VESSL AI Plans a Vera Rubin NVL72 Cluster in H1 2027 and 200,000 GPUs by 2030

VESSL AI, a South Korean neocloud company, told a strategy briefing in Seoul on 15 October 2026 that it plans to open a cluster based on Nvidia’s next-generation Vera Rubin systems in the first half of 2027 and to expand to 200,000 GPUs by 2030, according to Seoul Economic Daily and Edaily.

Those figures are company plans and targets reported by the two outlets, not completed purchases or live capacity TSN has verified independently.

What the company said it will build

Seoul Economic Daily reported that VESSL AI laid out the plan at a neocloud strategy briefing in Gangnam on 15 October. The Vera Rubin-based cluster is slated for the first half of next year. Edaily reported the same timing and named the system as Nvidia’s Vera Rubin NVL72 — a rack-scale design Edaily described as connecting 72 Rubin GPUs and 36 Vera CPUs with high-speed interconnect and HBM4 memory.

Neocloud, as Seoul Economic Daily explained, refers to services that specialise in supplying and operating GPU infrastructure for training and running AI models. VESSL AI presents itself as linking GPU capacity across partners and data centres and allocating it to customer demand.

GPU targets

Both outlets said VESSL AI currently has access to about 7,000 GPUs through partners and plans to buy GPUs directly from next year, reaching 200,000 units by 2030.

Edaily added intermediate targets of 10,000 GPUs by the end of 2026 and 50,000 by 2027. Seoul Economic Daily also said the company is targeting 50,000 GPUs next year and has already confirmed customer demand for 30,000 of them, with 70% to 80% of that demand from overseas companies.

CEO Ahn Jae-man said demand is so strong that GPUs “sell out entirely within one to two weeks of being secured,” according to Seoul Economic Daily. Edaily quoted him making the same one-to-two-week point.

Financing and revenue targets

Seoul Economic Daily reported that VESSL AI is pursuing 400 billion won in GPU financing together with Nvidia and global investment banks in the United States and Singapore. If the plan proceeds as intended, the paper estimated capital spending required through 2028 at a total of 5 trillion to 10 trillion won. That capex range is Seoul Economic Daily’s estimate, not a closed financing announcement.

Ahn said, per Seoul Economic Daily: “After posting 250 billion won in revenue this year, we will grow to more than 2 trillion won the year after next.” Edaily’s headline framing likewise put a 2028 revenue ambition in the multi-trillion-won range via the same briefing. These are management targets reported by the press, not audited results for future years.

Edaily also said the company plans to raise an additional 2 trillion won next year with global investment banks and major domestic securities firms. That, again, is a plan as reported, not a completed raise.

What we are not saying about data-centre megawatts

The two outlets gave sharply different figures for future data-centre capacity (Seoul Economic Daily described 50 megawatts across eight sites by 2028; Edaily referred to a 500 megawatt domestic AI data-centre aim by 2028). Because those numbers conflict, TSN is leaving megawatt capacity out of this account rather than choosing one.

Software pitch

Beyond hardware, Edaily reported that VESSL AI emphasises “Fluid Computing” software to operate distributed GPUs as one environment, claiming lower idle time and fault prediction. CTO Jeon Ji-hwan’s operational statistics in that article are company claims via Edaily; they are not independently audited here. Seoul Economic Daily similarly said the firm differentiates itself by linking scattered GPU infrastructure and matching it to demand.

Why it matters

Vera Rubin is Nvidia’s next major AI systems generation after Blackwell. A Korean neocloud saying it will field NVL72 capacity in the first half of 2027 is part of a wider race to secure next-wave racks, financing and overseas inference customers. The 200,000-GPU and multi-trillion-won revenue targets show how aggressive the forward math has become — and how much still depends on financing, Nvidia supply and actual customer contracts.

TSN has covered Nvidia’s own Vera Rubin rack disclosures and other companies’ Rubin GPU plans separately (see Related stories). VESSL AI’s briefing does not, by itself, confirm that Nvidia has allocated a specific NVL72 shipment schedule to the company.

What’s still unknown

• Whether the 400 billion won GPU financing closes, on what terms, and with which banks.
• Firm Nvidia allocation and delivery dates for Vera Rubin NVL72.
• How much of the “confirmed” demand for 30,000 GPUs is contracted versus indicative.
• Reconciled data-centre power capacity (omitted here because sources disagree).

What to watch

Watch for a financing close, a firm cluster opening date in H1 2027, and any Nvidia or partner confirmation of NVL72 deployments in Korea under VESSL AI’s brand.

What this means

• Plans, not installed base. H1 2027 Rubin cluster and 200,000 GPUs by 2030 are targets reported from a briefing.
• Credit the outlets. GPU, revenue and financing figures come from Seoul Economic Daily and Edaily’s accounts of the company’s event.
• Megawatts omitted. Sedaily and Edaily capacity figures conflict, so they are left out.

Sources

1. Kim Ji-young, “VESSL AI to Open Vera Rubin Cluster Next Year, Targets 2 Trillion Won in 2028 Revenue”, Seoul Economic Daily, 15 October 2026
2. Yun Junghoon, “Vessel AI to Introduce NVIDIA’s ‘Vera Rubin NVL72’ Next Year… Accelerating NeoCloud Business (Comprehensive)”, Edaily, 15 October 2026

Related stories

• NVIDIA Opens Vera Rubin Rack Designs to OCP, Meta Adopts Spectrum-6 and Google’s Falcon Comes to ConnectX-10
• AM Intelligence’s extra 20,000 Nvidia Rubin GPUs: company plans for India and Malaysia