Fuse Energy has confirmed its Fuse Compute AI cloud will bring Nvidia Blackwell Ultra GPUs online in the UK from December 2026, while Georgia's Silk Cloud readies a separate deployment for 2027. For UK buyers chasing GPU accelerators, the real question isn't just availability — it's how fast power and pricing actually scale.
View the data behind this chart
| Capacity (MW) | Dec 2026 | Q2 2027 | Q4 2027 | 2028+ |
|---|---|---|---|---|
| Fuse Compute capacity | 0.7 | 3.25 | 22 | 116 |
What's actually live, and what isn't
Two separate Blackwell Ultra stories broke this week, and UK buyers need to keep them apart. Silk Cloud, a Georgian cloud provider owned by Silk Road Group, plans to deploy HPE Nvidia HGX B300 systems at a Tbilisi data centre, but that launch is pegged to 2027 and is waiting on 'necessary permits'. No GPU count has been disclosed for that site.
The UK-relevant news is Fuse Energy's Fuse Compute offering, which is further along. Fuse says its first 128 Nvidia GPUs go online in December 2026, rising to 1,024 by Q2 2027. Like Silk Cloud, Fuse has opted for B300 hardware — the same Blackwell Ultra class chip, not a cut-down variant.
Fuse Energy's power-first pitch to GPU buyers
Fuse Energy describes itself as a full-stack energy company, and its entry into compute leans heavily on that identity. In a LinkedIn post announcing Fuse Compute, the firm argued: "The hardest part of scaling compute today is getting power to the site, which is the part we already do."
That's a credible differentiator in a market where grid connection delays, not chip supply, are often the real bottleneck for UK GPU capacity. Fuse cites 142MW of data centre connections across 15 projects in its pipeline — infrastructure it says it can now layer compute onto rather than build from scratch. Buyers evaluating providers should treat this power-led model as genuinely different from a traditional colo-plus-GPU deal, and worth scrutinising on hosting Blackwell AI racks grounds: power availability and cooling design, not just chip allocation, will decide who actually goes live on schedule.
The capacity ramp: kilowatts today, 116MW by 2028
Fuse's own published figures show a steep curve: 700kW live in December 2026, 3.25MW by Q2 2027, jumping to 22MW by Q4 2027, and 116MW from 2028 onward. That trajectory matters more to procurement teams than the initial headline figure, because it signals when meaningful, multi-megawatt capacity — not a pilot rack — actually becomes available.
Buyers sizing workloads should map their own GPU demand curve against this schedule rather than assuming December availability equals production-scale capacity. Our AI GPU calculator is a useful starting point for working out how many Blackwell Ultra-class units a given training or inference workload actually needs before committing to an early allocation.
Pricing and operator detail: still a blank page
Neither Silk Cloud nor Fuse Energy has published pricing for their Blackwell Ultra capacity, and Fuse has not named which data centres or operators will host its GPUs. For UK buyers used to transparent list pricing from established cloud providers, this is a real gap. Early capacity from power-led entrants often comes with bespoke commercial terms rather than standard on-demand rates.
That makes due diligence on contract structure essential before signing anything tied to the December or Q2 2027 milestones. Anyone weighing a capital purchase against a committed-capacity contract should also compare the underlying economics using server finance options, since the total cost of ownership on owned Blackwell Ultra hardware can look very different once power and cooling overheads are factored in.
Blackwell Ultra now, or wait for the next generation?
Both deployments confirm that B300-class Blackwell Ultra hardware is becoming the default choice for new entrants building AI cloud capacity outside the US hyperscalers. For UK buyers already tracking the Nvidia GPU roadmap, this is further evidence that Blackwell Ultra supply is widening even as attention starts to shift toward Nvidia's next architecture — a timing question worth working through properly rather than guessing.
- •Treat Fuse's December 2026 date as a pilot-scale milestone, not full production capacity
- •Ask any provider offering early Blackwell Ultra access to name the operating data centre and power source
- •Model your own GPU demand against the published capacity ramp, not the headline launch date
- •Compare committed-capacity contracts against outright purchase or finance routes before signing
- 01DataCenterDynamics — Georgia's Silk Cloud deploys Nvidia Blackwell Ultra GPUs · 24 October 2026