SanDisk’s Tokenomics 2030 report estimates that AI systems worldwide currently produce about five billion tokens per second, with potential growth to 317 billion tokens per second by 2030, reflecting a rapid expansion of AI computational output capacity.
- 2026 AI token production estimated at five billion tokens/second
- Projected 317 billion tokens/second capacity by 2030
- Growth projection based on infrastructure capability, not demand
What happened
SanDisk released its Tokenomics 2030 report, estimating that AI systems globally generated around 150 quadrillion tokens in 2026, roughly five billion tokens each second. The firm forecasts this number could escalate to as much as 10 quintillion tokens annually by 2030, equating to approximately 317 billion tokens per second. This projection represents a roughly 60-fold increase in token output over four years.
These figures derive from SanDisk’s internal analysis focusing on the potential capacity of global AI infrastructure, including compute power, memory bandwidth, and storage. However, the company has not disclosed detailed assumptions or external data sources behind these forecasts, positioning the numbers more as illustrative capacity limits rather than definitive demand predictions.
Why it matters
The report shines light on how AI expansion depends heavily on the growth and scaling of hardware and supporting infrastructure, especially the storage technologies crucial for managing vast token data. As AI applications proliferate, understanding the infrastructure capacity needed to handle token processing at scale becomes key for technology providers and investors.
Comparisons with other industry data reveal disparities in growth assumptions. For example, Google’s recently reported token processing rates suggest that SanDisk’s 2026 baseline is plausible, yet its 2030 projection assumes a deceleration in growth compared to recent trends. Goldman Sachs’ more conservative forecast also indicates varying expectations for token volume growth, underlining the uncertainties inherent in predicting AI compute demand.
What to watch next
Future AI infrastructure developments, including increased efficiency in compute and memory utilization, will directly impact token generation capacity. Monitoring disclosures from leading AI companies like Alphabet and industry analysts will help validate or challenge SanDisk’s projections as real-world demand unfolds.
Additionally, further clarity on token definition and measurement across different firms will be essential to accurately compare and assess token volume forecasts. The pace of AI adoption, regulatory developments, and advances in software optimization are all factors to watch that could significantly influence token generation trajectories toward 2030.