China’s DeepSeek prices new V4 AI model at 97% below OpenAI’s GPT-5.5

DeepSeek cut API prices sharply, taking input cache-hit costs to one-tenth of prior levels and offering an additional 75% discount on V4-Pro through May 5; the minimum input cost is now about US$0.14 per million tokens, while V4-Pro can be as low as US$0.0036 per million input tokens. The move makes DeepSeek-V4 far cheaper than OpenAI’s GPT-5.5 and could intensify price competition across the AI model market. Usage on OpenRouter jumped, with DeepSeek V4-Pro hitting 13.6 billion tokens on April 25, nearly 4x the prior day.

Analysis

This is less about a single model price cut and more about a structural reset in unit economics for agentic workloads. If DeepSeek can undercut incumbents by orders of magnitude while maintaining near-frontier performance on developer-relevant tasks, the competitive moat shifts away from raw model quality and toward distribution, workflow integration, and enterprise trust. That tends to compress margins first at the low end of the market, then forces incumbents to subsidize usage via bundling, which is negative for standalone API monetization across the stack.

The second-order winner is the infrastructure layer that can absorb volume growth cheaply, especially domestic compute ecosystems and orchestration platforms. Lower inference prices expand the addressable market for always-on agents, but that also increases token consumption per workflow; the net effect is likely higher aggregate demand for GPUs, networking, and inference software even as per-token prices fall. The near-term loser is any vendor exposed to usage-based pricing with weak differentiation, because procurement teams will now anchor on DeepSeek-level economics in renewal conversations.

The market is probably underestimating how fast this becomes an enterprise budgeting issue rather than a model race. In the next 1-2 quarters, the key catalyst is whether other Chinese labs follow with parallel cuts; if they do, it reinforces a deflationary spiral in API pricing and pressures global peers to defend share. The main contrarian risk is that ultra-low prices can be a loss-leader strategy that widens adoption but does not translate into durable pricing power, meaning the real monetization may accrue to the pick-and-shovel layer rather than the model providers themselves.

Key Decisions for Investors

• Long GS on a 3-6 month horizon as a relative beneficiary of accelerated enterprise AI adoption and capex/financing advisory flow; risk/reward improves if the price war drives more platform consolidation and M&A, but trim if AI margin pressure starts hitting broader tech multiples.
• Pair trade: long AI infrastructure enablers vs short API monetizers with weak moats over the next 1-2 quarters; prefer semis/networking/software orchestration over standalone model vendors, as volume growth should offset lower unit pricing at the infra layer.
• If available, buy 6-12 month call spreads on high-quality compute beneficiaries into any weakness caused by fears of AI price compression; thesis is that cheaper inference raises total token consumption faster than it destroys upstream demand.
• Avoid chasing long exposure to pure-play model providers with heavy reliance on premium API pricing for the next earnings cycle; the risk is renewal-rate compression and promotional spend rising faster than customer LTV.
• Set a 30-60 day alert for follow-on pricing actions from competing labs; if multiple peers cut simultaneously, consider adding to short baskets of software names with AI gross-margin sensitivity.