Companies Are Scrambling to Stop Spending So Much on AI
AI spending is creating a "tokenpocalypse" as companies and developers scramble to cut costs on large language models. Firms are exploring alternatives like smaller models, caching strategies, and open-source solutions to reduce reliance on expensive API calls from major AI providers.
Background
- "Token" here refers to the fundamental unit AI models process (roughly ¾ of a word). Every API call costs money per token — both input (your prompt) and output (the AI's reply).
- DeepSeek, a Chinese AI lab, recently released a model that reportedly achieves comparable performance to GPT-4 at a fraction of the cost, shaking assumptions about pricing.
- The article describes companies panicking as they realize their AI bills are unsustainable, and racing to adopt cheaper models, local hardware, or usage limits — a shift from the "spend whatever it takes" AI gold rush of 2023-2024.
- Key players: OpenAI (GPT-4, expensive), Anthropic (Claude, also pricey), DeepSeek (cheaper Chinese competitor), and various startups facing a "tokenpocalypse" — a Wall Street–style term for suddenly unaffordable AI costs.