70% of GPU capacity sits idle. packet.ai raised $19M from Creandum to fix that — and we're just getting started.
packet.ai has raised a $19M seed round led by Creandum to solve GPU infrastructure’s core problem: AI workloads use only around 30% of GPU capacity on average, meaning 70% of what service providers invest in — and customers pay for — sits idle.
Key takeaways
The announcement was made on 19 March 2026. The round was led by Creandum — the European VC behind Spotify and Klarna — with Repeat VC following, and participation from existing investors People Ventures, Z21 Ventures, Golden Sparrow, Hersir Ventures and Tekton.
The conventional narrative is that GPUs are scarce. That’s not wrong — but it misses the bigger picture. AI workloads consume only around 30% of GPU capacity on average. That means roughly 70% of the GPUs that service providers invest in — and customers pay for — sits idle.
This creates three compounding problems for the industry. Service providers must make enormous upfront CAPEX investments to meet peak demand, making profitability a persistent challenge. Customers pay for peak capacity they rarely use. And the resulting high prices restrict who can access GPU compute at all.
packet.ai CEO Ditlev Bredahl put it directly: “The GPU market has a waste problem, not a scarcity problem. We’ve spent 25 years building infrastructure software that makes service providers competitive — and the GPU opportunity is the biggest we’ve seen.”
The core insight
Unlike traditional cloud compute, which scales dynamically to match demand, GPUs are static: customers must rent fixed instances based on estimated peak workload requirements. hosted·ai’s software changes that equation without touching a single piece of hardware.
hosted·ai
Core platform delivering GPU pooling, optimised multi-tenant workload placement and overcommit. Up to 5× improvement in GPU utilisation — meaning 5× reduction in CAPEX requirements for service providers.
packet.ai
Our own neocloud, running on hosted·ai-optimised infrastructure. Delivers GPU compute at market-leading prices — H200 SXM from $2.25/hr, B200 SXM from $3.75/hr — 50%+ below hyperscale rates.
GPUaaS.com
Wholesale marketplace connecting enterprise buyers with hosted·ai customers and partners for custom GPU cluster requirements at scale.
Next on the roadmap: GPU Mesh — a resource exchange letting service providers buy and sell spare GPU capacity without additional hardware CAPEX.
As AI shifts from model training to inference, the market is changing in packet.ai’s favour. Inference workloads are bursty, latency-sensitive, and geographically distributed — exactly the conditions where static GPU allocation fails hardest and intelligent pooling wins.
Companies increasingly need local, low-latency, sovereign GPU infrastructure. The hyperscalers can’t be everywhere. Regional service providers can — but only if the economics work. hosted·ai’s software makes those economics work, turning 30% utilisation into something closer to the 85–95% that production inference clusters achieve on packet.ai today.
packet.ai’s approach delivers B200 SXM on-demand from $3.75/hr — the lowest confirmed rate across 26 tracked providers as of June 2026 — precisely because the underlying infrastructure runs at 5× higher utilisation than the industry average.
This round lets the team move faster: more platform capabilities, more partner regions, and continued investment in keeping packet.ai’s GPU prices the lowest on the market. The goal is to build the operating system for the GPU economy — the layer that makes regional, sovereign AI infrastructure viable for the thousands of service providers who can’t compete with hyperscaler CAPEX today.
If you’re an AI team evaluating GPU cloud options, the practical implication is straightforward: browse available GPU clusters on packet.ai, or see pricing on the H200 and B200 pages.
Last reviewed: 10 June 2026. Browse available GPU clusters on packet.ai →
Same models. Same API. Fraction of the cost. Start free — no credit card required.
Start Building →
