ai-costs

The author introduces a new financial metric for software companies called EBIT (Earnings Before Inference Tokens), aiming for a 75% growth margin.

Compares AI token consumption to digital employee salaries, predicts token costs will match or exceed employee wages, and discusses how businesses measure ROI and control costs.

A discussion about the AI industry bubble, noting that investment is not keeping up with consumer demand, leading to service limits and potential unaffordability. The author argues that overvaluation and speculative investment will cause a correction, affecting smaller players.

Microsoft has canceled internal licenses for Anthropic's Claude Code due to unexpectedly high costs from token-based billing, highlighting a broader industry shift where AI usage is blowing through annual budgets in months.

OpenAI is offering $2M in tokens to Y Combinator startups, which could make AI tokens much cheaper and solve the cost problem for consumer AI ideas.

Microsoft canceled internal Claude Code licenses due to untenable token-based costs; Uber burned through its 2026 AI budget in four months. This signals the end of the AI subsidy era as enterprise budgets clash with rising model prices.

The article discusses how the rising availability of cheap AI models from Chinese labs and other competitors threatens the valuation and market position of OpenAI and Anthropic ahead of their planned IPOs, as enterprise customers increasingly seek cost-effective alternatives.

A practical guide listing 10 strategies to reduce costs when using LLM APIs, including model selection, prompt caching, batch processing, and monitoring expenses.

A developer shares the hidden cost variables that cause AI bills to exceed estimates, including reasoning model chain-of-thought tokens, multimodal per-image charges, and function calling system tokens, and asks the community how they predict costs upfront.

Token costs are emerging as a key enterprise concern for AI adoption, with CIOs struggling to manage spending across different models and use cases. OpenAI announced Guaranteed Capacity to address long-term compute access.

A discussion on effective FinOps strategies for managing costs in large-scale AI agent operations, covering tactics like model routing, prompt trimming, caching, and the need to track cost by agent, workflow, and customer.

The article argues that the real challenge in AI isn't just building smarter models but making them cost-efficient at scale, highlighting the importance of reducing token usage, improving speed, and optimizing infrastructure.

A tweet highlights that companies are finding the cost of training employees in AI is higher than the salary of the employees they recently laid off.

Ex-Google CEO Eric Schmidt states that the real limit to AI is financial, not energy, estimating that 10 gigawatts of compute could cost half a trillion dollars, which only a few entities like the US or China can afford.

Research laboratories are grappling with rising AI subscription costs and usage limits from providers like OpenAI and GitHub, raising questions about the cost-benefit ratio for scientific research.