AI compute infrastructure stocks have had a rough few weeks. Grayscale thinks that’s the point.
In an analysis published on October 20, 2026, Grayscale Head of Research Zach Pandl said these stocks have fallen approximately 27% since the end of September 2026. He framed the slide as an opening, not a warning sign. The firm’s view is that demand for AI compute remains strong and is expected to keep growing. Grayscale sees the current dip as an opportunistic entry point into high-performance computing equities.
Why the stocks fell, and why Grayscale isn’t worried
According to the analysis, the decline stems from concerns that higher borrowing costs will weigh on capital-intensive businesses. Building and running compute infrastructure means spending enormous sums upfront on hardware, power, and facilities. When financing costs climb, those projects look less attractive on paper.
Grayscale’s argument is that the financing story and the demand story are two separate things. One affects how cheaply companies can build. The other determines whether anyone wants what they’re building. On the second question, Pandl’s note is unambiguous: the appetite for compute hasn’t gone anywhere.
Inference is doing the heavy lifting
The firm says AI compute demand is shifting toward inference workloads. Training is the expensive, one-time process of teaching an AI model. Inference is what happens every time someone actually uses that model, whether that’s asking a chatbot a question or having an agent complete a task.
Grayscale points to two forces pushing inference higher. The first is lower token prices. Tokens are the small units of text that AI models process and bill for, so cheaper tokens make AI usage more affordable. The second is rising utilization of AI agents, the software tools designed to carry out multi-step tasks on a user’s behalf.
The supply side can’t move fast
The analysis highlights two major bottlenecks. Access to power is one, since data centers need enormous amounts of electricity and securing it is not quick or easy. The other is construction time. Grayscale estimates that building a data center takes 2 to 5 years.
Grayscale expects this mismatch to likely sustain a demand-supply imbalance for the foreseeable future.
Background: a Bitcoin miners ETF with a new job
This isn’t Grayscale’s first word on the subject. The October note builds on research the firm published in September 2026 that examined compute bottlenecks and rising demand.
On September 22, 2026, the firm rebranded its Bitcoin Miners ETF as the Grayscale AI Compute ETF (GCPU), shifting its focus to AI computing. The rebrand landed just days before the selloff began at the end of September.
Grayscale also isn’t alone in its optimism. Morgan Stanley shared similarly bullish views on the AI infrastructure market in July 2026.
What this means for investors
The core question is whether the recent decline reflects a real change in the sector’s prospects or a temporary repricing driven by financing costs. Grayscale is firmly in the second camp.
Grayscale’s case leans heavily on falling token prices driving higher usage. Cheaper tokens are good for users. For providers, the bet only works if volume growth outpaces the drop in price per unit.
For the crypto industry, the GCPU rebrand shows a major crypto asset manager steering a mining-focused product toward the AI compute theme. Investors holding or considering that fund are effectively taking a view on the same thesis Pandl laid out: demand is durable, supply is constrained, and the recent drop is noise rather than signal.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.