Demand for computing power for artificial intelligence is still far from its peak. In the coming years it will not just grow, but accelerate — this is the forecast made by investor and host of the All-In podcast Jason Calacanis, and former Apple and Rivian engineer Phil Beisel backed it up with calculations.
According to Beisel's models, by 2035 demand for AI compute could grow hundreds and thousands of times relative to the current level. The key argument is that the industry is at the very starting point of the adoption curve.
Why talk of an AI peak is premature
Calacanis estimates that today artificial intelligence is used by only about 2% of corporate America's employees. At the same time, neither OpenAI nor Anthropic has yet launched AI agents on a mass scale. The analyst's conclusion is unambiguous: the industry has only scratched the surface of what is possible, and growth will be exponential rather than linear.
According to Calacanis's estimate, AI token consumption will grow roughly 500-fold from current levels. The investor calls those who talk about market saturation or a limit to compute demand out of touch with reality — in his view, such people simply don't use the technology.
Calacanis also noted that neither the U.S. midterm elections nor interest rates will stop this trend. The market will not move in a straight line upward, but investors should not be distracted by the noise — what is happening should be seen as the largest technological shift of their lifetime.
What Beisel's models show
Phil Beisel, founder of technical teams at Apple and Rivian, said Calacanis is "1000% right" and published a chart with several scenarios for compute demand growth. The starting point is 2026, and the forecast horizon extends to 2035.
The chart has four lines. The steepest is the scenario labeled Grok AI: demand grows about 4.34 times per year and already by 2031 goes off the scale. Beisel's own model without adjustments gives growth of 3.4 times per year. The version accounting for improvements in hardware and model efficiency grows more slowly — 2.27 times per year — but even it shows growth of roughly 1600 times by 2035.
For comparison, Beisel added a dotted line for historical growth in the CPU era — about 60% per year. By 2035 it rises only to around 70. The main thesis: wherever the market moves, the green line does not apply to it — that is what the previous stage of computing development looked like.
Short-term market fluctuations should not obscure the long-term trend: demand for AI infrastructure, according to the experts, is only just beginning to accelerate.
Cryptalist analysis: Forecasts of a 500-fold and even 1600-fold increase in demand for AI compute reflect a fundamental shift in the structure of capital expenditures in the technology sector, not speculative euphoria. For the crypto market, this is a signal of a long-term shortage of computing resources, which strengthens the investment appeal of decentralized computing and GPU infrastructure projects. However, investors should take into account the non-linearity of the trajectory: short-term volatility and cooling cycles do not cancel this scenario, but they complicate its realization.