Earn with Nexo
Hedge-fund manager says AI ‘bottleneck trade’ has run its course
Atreides Management's Gavin Baker argues easing supply constraints are reshaping who wins in the AI infrastructure gold rush, with implications for crypto miners pivoting to compute
For the past couple of years, the easiest money in AI investing came from a simple thesis: find the chokepoint, buy the company that controls it, and wait. That trade is over, according to one prominent hedge-fund manager.
Gavin Baker, chief investment officer of Atreides Management, declared on June 16 that the AI “bottleneck trade” has run its course. The firm manages between $4 billion and $7 billion in assets and focuses heavily on technology. Baker is also an early SpaceX investor.
What the bottleneck trade actually was
Baker identified the specific chokepoints that made this trade so lucrative. TSMC wafer capacity, power generation, cooling solutions, optics, and networking equipment were all in desperately short supply as AI buildout accelerated globally.
Companies that controlled any piece of that constrained supply chain could essentially name their price. Demand was so intense that Baker estimated unconstrained annual Nvidia GPU demand at $2 to $3 trillion.
Who wins in phase two
Baker indicated the next wave of AI investment winners will be hyperscalers, the massive cloud infrastructure operators, and specialized “neocloud” operators that have demonstrated robust return on investment and the ability to scale operations efficiently.
The crypto miner crossover episode
Several public Bitcoin miners have been pivoting their facilities toward AI and high-performance computing workloads. MARA and Iris Energy (IREN) are two notable examples. Both companies have been repositioning infrastructure originally built for Bitcoin mining to serve the booming demand for AI compute. The logic is straightforward: they already have the real estate, the power contracts, and the cooling infrastructure.
Baker’s thesis introduces a complication. If the bottleneck trade is ending, then simply having power and cooling capacity is no longer enough to command premium pricing. These crypto-turned-AI operators now need to compete on execution, customer relationships, and operational efficiency against hyperscalers with vastly deeper pockets.