Enterprises shift to cheaper DeepSeek models even as API prices jump

The core development is enterprise cost-driven model switching: reports say major companies like AT&T are moving production workloads to DeepSeek's cheaper models, trading brand-name frontier models (Claude, GPT) for substantial savings. This is a concrete data point in the week's dominant theme — the commoditization of mid-tier AI and enterprises' growing willingness to prioritize price over brand.

The twist is that DeepSeek is simultaneously raising prices dramatically — up to 1,100% on its V4-Flash and V4-Pro models — as it drives toward profitability ahead of a planned IPO. Even after such steep increases, the models reportedly remain cheaper than Western frontier alternatives, which underscores just how large the starting cost gap was. Founded by Liang Wenfeng in 2023, DeepSeek has been a defining disruptor in open-source and low-cost AI.

Competitively, DeepSeek's move fits a broader pattern of Chinese labs (alongside Alibaba's Qwen3.8-Max-0902) pressuring Western pricing while chasing sustainable economics. The price hikes are a maturation signal: the era of below-cost pricing to win share may be ending as these labs prepare for public markets.

The caveats enterprises must weigh are the same ones raised around Qwen: China-law compliance and data-governance considerations on every call, plus the risk of further price increases eroding the savings that justified the switch. Readers should watch whether the AT&T-style migrations stick after the 1,100% hikes, and how DeepSeek balances IPO-driven margin pressure against the cost advantage that built its enterprise traction in the first place.