Gavin Baker, managing partner and CIO of Atreides Management, has decided to step back from what has been one of the most lucrative AI infrastructure investments over the past year. The strategy, known as the bottleneck trade, focused on companies that held critical positions in the AI supply chain. However, as physical shortages improve, this investment style is losing its effectiveness.
#What Is the Bottleneck Thesis?
The bottleneck thesis centers around key chokepoints in the AI supply process. Important areas include TSMC wafer capacity, power generation, cooling systems, optics, and networking equipment. Firms that occupy these critical positions could dictate their pricing, driven by demand for AI computing that has vastly exceeded supply across all levels.
To highlight the imbalance, Baker provided a significant statistic, estimating that the unrestrained demand for Nvidia GPUs could potentially reach $2 to $3 trillion annually.
A noteworthy aspect of this situation is TSMC’s cautious management of wafer capacity. By carefully regulating supply, TSMC has played a crucial role in averting what Baker warns could have evolved into an exaggerated AI bubble.
#Where Should Investors Focus Next?
As the market shifts, Baker indicates a preference for two specific categories of investments. First, investors should look at hyperscalers—companies with the financial resources and customer relationships necessary to scale AI deployment effectively. Second, those interested in the evolving technology landscape should consider neocloud operators. These newer cloud service providers are designed specifically for AI workloads and have showcased strong return metrics.
Atreides Management, which manages assets between $4 and $7 billion, appears to be aligning itself with these emerging trends, reflecting a strategic repositioning in its investment approach.
#What Are the Implications for Crypto Investors?
In the crypto arena, several publicly listed Bitcoin mining companies, such as MARA and IREN, are redirecting their data center assets from mining towards AI computing. The rationale is straightforward: leveraging existing power-intensive facilities with cooling setups to pivot to AI workloads may yield higher returns than continuing to mine Bitcoin, especially given the pressure on profit margins following events like halvings.
If Baker's insights hold true and the bottleneck trade's advantages are diminishing, companies adjacent to the crypto space will face a complicated environment. On one hand, the alleviation of constraints in power and cooling suggests increased competition from dedicated AI data centers. Conversely, companies that have already transitioned may enjoy the benefits of a larger shift towards operators demonstrating solid AI infrastructure capabilities.