Two semiconductor stories this week that seem like business news are actually foreign policy. Intel has joined Elon Musk's Terafab project, an effort to build a new U.S. semiconductor factory in Texas alongside SpaceX and Tesla. Meanwhile, Firmus, Nvidia's Asia-based AI data center bet, has hit a $5.5 billion valuation after raising $1.35 billion in six months. The same capital moment, two geographies, one underlying panic: whoever controls the physical substrate of AI controls the future of computation, and right now nobody is sure which side of which border that substrate should sit on.
The Physical AI Investment Thesis
Eclipse's $1.3B physical AI fund, the Terafab announcement, and Firmus's rapid capitalization all belong to the same investment thesis: that the next decade of AI value creation is not in software but in hardware, land, power, and fabrication. This is a significant departure from the asset-light VC playbook of the 2010s. TurboFund's April 2026 AI VC roundup documents how dramatically the investor appetite for physical infrastructure has shifted in just 18 months. What is notable about Terafab is that it is being assembled from existing industrial players, Intel, SpaceX, Tesla, rather than purpose-built. It is a consortium model dressed as a startup.
Nationalism as a Product Roadmap
The chip nationalism framing is not incidental. The CHIPS Act created the incentive structure. The Iran conflict and the Iranian hacker advisory, with U.S. agencies warning of escalated attacks on critical infrastructure, supplied the urgency narrative. Firmus going deep in Asia while Terafab anchors in Texas is not a contradiction. It is a hedge. The companies most exposed to geopolitical risk are building on both sides of the emerging tech blocs, which means they are not betting on an outcome. They are betting that fragmentation itself is the durable condition. A 2025 paper in World Development by Mazzucato and Dibb found that state-backed industrial policy in semiconductors consistently underperforms private coordination timelines but generates path-dependent infrastructure advantages that outlast any single administration. Terafab may be chaotic. The fab itself, if built, will matter for decades.