ReadingResearchRadarInvestment framework
Sign in / Sign up
← Back to reading
NewsMaterial2 min readDisconfirmAI Editorial

SanDisk estimates global AI token output could reach 317 billion per second by 2030

Flash memory maker SanDisk projects a 60-fold increase in AI token generation by 2030, though the figures are an internal capacity ceiling rather than independent demand.

2026-11-21

In its Tokenomics 2030 outlook published on November 21, 2026, flash memory manufacturer SanDisk estimated that global AI systems will generate approximately 150 quadrillion tokens this year, or roughly five billion every second. By 2030, the company believes annual output could approach 10 quintillion tokens, equivalent to about 317 billion per second. This implies a more than 60-fold growth over four years, requiring global token output to nearly triple annually until the end of the decade.

The report explicitly states these figures are drawn from SanDisk's own internal analysis and are "intended for illustrative purposes only," with no outside data sources named. The model calculates how much token output the world's infrastructure could realistically support based on power directed toward AI inference, accelerator fleet size, compute and memory bandwidth, networking, storage, and expected software efficiency gains. Consequently, the 317 billion figure represents a capacity ceiling rather than a demand forecast.

For context, Alphabet CEO Sundar Pichai disclosed at Google I/O in May 2026 that Google alone was handling 3.2 quadrillion tokens a month, which translates to roughly 1.2 billion tokens per second—about a quarter of SanDisk's global estimate from a single company. Independent tallies from I/O Fund placed industry-wide volume near 370 trillion tokens a day, or about 135 quadrillion a year. Goldman Sachs Research projected in May that token consumption would grow 24-fold between 2026 and 2030 to reach 120 quadrillion tokens per month; SanDisk's 2030 projection is roughly seven times higher than Goldman's monthly rate.

The commercial logic behind SanDisk's outlook centers on the KV cache, the attention state a model builds as it reads a prompt, which agents need to keep around across long, multi-step sessions. SanDisk argues that keeping all of it in scarce, high-bandwidth memory will not always be practical or economical, and that enterprise SSDs can serve as a larger tier for selected cache data so systems avoid recomputing work they have already done. However, the outlook's most important input is also its weakest link: power availability. Gartner analyst Linglan Wang has warned that "AI capacity is now constrained by power availability," and Phison CEO Pua Khein-Seng relayed a NAND maker's estimate that new capacity takes around four years to reach mass production, pushing supply difficulties out to 2030.

Sign in / Sign up
ReadingResearchRadarInvestment framework

Sources

  • techradar.comTechRadar↗
Continue researching →