Neocloud operators: from niche upstarts to AI infrastructure backbone
Neocloud operators are specialized AI infrastructure providers that scale shared GPU fleets for multiple customers through contracted access, aiming to deliver enterprise AI compute on demand without owning every layer of traditional cloud platforms, and they now sit at the center of a race to deploy capacity fast enough to keep up with generative AI workloads and long-term inference growth. At the heart of this race is Sharon AI, whose contracted value has climbed to about $8.8 billion year-to-date while it targets a GPU fleet deployment of more than 64,000 NVIDIA units by mid-2027. That headline number is not just a bragging point; it signals that the balance of power in AI infrastructure scaling is shifting toward operators that treat GPUs like long-lived industrial assets rather than short-term cloud inventory.
Why demand is outpacing supply: AI factories as long-term bets
The core reason neocloud operators are on a GPU buying spree is simple: demand for high-density AI infrastructure scaling is rising faster than traditional cloud can add supply. Sharon AI says its deployment pipeline is expanding as demand for high-density AI computing capacity exceeds available supply. Naver’s CEO frames the same trend in more strategic terms, calling its AI factory project “an investment in a structurally growing market” driven by generative AI, inference, and AI agents over the mid to long term. Naver expects its AI factory with NVIDIA to start generating revenue in the first half of 2027, launching with 55 megawatts of capacity and scaling to 100 MW by year-end and 200 MW in 2028. This kind of staged ramp shows that operators now think in multi-year power and capacity targets, not quarterly server shipments.
| AI Factory Plan | Initial Capacity | Target Expansion |
|---|---|---|
| Naver–NVIDIA AI factory launch | 55 MW in 2027 | 100 MW by end-2027, 200 MW in 2028 |
| Sharon AI GPU fleet | Growing from current base | More than 64,000 GPUs by mid-2027 |
Take-or-pay contracts: the financial engine behind GPU fleet deployment
The real innovation behind neocloud operators is not only technical; it is financial. Sharon AI’s AI infrastructure scaling strategy depends on turning massive GPU investments into predictable cash flows through take-or-pay contracts. Under this model, customers pay monthly for reserved capacity once systems are deployed, whether or not they use the full amount. That transforms GPUs into quasi-utility assets with bond-like revenue visibility instead of speculative bets. A six-year master services agreement with NVIDIA provides access to up to 40,000 GB300 GPUs and helps reduce hardware rollout risk while leaving room to sell reserved capacity to other customers. Sharon AI further argues that GPUs retain at least two years of useful economic life after typical contracts expire, opening options to re-contract, sell on-demand capacity, or pivot hardware toward inference workloads. In effect, fleet operators are building layered revenue stacks on the same silicon.
From early-stage ramp to AI factory maturity
For all the bold numbers, this AI factory buildout is still early-stage, and the income statements show it. Sharon AI’s revenue remains modest relative to its contracted pipeline, with financials reflecting an infrastructure ramp rather than a steady-state utility. Naver’s leadership echoes this pattern, noting that margins in its AI factory business are likely to start low but are expected to reach double digits and eventually exceed 20 percent as the business matures. Both operators are betting that once their GPU fleet deployment and power footprints are in place, utilization will rise, cost per unit of compute will fall, and more enterprise AI compute workloads will move onto their platforms. Crucially, they assume cheaper and more efficient AI models will increase, not reduce, overall compute demand as usage expands. If they are right, today’s thin margins are the price of gaining durable control over tomorrow’s AI infrastructure.
The emerging playbook for AI infrastructure platforms
What Sharon AI and Naver show is an emerging playbook that others will copy. First, lock in long-term GPU access through strategic collaborations that make hardware risk manageable. Second, secure power and data center capacity in advance, as Naver has done by lining up sites, capital partners, and operational capabilities to run the full AI stack. Third, design contract structures that guarantee payment for reserved capacity and extend visibility years into the future. This playbook is opinionated: it assumes AI infrastructure is closer to an electricity grid than a conventional cloud service. The operators willing to embrace that view, treat GPUs as multi-cycle industrial assets, and anchor their business on enterprise AI compute commitments will define how and where the next generation of AI factories are built. Those that hesitate risk watching the AI infrastructure backbone being written without them.
