pressure, but they do not resolve the geopolitical risk premium now embedded in the sector.
The AI trade is maturing. Geopolitical risk is the new valuation variable.