While hyperscalers battle for data center supremacy, a startup just raised $30M to make the fragmented mess of 200+ smaller clouds actually usable.

The Summary

  • Hyperbolic raised $30M led by Intel CEO Lip-Bu Tan's VC firm to build infrastructure unifying hundreds of AI neoclouds into one marketplace
  • The startup creates abstraction software so companies can rent compute across different providers without managing each separately, plus lets customers resell unused capacity
  • Hyperbolic now connects $200M in annual compute value across 30 neoclouds, targeting $150M revenue as GPU shortage forces buyers beyond AWS/Azure/GCP

The Signal

The neocloud explosion created a paradox. Companies desperate for GPUs can now choose from 200+ providers, each with different chips, networking stacks, and APIs. More supply somehow made procurement harder. Hyperbolic's pitch is dead simple: one interface for all of them.

Founder Jasper Zhang, 31, spent four years watching AI shops waste engineering hours duct-taping together compute from WhiteFiber, Vector Core Compute, and dozens of others. Each neocloud speaks a different dialect. Most run custom networking. Some use Nvidia chips, others AMD or custom silicon. The customer experience, Zhang says, has been "really painful."

"It's been a really fragmented market that makes the customer experience really painful."

Hyperbolic's software sits between buyers and the neocloud zoo. Companies specify what they need, Hyperbolic finds capacity across its network of 30 providers, and abstracts away the differences. Train a model on Provider A's H100s today, run inference on Provider B's MI300s tomorrow, same API. The company already routes $200M in annual compute spend.

But the real angle is the resale market. Hyperbolic wants customers to flip unused capacity they've already paid for. Cloud contracts lock in minimums. Workloads spike and crash. Right now, that unused compute just bleeds money. Hyperbolic's building a secondary market where your idle GPUs become someone else's training run. It's Airbnb for AI infrastructure, except the hosts are enterprises with lumpy workloads and the guests are startups who'll take compute however they can get it.

Key dynamics at play:

  • Neocloud providers get aggregated demand without building their own sales teams
  • Buyers get optionality and price competition without integration hell
  • Intel's CEO backing this is a tell: chipmakers need software to make their non-Nvidia silicon viable at scale

The funding round pulled in Dell Technologies Capital, General Catalyst, and Vista Equity. That's hardware, infrastructure, and enterprise software money converging on the same thesis: the hyperscaler oligopoly is cracking, but someone needs to make the alternative actually work. Hyperbolic raised $12M in 2024, then $7M before that. Now $30M more. Scaling from 30 clouds to hundreds while building a functioning spot market for GPU time.

The Implication

If Hyperbolic works, it proves the neocloud thesis without requiring any single neocloud to win. The market doesn't need a fifth hyperscaler. It needs infrastructure that makes 200 small clouds act like one big one. Watch whether the resale marketplace actually ships and gets adoption. That's where this goes from middleware play to market maker. If enterprises start trading GPU capacity like bandwidth contracts, the power dynamic in AI infrastructure shifts permanently. The hyperscalers priced on scarcity. Hyperbolic is pricing on liquidity.

Sources

Business Insider Tech