The managing partner argues supply constraints that fueled a year-long trade are easing, forcing investors to rethink what actually has lasting value in AI infrastructure
Gavin Baker, managing partner and CIO of Atreides Management, is calling time on one of the most profitable AI infrastructure trades of the past year. The so-called “bottleneck trade,” built around betting on companies that controlled scarce resources in the AI supply chain, is losing its edge as physical shortages ease.
The bottleneck thesis, explained #
The critical chokepoints included TSMC wafer capacity, power generation, cooling systems, optics, and networking equipment. Companies positioned at these bottlenecks could essentially name their price because demand for AI compute far outstripped supply at every level.
Baker offered a striking data point to illustrate just how severe the demand-supply mismatch has been. He estimated that unconstrained Nvidia GPU demand could reach $2-3 trillion annually. Baker specifically highlighted TSMC’s disciplined approach to wafer capacity as a critical factor. By not flooding the market with supply, TSMC effectively prevented what Baker suggests could have been a full-blown AI bubble.
Where the smart money goes next #
The rotation Baker describes favors two categories. First, hyperscalers with the balance sheets and customer relationships to dominate AI deployment at scale. Second, neocloud operators, the newer breed of cloud providers specifically built around AI workloads, that have demonstrated strong return on investment metrics.
Atreides Management, which oversees assets in the range of $4-7 billion and specializes in technology investments, appears to be positioning accordingly.
What this means for crypto investors #
Several publicly traded Bitcoin mining companies, including MARA and IREN, have been reallocating data center resources away from mining and toward AI compute. The logic is simple: if you already own power-hungry facilities with cooling infrastructure in place, pivoting to AI workloads can generate better returns than mining Bitcoin, especially when mining margins compress after halvings.
If Baker is right that the bottleneck trade is winding down, these crypto-adjacent firms face a nuanced reality. On one hand, the easing of power and cooling constraints means they’ll face more competition from purpose-built AI data centers. On the other hand, the companies that have already completed their transitions may benefit from the broader market rotation toward operators with proven AI infrastructure capabilities. Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our