The availability of low-cost AI models from companies like DeepSeek and Baidu is leaving technology chiefs wondering when enterprise AI deployment costs will materially drop.

But the answer is far from certain.

DeepSeek’s rollout early this year might have put pressure on incumbents like OpenAI, Anthropic and Google, but analysts say it’s unclear if CIOs will see lower AI costs. This is because AI product development and use patterns are constantly evolving, adding complexity to vendors’ pricing strategies.

If the underlying infrastructure gets cheaper but usage patterns increase, companies won’t necessarily pay less, analysts say. This dynamic creates a challenge for CIOs and CFOs who want to maintain stable AI costs while deploying AI use cases that will set them apart from their competitors.

CIOs “want to see predictability in how much AI is going to cost today versus how much AI is going to cost tomorrow,” said Chirag Mehta, vice president and principal analyst at Constellation Research. “Vendors are trying to figure out how to charge for features that they have built, not just to cover the cost of the AI [models]…it’s value-based pricing.” 

Consequently, AI vendors are moving away from flat-fee subscription models to usage-based pricing to cover their costs. They’re also rolling out bespoke AI products. OpenAI, for example, reportedly plans to launch a $20,000 per month AI agent capable of “PhD-level research.”

Cheaper Infrastructure = Higher Customer Costs?

The rollout of DeepSeek’s open source R1 AI model, which reportedly cost just under $6 million to build, initially generated enthusiasm about the potential for new use cases to reach the market more quickly. 

Despite its availability, enterprise pricing won’t necessarily go down, according to Kyle Poyar, co-founder of Boston-based venture capital firm Tremont and author of a SaaS-focused newsletter.

With lower underlying costs, “AI is commoditized. It becomes more like a utility,” he said. “What often happens as costs come down is that consumption rises dramatically, and so spend doesn’t necessarily decrease,” he added.

With a range of AI product offerings on the market, CIOs are likely focusing on ROI and determining the right “price-value exchange,” Poyar said. The degree to which AI tools are developed internally versus through vendors is also a likely preoccupation for CIOs. Open source models may be more costly to maintain than relying exclusively on third-party vendors for AI offerings.

The CIO View

Lonnie Garris, CEO and co-founder of Cool Amps energy solutions and director of information security at Riomar Group, suggests technology leaders shouldn’t consider solutions like DeepSeek just because of lower costs and should consider the security implications of using such a tool. 

When deciding which AI products to buy, companies should carefully consider the ROI; the most expensive offering may not be the best choice, he argues.

“Look at how you can make workflows faster and more streamlined,” he said. “Right now, if OpenAI is offering $20,000 a month for PhD-level [agents], I would just hire a human PhD.”

Others say cost savings from using open source models like DeepSeek don’t always manifest for smaller firms. CIO Jerry Heinz said that based on his client experiences, running DeepSeek’s open-source model through an API may end up costing some companies more than outsourcing the effort to an external vendor.

Though transaction costs might be cheaper through DeepSeek, “if I were to pull it in-house, I would have to buy the hardware [data center server], the racking process, the networking, it’s all of that,” making it pricier than working with a third-party AI vendor.

One concrete use case CIOs might consider, despite variable cost implications, is agents that can handle certain discrete tasks, said Poyar.

“Where I’m seeing the most traction right now is in vendors that are building vertical AI agents that are often highly targeted toward specific workflows, usually in a specific industry,” he said. “They can deliver a strong value proposition.”

The Longer-Term Outlook 

While the rollout of DeepSeek won’t immediately lower costs for enterprise customers, it shows that AI models can be developed at a lower cost, said Mehta. Over time, the entry of additional market participants may help bring clarity to pricing models.

“DeepSeek was one of the inflection points where you can actually do things in a cheaper way,” said Mehta. “Eventually, maybe two or three years from now… the large language model will become a commodity. People will figure out an interesting way to build things on top of that and then price it based on a value-based model.”