The AI race is raising companies' bills, from GPUs to memory to Microsoft 365

OpenAI, Anthropic and the largest cloud providers are racing to build ever more powerful AI, and some of the leaders are raising outside money to help pay for it. Companies outside that race, including those adopting AI as fast as they can, are also facing higher prices for memory chips, reserved cloud GPUs and software.

Anthropic disclosed approximately $895.4 billion in purchase and lease commitments, substantially all tied to computing infrastructure, in an amended IPO prospectus filed Nov. 2. About $214 billion of that total is subject to termination rights or conditions; the commitments extend through 2045. The company reported $33.2 billion in revenue in the first nine months of 2026. Some agreements require Anthropic to pay for resources "regardless of actual utilization," meaning even for capacity Anthropic does not use. "We must grow our customer base and expand revenue rapidly in order to generate the cash flows necessary to satisfy these commitments," Anthropic wrote.

On Oct. 28, Alphabet reported negative free cash flow for the second quarter in a row, after $51.4 billion of capital spending in the third quarter. The company also sold $25 billion in bonds in August amid its AI infrastructure expansion, Bloomberg reported.

The cost pressure is showing up in corporate budgets. A majority — 56% — of respondents said a software vendor added an AI surcharge, raised a renewal price citing AI, or moved features the organization already had into a higher-priced tier in the past 12 months, according to VentureBeat Intelligence's October AI Infrastructure and Compute survey. And 38% of respondents named a cost overrun as a reason their organization delayed, scaled back or canceled an AI project in the past 12 months. Only 12% named a shortage of computing capacity.

Microsoft 365 list prices rose as much as 43% on July 1

Among respondents who reported at least one of those vendor changes, 40% said the added cost amounted to 10% or more of their annual software spending.

Microsoft raised list prices on select commercial Microsoft 365 subscriptions on July 1, with increases of 5% to 13% for enterprise suites with Teams and as much as 43% for Microsoft 365 F1 without Teams, and justified the increases with new features in AI, security and management. Existing commercial customers on affected plans see the new price at their next renewal after July 1, regardless of how they buy.

Salesforce raised Enterprise and Unlimited edition list prices for Sales Cloud, Service Cloud, Field Service and select Industries Clouds by an average of 6% on Aug. 1, 2025, citing “the significant ongoing innovation and customer value delivered through our products” as it rolled out Agentforce packaging.

Cloud providers' backlogs approach $2.5 trillion

The four providers reported approximately $2.48 trillion in combined backlog in their latest results: Microsoft, $677 billion; Oracle, $664 billion; Amazon, $578 billion, primarily from AWS; and Google Cloud, $561 billion. The companies use different backlog definitions, and the figures include contracts unrelated to AI. Microsoft CFO Amy Hood told analysts in January that OpenAI accounted for about 45% of Microsoft's commercial backlog at the time. The Wall Street Journal reported that OpenAI signed a five-year, $300 billion computing contract with Oracle. Amazon's latest SEC filing also identifies $138 billion in AWS commitments from OpenAI.

Alphabet guided 2026 capital spending to $195 billion to $205 billion, and CFO Anat Ashkenazi expects it to rise “significantly” in 2027.

Memory prices more than doubled in 12 months

Gartner said on Oct. 21 that “over the past 12 months, demand from data center systems has driven memory prices to more than double, with cost increases cascading throughout storage, server, networking, PC and device supply chains.” “This ‘memflation’ effect is expected to persist through 2027 before hardware pricing begins to stabilize,” said Gartner analyst John-David Lovelock.

Hood told investors on April 29 that the company expected roughly $190 billion in calendar 2026 capital expenditures, including approximately $25 billion from the impact of higher component pricing. Microsoft lowered its forecast to about $175 billion in July after a change in how its data center leases are classified, then to about $170 billion in October, citing delays in third-party capacity.

AWS raised reserved GPU prices four times in 2026

AWS raised prices on EC2 Capacity Blocks for ML, its reserved GPU product, by about 15% on Jan. 4, roughly 10% in April for some GPU configurations and about 20% on July 1. It raised rates again on Oct. 7, by approximately 15% for most U.S. GPU configurations. AWS’s pricing page says reservation prices are updated regularly based on supply and demand.

Of the 128 respondents whose organizations operate GPUs or accelerators, 73 put average utilization in production at half or less, 13 put it above half and 42 do not measure it. Among the 114 that reserve capacity in advance, 31% reported that more than a quarter of their AI compute spend goes to capacity they reserved but did not use, and 22% did not know.

Gartner predicts AI inference costs per agentic workflow will increase more than fivefold by 2028

Token prices are headed down, and Gartner expects token costs to fall 95% by 2030. It predicted on Aug. 17 that AI inference costs per agentic workflow will increase more than fivefold through 2028 anyway.

“Product leaders cannot rely on more efficient token economics to rationalize AI costs,” said Will Sommer, senior director analyst at Gartner. “Each successive generation of AI capability will necessitate more, and often more expensive, tokens.”

Soft costs on top

Running AI infrastructure also requires specialized engineering expertise. “Planning GPU and infrastructure is a very complex problem,” said Sudhanshu Bist, vice president of platform infrastructure engineering at Upwork, at VentureBeat’s Agentic Infrastructure Stack Summit in San Francisco on Oct. 28. The newest chips “are not easily available on demand,” so “you have to commit upfront.” Commit for the peak, he said, and that capacity sits idle in the valleys. Asked on stage about the idle-GPU finding, Val Bercovici, chief AI officer at WEKA, called running AI infrastructure a steep engineering problem for companies whose core business is something else. “Either they don’t have the talent or it’s not their business to be experts at running infrastructure for AI, for GPUs,” he said.

Changing AI models can also mean additional engineering work. In VentureBeat Intelligence’s September Agentic Orchestration report, 59% of respondents whose organizations switched AI models in a live application reported that performance dropped and they had to rework the application or switch back.

124 of 156 respondents expect AI budgets to rise

One hundred twenty-four respondents expect their AI budget to increase over the next 12 months, and 29 expect it to grow by more than half. Among those 124, 27% judge whether AI spending is worth it by a measured financial return they could defend to their finance team. Across the whole sample, 40 use a financial return, 69 use a measured operational result such as time saved, and 40 use leadership judgment without a formal measure.

Asked separately about AI infrastructure, 21% of respondents track the return rigorously. Of the 91 that track it rigorously or partially, 34% have a figure their finance organization has reviewed and signed off.

The AI bill by line

What buyers need to do about it

• Check Microsoft 365 renewals. Confirm which plans are affected before your next renewal. Higher list prices apply to affected subscriptions at their first renewal after July 1.

• Count the vendor changes. Track AI surcharges, renewal increases and features moved to higher-priced tiers, one line per vendor.

• Measure utilization before reserving more. Of the 86 respondents who reported a GPU utilization figure, 20 said they use continuous telemetry across their entire fleet.

• Check AWS Capacity Block rates before buying. AWS updates reservation prices regularly and has increased them four times this year.

• Budget agents per successful task. Gartner says comparing token prices alone misses two important cost drivers: token consumption and task success rates.

• Put a finance-reviewed return on the increase. Of the 91 respondents who track infrastructure returns rigorously or partially, only 34% have a figure finance has signed off on.

VentureBeat Intelligence surveyed people at organizations with 100 or more employees, and consultants serving such organizations, in October 2026. The survey received 172 responses, of which 156 qualified. Of the qualified respondents, 87% came through a research panel and 10% are consultants answering about their main client. Respondents chose to take the survey, so the figures describe those respondents only. The August and July editions of the same series are published on VentureBeat.