Artificial Intelligence

Agents Burn 50x Tokens, HPE Pitches On-Prem Fix

Agents use 20x to 50x more tokens than chatbots, and HPE claims on-prem routing can cut 60% of cost.

StartupHub.ai Staff

By StartupHub.ai Staff, reviewed by Daniel Singer, Editor

2 min read

Agents consume 20x to 50x more tokens than chatbots, Sadeepa Wijesekara told The TWIML AI Podcast with Sam Charrington, a gap that wipes out falling per-token prices. There is no venue, odds or volume to quote because this is an enterprise interview, not a traded market. Not financial advice. Markets move fast.

The bill still rises even as list prices fall.

Buying Tokens vs Building an AI Factory - TWIML
Buying Tokens vs Building an AI Factory, TWIML

Wijesekara, vice president of Private Cloud AI at HPE, said HPE forecasts a large token spend internally and sees the same surge at customers once developers get broad access. He said running suitable work on HPE Private Cloud AI on prem could save up to 60% of token cost, a claim made after host Sam Charrington raised Apple, Uber and Meta moves to cap internal token budgets.

The second number is 60 to 70% utilization, the level at which Wijesekara said HPE still calculates a return against public frontier models. He tied that math to GPU choice and scheduling, with H200, RTX 6000 and B300 class hardware matched to memory needs and night jobs filling idle capacity. The figures do not show measured value per task, and Wijesekara said most firms lack agent-level telemetry to prove it.

Routing is where he would start. Customers pick deterministic rules for which prompts stay local on open-weight models and which need frontier reasoning, then watch the split before trusting automated routers, Wijesekara said. Compliance and data gravity often force the call, since some data cannot leave the building even when a frontier model would help.

HPE frames that fix as a turnkey appliance with lifecycle management and solution engineers, while Iternal brings Blockify ingestion to the same platform to turn unstructured content into governed IdeaBlocks for every app. What would change the read is audited cost per completed job, including SaaS AI add-ons that sit outside tagging today. Will call-by-call placement keep agent bills flat as usage climbs 50-fold?

Mentioned in this story

Markets

Sources

Get the free daily brief every morning

Every morning the day's AI news, new AI startups and new funding rounds. Free and takes a few minutes to read, unsubscribe at any time.

Join 64,000 other subscribers.

TopicsHPESadeepa WijesekaraSam CharringtonPrivate Cloud AIIternalAppleNvidia

StartupHub.ai StaffDaniel SingerWritten by StartupHub.ai Staff, Editorial teamReviewed by Daniel Singer, EditorAbout

StartupHub.ai Staff, Editorial team

The staff writers of StartupHub.ai, ranging from investment analysts to avid AI tool users, early adopters and critical enthusiasts. Backgrounds span engineering, business and the arts. We hold every piece to rigorous standards of research and review.

Daniel Singer, Editor

Daniel Singer is an AI expert who has been writing about, covering and building AI technology since 2018.

More from Daniel Singer

Test your Agent Readiness now

Paste your website. How well ChatGPT, Claude and the other agents can read and use a site, in about a minute.

Watch

The videos behind the news, one flip at a time

The feed mixes the videos from our articles with startup demos and you flip through them like TV channels for free with no signup.