Future generations of AI systems will continue to accelerate': New report claims that by 2030, the world could generate approximately 317 billion AI tokens every second
Five billion AI tokens a second today, 317 billion by 2030? SanDisk's forecast starts on solid ground but ends on assumptions it hasn't published.
New research suggests AI systems could generate an astounding 317 billion tokens per second by 2030, according to a report from SanDisk. This figure represents an increase of more than 60 times in just four years, requiring global token output to nearly triple each year. SanDisk's calculations are based on internal analysis, focusing on the capacity of the world's AI infrastructure rather than demand forecasts.
The company notes that this is an illustrative figure and does not include specific assumptions about future power, compute, memory, networking, and storage capabilities. The 2026 projection aligns with Google's disclosures, but independent estimates also point to a similar range. The growth rate is highly ambitious, with some experts questioning the feasibility of such rapid expansion.
SanDisk's interest in this projection stems from its focus on enterprise SSDs, which can store and serve as a larger tier for AI model cache data, helping to avoid redundant computations. While the technology exists to support these projections, the sheer scale of the growth required remains a significant challenge.
Rewritten by urgent.news; may contain errors — check the original.