AI Running Costs Fell This Year Even as Server Prices Rose, but Data Center Curbs Test the Trend

The cost of running AI has fallen this year even as the price of renting servers has risen, and some tech executives and investors now worry that data center restrictions could end that run, The Information reports. The concern follows Tuesday's U.S. elections, in which voters elected officials who are pushing to restrict data center expansion.

The cost trend so far

The Information cites OpenAI's GPT-6.1 Sol, released in late September, as an example. With a high reasoning setting, the model can perform comparably to the larger OpenAI flagship GPT-6 Astra. Research firm Artificial Analysis said the newer model costs about a fifth as much per task as the older one. Per-token pricing, context window and benchmark scores for either model were not provided.

According to the report, Anthropic has touted similar efficiencies with its latest models. This has happened even though buying or renting Nvidia hardware has become a lot costlier this year.

Why the trend matters

The Information notes that AI costs are becoming a major concern at enterprises. If the price-drop trend does not persist, the report says, customers might pull back.

A senior executive at a private equity firm that owns and develops data centers told the reporter this week that AI model prices will keep falling because of "token efficiencies," meaning AI doing more work with less compute, driven by hardware and software advances. The executive argued this would offset constraints on data center expansion. The Information points out that the executive is heavily incentivized to say so, given the financial bets the firm has already made.

Cloud providers and markets

On Wednesday, Amazon Web Services CEO Matt Garman said at an event hosted by Blackstone that he is optimistic about the company's data center expansion despite the election results. "We feel like we're in a pretty good spot, and I think others may have some more work to do," he said. He did not say whether he expects server rental prices to rise further. The Information notes that Amazon has faced significant local resistance to some projects and dropped one in Virginia this year.

Stock investors initially appeared to expect higher rental prices, according to the report. They bid up shares of major cloud providers on Wednesday and sold off shares of smaller cloud providers that still need approvals for new facilities. The major providers pared back some of those gains in a selloff on Thursday.

The Information also reports that many facilities due to come online next year already have some or all of the necessary state or local approvals. The electoral backlash may therefore matter most for facilities aiming to launch in 2028 and beyond.

Anthropic's margins

Anthropic's initial public offering prospectus offers another signal, according to the report. The company's gross profit margins have hovered around 85% in the first nine months of the year, excluding the cost of developing the models. The Information says that may give Anthropic slack to lower prices if it needs to boost demand.

What this means

Two cost curves are moving in opposite directions: hardware and server rental prices are rising, while the price of getting work done with a model, measured per task, has fallen. The GPT-6.1 Sol comparison shows that gap can come from model efficiency rather than cheaper compute. Anything that tightens compute supply puts pressure on the hardware side of that equation.

The report suggests the timing of that pressure is not immediate. Facilities due next year largely have approvals in hand, so the exposure sits further out, with projects targeting 2028 and beyond. The optimism of the private equity executive should be read in light of the financial stake the report attributes to them. Anthropic's reported 85% gross margin indicates headroom to cut prices, though the report frames that only as a possibility if demand needs a boost.