Token economics are becoming a defining measure of artificial intelligence infrastructure efficiency.
As AI workloads move beyond training, the economics of producing useful intelligence are starting to shape infrastructure decisions. CoreWeave is building around that shift, expanding beyond GPU compute into networking, storage and software while inference demand grows faster than training. That transition is changing how customers weigh capacity, utilization and the cost of producing tokens at scale, according to Dave Vellante (pictured, left), chief analyst at theCUBE Research.
“Today they’re at 50/50, and they expect to be 10/90 by next year,” Vellante said. “Both curves are growing. If you look at CoreWeave’s numbers, they’re off the charts.”
Vellante spoke with Executive Analyst John Furrier (right) as a part of the keynote analysis at the Fully Connected event, during an exclusive broadcast on theCUBE, SiliconANGLE Media’s livestreaming studio. They discussed how inference growth and system-level optimization are changing the economics of AI infrastructure. (* Disclosure below.)
Token economics shift the infrastructure equation
The shift toward inference puts more emphasis on how efficiently an entire AI system can turn compute into usable output. That favors architectures optimized across silicon, networking, storage and software rather than around a single component. That transition is a move toward system-level performance, Furrier explained.
“If that shift happens, these data centers will look a lot different with the game still the same,” he said. “Pump out as many tokens per watt as possible. Get the intelligence shipping. You got the perfect storm on the supplier side with Dell, ecosystems booming and CoreWeave in pole position.”
The commercial implications extend beyond performance. CoreWeave’s customers are also pushing for shorter contracts, spot pricing and on-demand access, Vellante noted. Those models carry higher prices than long-term take-or-pay agreements, suggesting that flexibility itself is becoming part of the economics surrounding AI capacity.
“The prices for those types of structures are much, much higher, and people are willing to pay,” Vellante said. “It’s going to be really interesting to see as that 98% starts to go down toward 50/50, how that’s going to sort of affect the market.”
CoreWeave builds beyond GPU capacity
GPU availability may bring customers to CoreWeave, but the company is broadening its stack to keep them there. Its strategy now includes networking, storage and a software layer built around observability, security and continuous improvement, giving it more control over the infrastructure surrounding AI workloads, Vellante added.
“What you’re seeing CoreWeave do strategically is they’re expanding out beyond compute,” he said. “They’ve got networking; they’ve got storage. They announced [CoreWeave] Forge today … they’ve got this software layer, which is observability. They’ve got security in there. They’ve got this closed-loop system between evaluation, observation, runtime curating … that’s sort of how they’re saying their software layer is now intact.”
That broader systems approach also reinforces CoreWeave’s relationship with Nvidia Corp. That silicon remains critical, but commercial advantage increasingly depends on how efficiently the full infrastructure stack turns those resources into useful tokens, Furrier explained. That makes hardware availability only one part of a larger performance equation.
“This is becoming increasingly a systems game; it’s a systems race,” he said. “Silicon matters enormously. What’s useful commercially is how quickly the entire system turns into useful tokens. At the end of the day, that’s the key.”
Here’s the complete video interview, part of SiliconANGLE’s and theCUBE’s coverage of the Fully Connected event:
(* Disclosure: TheCUBE is a paid media partner for the Fully Connected event. Sponsors of theCUBE’s event coverage do not have editorial control over content on theCUBE or SiliconANGLE.)