The past few weeks have been particularly challenging for AI infrastructure stocks, with the sector correcting by more than 20%. While the magnitude of the sell-off has been significant, we have witnessed similar corrections several times over the past three years, and each has ultimately been followed by a sharp recovery. We believe this episode will prove no different and view the current weakness as a compelling buying opportunity.
The correction was initially triggered by reports suggesting that Meta was looking to sell “excess” computing capacity, raising concerns that AI infrastructure demand—and by extension hyperscaler capex—might be approaching a peak.
Those fears now appear increasingly misplaced. In recent days, Meta CEO Mark Zuckerberg clarified in an interview with Bloomberg that the company has no excess compute and, in fact, needs every unit of computing capacity it can secure. He added that a portion of Meta’s infrastructure could generate higher economic returns if temporarily rented to third parties in a market where AI compute remains severely constrained.
Zuckerberg’s words have been followed by actions. Earlier this week, the company announced that it would invest $50 billion in its Hyperion AI supercluster in Louisiana, expanding the project to approximately 5 GW of computing capacity. That represents a dramatic increase from the $27 billion investment outlined last October and the 2 GW capacity originally planned.
Meta’s seemingly insatiable demand for compute was further corroborated by an internal memo reviewed by Reuters, indicating that the company expects to finish 2026 with approximately 7 GW of AI computing capacity before adding another 7 GW during 2027, bringing total installed capacity to roughly 14 GW by the end of 2027. This compares with an estimated 3–4 GW at the end of 2025 and illustrates just how rapidly Meta’s AI infrastructure ambitions continue to expand.
Rather than pointing to a slowdown in AI investments, these developments suggest that expectations for Meta’s capex—particularly in 2027—may ultimately prove too conservative. The same conclusion can be drawn from recent capital-raising initiatives by Google and SpaceX, which owns xAI. Both companies have raised more than $80 billion, providing substantial financial firepower to accelerate AI infrastructure deployment and increasing the likelihood that hyperscaler capex once again surprises to the upside when results are reported over the coming weeks.
Additional industry data points continue to support this constructive view. Strong earnings and robust order books from semiconductor equipment leaders such as ASML and Aehr Test Systems point to continued capacity expansion across the AI/photonics semiconductor supply chain. Meanwhile, IBM’s underwhelming quarterly results were, somewhat paradoxically, supportive of the AI infrastructure investment thesis. IBM noted that portions of its traditional hardware and software businesses were being pressured as enterprise customers increasingly redirected IT budgets toward servers and storage, in other words AI infrastructure.
Taken together, these developments reinforce our conviction that the recent correction reflects a temporary deterioration in sentiment rather than a weakening of underlying fundamentals. If anything, the evidence continues to point toward an acceleration in AI infrastructure investment. Against this backdrop, we continue to view the current pullback as an attractive opportunity to add exposure to high-quality AI infrastructure companies.
