Breaking

DigitalOcean CEO Projects Token Demand Surge by 2030, With Agents Generating 70%

June 30, 2026 at 10:08 EDT

AI Agents
Infra & Chips
Sales & Business

DigitalOcean CEO Paddy Srinivasan projected that AI token consumption will reach 4 to 5 quintillion (10^18) per year by 2030, with roughly 70% of those tokens generated by autonomous AI agents. As AI shifts from conversational chatbots to "doing"—systems that plan and execute autonomously—he framed the forecast as a structural change in cloud demand.

April 28, 2026 · DigitalOcean Deploy 2026

Agents Will Eat the Tokens: 70% of a Quintillion-Scale Demand

CEO Paddy Srinivasan projects that once AI consumption hits 4-5 quintillion (10¹⁸) tokens per year, autonomous agents will generate roughly 70% of them — workloads "multiple orders of magnitude bigger" than human-driven chat.

70%

of all tokens generated by agents at quintillion scale

15×

tokens of agentic vs. human users

4×

CPU capacity vs. traditional workloads

50-90%

of agent work runs on CPUs for orchestration

Global token demand toward 2030

Daily consumption is expected to grow roughly tenfold — a ~10× jump.

~50T / day

Today

→

500T+ / day

By 2030

One agent task ≠ one chat reply

A single agentic task chains together far more than a one-shot response.

100s

model calls

100s

database queries

1M+

tokens per task

Reported deployment results

-42%

LawVo cut inference cost with no code changes (500M+ tokens/week, 130+ agents)

71%→94%

RAG accuracy gain using Knowledge Bases

20-40%

cheaper than rivals — 1M bookings/mo travel agent at $67,727

The opportunity

Token efficiency in agentic loops — via caching and MoE routing — is seen as a practical lever as inference overtakes training and open models reach quality parity at lower cost.

The caveats

Long-term pricing is hard to forecast amid surging consumption, shifting inference onto CPUs adds challenges, and many cited cases are vendor-provided — broad independent validation is still pending.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.