DeepSeek had a "discount" running on its flagship AI model. It was supposed to end May 31. Now it isn't ending at all.
The Chinese AI startup just made the 75% price cut on its V4 Pro model a permanent number, and the rest of the industry has to react.
DeepSeek V4 Pro's top-tier price just dropped from $3.48 to $0.87 per million tokens. The cheap end fell just as far, settling at a fraction of a cent per million tokens.
In plain English: enterprise users burning through billions of tokens a day got a big bill cut. For the rest of the AI industry, this is a price floor that didn't exist a week ago.
The company released V4 Pro and V4 Flash about a month ago. DeepSeek marketed them as the start of the "era of cost-effective" long-context AI, and the new rate card bakes that pitch into pricing.
Long context matters here because pricing scales with how many tokens a model has to handle. A discount that hits hardest on huge inputs is a discount built for enterprise workloads.
Want the daily breakdown of what's moving in AI and tech? Market Briefs lands in your inbox every weekday morning, and a free investing masterclass comes with sign-up.
The list of names sitting at higher prices is short and familiar. OpenAI's GPT-5, Google's Gemini 3.5 Flash, and Anthropic's Claude.
Anthropic has accused DeepSeek of "distillation attacks," meaning training cheaper models by copying outputs from Claude. The pricing gap just got wider regardless of who's right.
For enterprise customers, the math gets simple. If two models do the same job, the one charging a fourth of the price wins more deals.
This isn't an academic comparison anymore. Buyers running AI agents at scale are now looking at swap-in options that quietly cut their inference bills by 75%.
DeepSeek is reportedly chasing a $45 billion valuation. Cutting prices and giving up margin to grab market share lines up with that pitch - own the demand now, monetize later.
It's the same playbook Amazon ran in retail. Lose money on price, win the customer, build the moat.
The move also fits the broader Chinese AI strategy. State-level support and access to home-built compute let the model providers absorb hits to margin that Western rivals can't easily match.
That's a structural change. Rivals like Anthropic and OpenAI are still selling to enterprises one big deal at a time, while DeepSeek competes on rate card.
The pressure point isn't OpenAI. It's the smaller AI startups stuck in the same price band, because they can't undercut a Chinese giant with deep funding.
For investors, the trade splits in two. AI compute and chips still ride the demand wave, because more usage means more tokens. AI software margins, on the other hand, just got harder to defend.
Watch the next earnings call from any AI-pure software name. The first time a CFO blames DeepSeek pricing for soft guidance, the narrative shifts.
Join Market Briefs for the daily five-minute read - and grab a 45-minute course on finding investments as a free bonus.
