90% Off and $600M In: The Week Enterprise AI's Economics Shifted
DeepSeek cut API cache prices to a tenth of their original cost and launched V4 Preview. Days later, Cohere and Aleph Alpha announced a merger. The timing is not a coincidence.
The price floor for AI API access just dropped again — and this time it's not a promotional tease. DeepSeek's simultaneous launch of V4 Preview and a 90% cut to input cache prices (as of April 26, 2026) signals that the Chinese lab is done competing on capability alone; it's now competing on cost. That same week, Cohere and Aleph Alpha announced plans to merge with $600M in new funding — a data point that's hard to read as coincidence.
The DeepSeek moves are worth examining together. DeepSeek announced on X that cache hit pricing across its entire API series is now 1/10th of the original price, effective immediately. Simultaneously, DeepSeek V4 Preview went live on web, app, and API, promising top-tier reasoning at a fraction of what Western labs charge for comparable outputs. The 75% off promotion on V4-Pro was already running; the cache price cut on top of it is a structural shift, not a promotional one.
Meanwhile, SiliconAngle reported that Cohere and Aleph Alpha intend to merge, pairing Cohere's North American enterprise NLP strength with Aleph Alpha's European sovereign AI positioning, backed by $600M in fresh capital. Neither company has competed primarily on raw model capability — their value proposition has always been enterprise trust, data sovereignty, and deployment flexibility. That framing matters less when the underlying model economics collapse beneath you.
Yann LeCun offered a useful frame this week: innovation from industry is enabled by scientific breakthroughs that precede it by 5, 10, or 20 years. The implication is that what looks like a pricing war is actually the harvesting of earlier academic work — and the companies best positioned to harvest it efficiently win on cost before they win on features. DeepSeek has been explicit about this approach from the start.
For builders, the cache price drop is immediately practical: workflows that rely on repeated context — RAG pipelines, long-session agents, document Q&A — just got materially cheaper to run on DeepSeek's API. For enterprise buyers evaluating Cohere or Aleph Alpha, the merger adds near-term procurement complexity but may produce a stronger, unified vendor within 12-18 months. For the broader market, the consolidation is a signal: when 2 well-funded, domain-specific companies merge rather than compete, the pricing pressure from foundation model providers has become structural, not cyclical.
The bottom line: If you're building on third-party model APIs, benchmark your costs against DeepSeek V4's April 2026 pricing — the gap to Western providers has grown wide enough to justify a serious evaluation.
Sources
- Shooting oneself in the foot? Nope. Shooting oneself in the prefrontal cortex. https://t.co/3FW2kt2b
- @techpupparent Why don't backups fix this? Good backups should be the ultimate protection against
- btw we are cooking something with @hhua_ (not final yet but keep calendar open after ICML in Seoul
- @theo @uwukko I’m not giving you shit. It’s for your followers.
- Had to record this one for posterity on my blog https://t.co/ht9EtrXV4R
- DeepSeek 🎉 DeepSeek-V4 Preview is here ... top-tier reasoning. Now available on web, app, and API.
- @theo @uwukko Are they on the Codex open source plan? Do you know?
- @theo @uwukko when are we doing the codex plan giveaway stream