- Businesses are cutting AI spending by dynamically routing workloads to cheaper open-source models — including DeepSeek and Alibaba’s GLM — reserving premium OpenAI and Anthropic models only for complex tasks, achieving cost reductions of up to 95%.
- DeepSeek’s share of AI usage jumped from 1% to 17% in one month on Vercel’s platform; on OpenRouter, it’s been the most-used AI company since mid-May, with open-source usage among top customers growing 4x faster than closed-source.
- Anthropic’s Fable 5 model is more than 50 times more expensive per token than DeepSeek’s V4 Pro, highlighting the magnitude of the price gap that enterprise buyers are now actively exploiting.
- OpenAI is weighing drastic token price cuts ahead of expected similar moves by Anthropic — a price war that would deepen losses at two companies already burning billions annually, ahead of their anticipated IPOs.
What Happened?
A new layer of AI infrastructure has emerged — model-routing platforms and agent orchestration tools that dynamically assign tasks to the cheapest AI model capable of handling them. For simple tasks like email triage, meeting transcription, or bug identification, companies are routing to open-source Chinese models like DeepSeek V4 or Alibaba’s GLM, which cost a fraction of frontier US models. Only for complex reasoning or coding tasks do these systems escalate to OpenAI’s ChatGPT or Anthropic’s Claude. Startups like Lindy, Factory, and Lovelace AI are building this routing logic into their products, and executives report AI cost reductions of up to 95% for some workloads. A Citadel Securities report this week cited this dynamic as a driver of recent declines in widely tracked AI spending indices.
Why It Matters?
This is the commoditization scenario that investors in OpenAI and Anthropic have feared since both companies first sought billion-dollar valuations. The interchangeability of AI models — at least for a large swath of enterprise tasks — is no longer theoretical; it’s being operationalized at scale by real companies saving real money. DeepSeek’s explosive growth on developer platforms in a single month illustrates how quickly market share can shift when price differentials are extreme. The timing is particularly painful: both OpenAI and Anthropic have filed confidentially for IPOs, meaning any further margin compression or evidence of commoditization directly undermines their public market valuations. Sam Altman himself has acknowledged costs are “a huge issue” — a remarkable admission from the CEO of the world’s highest-valued AI startup.
What’s Next?
OpenAI’s anticipated token price cuts will be the next catalyst — if Anthropic matches, a full price war begins and the path to profitability for both companies extends further out. Frontier models from US labs remain approximately four to six months ahead of open-source competitors on benchmarks, which provides some justification for premium pricing on genuinely complex tasks. But the addressable market for “frontier-only” tasks is narrowing as open-source quality improves. Microsoft’s new smaller efficient models and Nvidia’s Nemotron family add more competition from the domestic side. The structural question for investors is whether OpenAI and Anthropic can race down the cost curve fast enough to maintain margins as the market bifurcates into commodity compute and a shrinking premium tier.
Source: The Wall Street Journal
