Shares in companies tied to the physical infrastructure of artificial intelligence fell sharply this past week, with crypto-and-compute operator Hut 8 dropping 23% as investors recalculated how much political support the AI buildout can count on going forward. The trigger was last week's US midterm elections, in which Democrats won control of both chambers of Congress, a result markets read less as a verdict on technology policy specifically and more as a signal that the federal environment shaping energy permits, tax incentives and infrastructure spending tied to data centres is about to get more complicated.
The sell-off is notable because it targets the physical layer of the AI economy rather than the software and model companies that have dominated headlines for the past several years. Data-centre operators, power-infrastructure firms and the companies that finance their construction have effectively been pricing in years of uninterrupted, federally supported expansion, and any signal that the political consensus underpinning that expansion is shakier than assumed hits their valuations disproportionately, since their business models depend on long planning horizons and predictable permitting environments.
That the sell-off coincided with continued large capital commitments elsewhere in the AI sector only sharpens the contrast. Defence AI company Shield AI closed a $2.5 billion raise, including a $1 billion Series H at a $28.8 billion valuation, and a separate open-weights chip-design model was released under a permissive licence at a fraction of the cost of comparable commercial tools. Money and technical progress are still flowing into the sector at pace; what moved this week was narrower and more specific, a repricing of political risk around the physical infrastructure that AI's continued growth depends on.
For data-centre operators with South African or African exposure, the read-through is more about financing conditions than direct impact. A cooling in US investor appetite for infrastructure plays tends to tighten the capital available for similar projects globally, even in markets where the underlying demand for compute capacity, including from businesses adopting generative AI tools, continues to grow. South Africa's own national AI policy, already delayed into next year, sits at the intersection of exactly this kind of global capital sentiment and local infrastructure constraint.
Whether this week's move proves to be a temporary repricing or the start of a more sustained reassessment of AI infrastructure valuations will depend largely on how the new Congress approaches energy and technology policy once it is seated. For now, the market has drawn a clear distinction between confidence in AI's underlying technology, still attracting billion-dollar bets, and confidence in the political conditions needed to keep building the physical capacity that technology runs on.