“The first wave of enterprise AI consisted of impressive demonstrations and obnoxiously high bills,” says Gogia. “CIOs quickly learned that the cost of AI is never just model invocation, but also includes retrieval, orchestration, and other components.”
However, the 75 percent reduction is only really relevant if CIOs can use the model on a large scale, he continues.
“For most companies, the relevant comparison is not DeepSeek’s direct API, but rather what it costs to deploy on-premises versus an external inference provider,” explains Amit Jaju, senior managing director at Ankura Consulting. “If a CIO can host DeepSeek V4‑Pro on their own infrastructure, inference costs drop dramatically and many projects that were previously economically unviable suddenly become realistic. These include permanently active copilots, large document review, code generation, L1 support and multi-agent workflows.
However, if the model is used through third parties, the actual costs may be higher and the ROI benefits may be lower, Jaju said.
The pressure on AI prices continues to increase
DeepSeek’s aggressive pricing strategy is likely to increase pressure on large AI providers, whose models often command premium prices in the enterprise environment. This could force OpenAI, Anthropic and Google to offer more attractive pricing and package models.
Shah notes that high-margin, token-based pricing models from Anthropic and OpenAI have become increasingly difficult to justify for many business applications. The existence of a powerful open weights alternative gives companies considerable room for maneuver. This will likely encourage Western AI labs to gradually move away from purely consumption-based pricing towards more value- or outcome-based monetization models.
