Databricks CEO Ali Ghodsi says enterprises should move from tokenmaxxing to valuemaxxing
Enterprises that launched AI projects with the biggest, most capable models are now confronting rising bills. Databricks co‑founder and CEO Ali Ghodsi calls the early phase “tokenmaxxing,” where teams equate high token usage with progress. He argues the next phase, “valuemaxxing,” should focus on what the spend actually delivers.
Key points
- Enterprises are shifting from using only large frontier models to pairing smaller models with routine tasks
- Ghodsi defines "valuemaxxing" as measuring outcomes, not just token consumption or cost
- Effective AI governance requires linking token use, infrastructure costs, and result quality
Ghodsi notes that many firms are now matching smaller, specialized models to routine tasks while reserving frontier models for complex reasoning or high‑risk work. He warns that dashboards that only track token counts and spend miss the broader picture: an AI workflow often includes document retrieval, database queries, external API calls, and compute that are not reflected in model‑cost metrics. To assess true value, organizations need to connect three data streams—AI activity, infrastructure/financial cost, and outcome quality—so they can decide when a larger model is justified and when a cheaper one suffices.
The piece stresses that visibility must stay within an organization’s security boundary, avoiding the need to export sensitive prompts or data to external services. By aligning model choice with measurable outcomes, companies can keep experimenting without uncontrolled spending or data‑privacy risks.
Why Token Costs Don’t Tell You Whether Enterprise AI Is Working
Unite.AI · 30 October 2026
Loading the full article…
This text was published by Unite.AI and written by Nils Bunge, Head of Product, Tsuga. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Enterprise & Industry
All →- Cisco assigns 27-year-old engineer to rewrite 98% of legacy code with AI in 24 months · 1 src
- Cognizant dedicates 25,000 staff to OpenAI deployments for clients · 2 src
- Ravical raises $17M seed to build AI-powered shared workflows for accounting firms · 2 src
- Verosoft launches Verovia AI platform for industrial automation · 1 src
- Microsoft adds 25 finance skills to Copilot for finance teams · 1 src
Comments
via GitHub Discussions