All articles tagged with #tokenmaxxing
OpenAI CFO Sarah Friar told investors that enterprise customers now account for the majority of revenue, surpassing the ChatGPT-driven consumer business, with a $40 billion annualized run rate and rapid enterprise growth; executives addressed leadership changes, open-source competition, and a growing advertising business while signaling pricing efficiency and continued focus on monetizing enterprise use.
AI teams are moving from aggressive token budgeting to modelmaxxing—using cheaper models for routine work while reserving frontier models for high-IQ tasks—driven by new routing tools from startups that automatically steer tasks to cost-effective options. With leaders like Coinbase’s Brian Armstrong predicting most workloads will run on cheaper models within 12–18 months, the industry is shifting how companies manage AI budgets and model selection.
Meta’s heavy use of Gemini AI tokens prompted Google to impose limits, disrupting Meta’s internal AI projects; Meta shifted from tokenmaxxing to token-efficient usage as part of a broader cost-cutting push, while Google’s capacity constraints reportedly spurred SpaceX/xAI compute deals, with other large firms also facing caps.
Microsoft CEO Satya Nadella cautioned that tokenmaxxing isn’t always valuable and urged employees to match AI models to the task, avoiding frontier-model panics for non-frontier problems, while highlighting practical tools like Copilot auto mode and a vibe-coded project updater that generates plans and updates from workplace conversations.
Facing soaring token costs, Big Tech is rethinking AI adoption: Amazon and Uber cap usage, OpenAI's Sam Altman calls token usage a huge issue, and firms pivot toward leaner edge‑computing models to cut token bills; GitHub's pay‑per‑token plan struggles, while Microsoft and Google push edge solutions to reduce cloud reliance, even as data‑center energy and water concerns linger.
A viral remark by Uber COO Andrew Macdonald about AI token usage not clearly delivering productivity gains has intensified Silicon Valley’s backlash against tokenmaxxing. As major firms push AI internally, reports show ballooning budgets and unclear ROI, with Google CEO Sundar Pichai warning about rising costs. While some defend tokenmaxxing as a necessary phase, others urge tying spending to concrete metrics rather than token counts; Jellyfish’s study suggests top token users don’t proportionally outperform peers, underscoring calls for better cost controls. Investors are divided, from caution about an AI bubble to defenders like Garry Tan, reflecting a wider debate over AI’s financial sustainability and real-world impact.
Visa reports its internal AI token consumption has surged to about 1.9 trillion per month by March, doubling February, and the company is rewarding teams for AI-driven results rather than output alone. Most usage is in software engineering, but AI adoption is growing in marketing and other areas, with tools like Claude Sonnet, ChatGPT, and Google Gemini popular internally. About 89% of Visa employees are active AI users and 44% are power users (≥25 prompts per day for 15 days a month). The trend—dubbed tokenmaxxing—signals broader corporate AI adoption beyond Silicon Valley, echoing high token usage seen at other firms like Meta.
NYT columnist Kevin Roose reports that employees at Meta and OpenAI are being evaluated on how quickly they burn through AI tokens, with managers rewarding heavy AI usage and chastening those who use AI less. The trend, tied to agentic AI tools like OpenClaw and Claude Code, has led to staggering costs (one OpenAI engineer reportedly used 210 billion tokens) and high-profile claims like GPT-5.4 processing 5 trillion tokens per day. Critics say token-based metrics distort incentives and budgets and may incentivize unnecessary AI use.