A new breed of providers known as "neoclouds"—specialized in renting out GPU compute—is gaining ground fast, with players ranging from Groq to footwear maker Allbirds entering the field, even as OpenAI pushes to build its own chips, intensifying competition over AI compute supply.
AI Infrastructure · The Neocloud Surge
The Rise of "Neoclouds": Renting AI Compute Instead of Owning It
Amid chronic GPU scarcity, AI-specialized providers — and chip players like Groq and SpaceX/xAI — are racing to lease out compute, undercutting hyperscalers on price, speed, and immediacy.
$2.99
Neocloud H100, per GPU-hour on demand — often under half the hyperscaler rate
3.5M+
Developers on GroqCloud's LPU-powered inference services
$20B
NVIDIA licensing / tech-transfer deal that pushed Groq into the neocloud business
SpaceX / xAI Compute Leases — Monthly Cost
Surplus Colossus capacity rented externally, monetized ahead of an IPO
$1.25B
→ AnthropicColossus 1 (220,000+ GPUs)
$920M
→ Google~110,000 GPU equiv.
$150M
→ Reflection AIColossus 2 GB300 (up to $6.3B)
Column heights proportional to monthly lease cost
Why builders are switching
Cheaper, faster GPU provisioning vs AWS/Azure/Google Cloud
GroqCloud praised for dramatic inference speed & cost cuts
CoreWeave / Lambda valued for reliability, Kubernetes, Slurm & Jupyter support
The constraints
Shifting model catalogs & free-tier rate limits
Speed varies with context length
Stockouts amid excess demand
Usage-proportional cost growth instead of fixed capital
The shift
From owning compute to renting it — a "new generation of AI infrastructure."
Bare-metal
→
GPUaaS
→
Managed / inference-focused offerings
Eases the capital burden on AI startups — but how long demand lasts hinges on the pace of the broader AI market's expansion.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article…