Enterprise

CoreWeave Signs AI Compute Contracts at $40M Per Megawatt

The GPU cloud provider is winning customers at higher prices despite widening losses and a $10 billion debt raise.

Omega Editorial· September 19, 2026· 3 min read

CoreWeave raises prices as AI compute demand stays tight

CoreWeave disclosed on September 17 that contracts signed since June 30 are generating approximately $40 million in annualized revenue per megawatt of power capacity. The AI infrastructure provider is securing these deals on short three-to-six-month terms, a structure that allows frequent repricing as long as compute scarcity persists.

The company's revenue backlog stood at roughly $104 billion as of June 30, according to details first reported by AI Watch. That figure excludes more than $25 billion in additional commitments added early in the third quarter. Contracted power capacity climbed to about 4.2 gigawatts by the August 11 earnings call, up from 3.7 gigawatts at quarter-end.

Second-quarter revenue reached $2.575 billion, more than double the $1.212 billion recorded a year earlier. CoreWeave's customer base is expanding beyond AI research labs to include enterprises such as Caterpillar and Bentley Systems.

Why it matters

CoreWeave's pricing trajectory offers a real-time measure of how much organizations will pay to secure scarce GPU capacity. The $40 million-per-megawatt benchmark and the willingness to sign short-duration contracts both signal that supply constraints remain binding across the AI infrastructure market. For CIOs and infrastructure planners, these data points clarify the premium attached to guaranteed compute access—and the risk that pricing could shift quickly if capacity loosens.

Losses widen as infrastructure costs outpace revenue growth

Higher contract values have not yet translated into profitability. CoreWeave posted an operating loss of $49 million in the second quarter, reversing a $19 million operating profit from the prior year. Adjusted operating margin fell to 5 percent from 16 percent, indicating that costs are rising faster than revenue even after adjustments.

Net interest expense more than doubled to $640 million, contributing to a net loss of $626 million compared with $290 million a year earlier. The company raised over $10 billion through convertible bonds and unsecured debt during the second quarter to fund its buildout.

Active power capacity stood at 1.5 gigawatts against 3.7 gigawatts under contract as of June 30, meaning CoreWeave must deploy significant additional infrastructure to fulfill its backlog. The short contract durations add execution risk, since the $40 million rate must hold through each renewal cycle.

Hedge funds increase positions while short interest climbs

Hedge fund ownership of CoreWeave rose to 71 funds from 63 in the prior quarter. At the same time, short interest reached 17.84 percent of the float, reflecting skepticism about the company's path to profitability despite strong demand signals. Some portion of that short position likely represents hedging activity rather than outright bearish bets.

The combination of rising institutional ownership and elevated short interest creates the potential for sharp price moves in either direction as CoreWeave reports progress—or setbacks—in converting its backlog into operating cash flow.

These details were first reported by AI Watch.

#coreweave#ai infrastructure#gpu cloud#data center capacity#ai compute pricing#enterprise ai

This is an original analysis by the Omega editorial team. Source reporting: AI Watch.

Want systems like this working for your business?

Book a Call

More in Enterprise

Enterprise· 3 min read

Workday Scales AI Agents with Iceberg and Unity Catalog

The enterprise software giant solved the multiplicative complexity problem of agentic AI by building a centralized, governed data layer that brings agents to data instead of replicating data to agents.

Via AI Watch · Sep 18, 2026
Enterprise· 3 min read

Disney Names Character.AI's Former CEO as First CTO

Karandeep Anand joins despite legal dispute last year over unauthorized use of Disney characters.

Via AI Watch · Sep 18, 2026
Enterprise· 3 min read

U.S. carriers split on GPU role in AI-RAN deployments

Verizon, T-Mobile and AT&T are testing AI-powered radio access networks with sharply different hardware strategies.

Via AI Watch · Sep 18, 2026