PitchBook: global AI chip sales grew 60% YoY to $35.9B in 2021; semiconductor startups got $1.8B in VC funding in 2021, up from $1.4B in 2020 and $1.1B in 2019
The raw computational power necessary to use machine learning has dwarfed everything else we use computer chips to accomplish by an order of magnitude.
Context & Ripple Effects
PitchBook's 2021 tally extends a funding arc that began with VCs putting over $1.5B into chip startups in 2017, nearly double two years prior, as AI workloads started reshaping silicon demand. What the new numbers expose is a mismatch: AI chip sales hit $35.9B, up 60% year-over-year, while semiconductor startups attracted only $1.8B — the demand for AI compute was growing far faster than the venture capital flowing to the companies trying to build it.
That gap explains why the money eventually moved up the stack. By H1 2025, AI startups were capturing 53% of all global VC dollars, and by Q1 2026 that concentration had deepened further — capital chased the model builders and applications rather than the chipmakers underneath them.
First-order effects
- Established chip vendors, not startups, absorbed the bulk of the $35.9B in 2021 AI chip revenue — the 60% growth rate accrued mostly to incumbents with fabs and existing product lines, leaving the $1.8B in startup funding competing for a narrow design-win pipeline.
- Buyers of AI compute faced a market where demand was compounding at 60% annually against a startup ecosystem funded at roughly 5% of one year's chip sales, tightening supply and pricing power for whoever could ship.
Second-order effects
- Venture allocators responded by funding the application layer instead: once AI companies proved they could absorb capital at scale, the share of VC dollars going to AI climbed from single-digit chip-startup allocations to a majority of all global venture funding within four years.
- The imbalance pushed AI companies themselves into becoming chip buyers of record — their model-training needs, not traditional end markets, became the marginal driver of semiconductor orders, shifting pricing leverage toward compute suppliers.
Third-order effects
- The pattern points to a structural split between who builds compute and who pays for it: as later data showed, quarterly AI sales only narrowly exceeded estimated data center and chip depreciation costs, meaning the sector's margins rest on sustaining the very capex cycle its own growth created.
- If capital keeps concentrating in AI while chip startup formation stays thin, the industry consolidates around a few integrated compute providers — with the financing terms of that compute, not chip design alone, determining which AI companies survive downturns.
The trend: This is an early data point in the AI infrastructure capital cycle, where compute demand compounds faster than the capital and capacity supplying it, concentrating both revenue and risk in whoever finances the hardware layer.