AI companies pay for computing power to generate answers, images, and code. Now you can own a share of it.
Powerful AI servers cost hundreds of thousands of dollars. Fractional GPU makes their computing power available in smaller shares that work for you.
Select a server to see how its capacity is shared
The selected server's capacity is shared among multiple owners. Your share is highlighted in violet
Example allocations only. Names do not indicate Fractual ownership or endorsement
You choose how your share is used. Professionals handle the computers, cooling, and day-to-day operations.
Lease and mining calculations use first-pool reference terms, not confirmed current opportunities. Availability, accepted terms and costs depend on agreements. Lease proceeds may be lower or zero; mining rewards are not fixed cash payments.
Personal use and transfers are subject to the terms of the applicable agreement.
Get early access to Fractual’s Fractional GPUs. Leave your email to request a reservation and learn about availability.
A Fractional GPU is an asset representing a share of computing capacity from pooled AI servers.
Enter your email to request access to Fractional GPUs. You’ll review available terms before any purchase.
New chips arrived. The older asset kept earning.
Adapted from a16z, State of Markets II / Silicon Data (September 7, 2026). Each series begins at 100 for its first observed rental rate; chart approximately reproduces the published index. Rental rates are not resale prices or guaranteed future income.
How to read the scenarios: A conditional median is the midpoint of modeled outcomes within a particular scenario (supply-shock or low-range). The full-mixture median is the midpoint across the entire modeled distribution. The chart does not show the probability of each scenario.
Barkr model: 100,000 simulated paths over 10 years; each series begins at 100% of initial system value. Source: Barkr, The Resale Standard (September 29, 2026), Figure 3, GB300 NVL72 (Blackwell Ultra); curves rebased to 100%, intermediate values approximate. Estimated useful life cited by NVIDIA (October 1, 2026). These are modeled system-value scenarios, not fixed depreciation or guaranteed resale prices. They are not B200 GPU, Fractional GPU, or operating-income valuations.
Answers about ownership, operations, costs, and what happens next.
A Fractional GPU is an asset representing contractual rights to a share of computing capacity from pooled AI servers, rather than direct ownership of a particular physical GPU. It remains yours while you hold it, and you choose how to use your capacity under the applicable agreements.
Every AI answer, image, or generated piece of code requires computing work. AI companies may rent powerful computer capacity to deliver those services to their customers. The demand for AI computing infrastructure is growing, although rental prices and utilization can change.
Selected 2026 market indicators include U.S. AI investment estimated at 1.8% of GDP, expected AI-infrastructure capital-spending growth of 75%, and a NVIDIA B200 rental-price index that rose 24.5% from January through March 2026. These indicators use different methods and periods. They describe market conditions - not future GPU prices, utilization, or earnings. Sources cited in the underlying market research: Goldman Sachs Research (August 2026), Brookings (October 2026), IEA (2026), NVIDIA, and Silicon Data.
GPU purchases can be affected by limited hardware allocations, distributor priority, buyer eligibility, and export restrictions. Purchasing and deployment depend on actual availability, distributor terms, and applicable rules; no future access window is guaranteed.
Datacenter providers and professional system administrators manage installation, electricity, cooling, networking, monitoring, maintenance, and technical operations. You choose among the uses available under your agreement.
AI companies post requests with a price and duration. You decide whether to participate. A lease starts only when enough capacity has been committed to fill the request. Proceeds are allocated among participating owners in proportion to their contribution.
Your asset remains yours. You can continue the same use if available, choose another option, use it yourself, or seek to transfer it under the agreement. Existing commitments may restrict when changes take effect.
The first-pool economics use an eight-NVIDIA-B200 reference server priced at $718,750. It shows gross annual proceeds of $266,304 from the first-pool lease example and $420,480 from the mining example, with $54,750 in annual hosting and managed-operations charges in each case. Choose one example at a time to see how those per-server figures scale with the amount selected, including the calculation breakdown. The lease and mining calculations use first-pool reference terms; current availability is not confirmed.
No. These figures are calculated from first-pool reference terms, not promised cash flow or returns. The lease scenario depends on an available contract, capacity commitments, usage, and costs. Mining rewards are paid in GNK, and their USD value changes with the token price. The mining example annualizes September 2026 observations and assumes 8,760 earning hours per GPU yearly. Taxes, financing, depreciation, other fees, and resale are excluded.
Datacenter space, power, networking, maintenance, and management software are shared in proportion to your share. The first-pool model includes $54,750 per year for each eight-B200 reference server in hosting and managed operations. The lease and mining example calculations already deduct that operating cost. Personal-use access fees depend on the applicable agreement.
Yes, if the option is available for your pool. The first-pool example uses Gonka, with a 24-hour network term. Network rewards are not fixed; the value of GNK can increase or decrease.
Yes, under the applicable scheduling and access terms. Personal use gives you computing resources for your own workloads, not lease or mining proceeds.
You can request a change of datacenter provider, subject to the service and use agreements. You may also request temporary disconnection, which can pause future datacenter charges once effective. Timing, notice, and existing commitments matter.
You can arrange a direct resale or transfer of your Fractional GPU at any time, subject to the terms and restrictions of your agreement, existing use commitments, applicable regulations, and manufacturer restrictions. A secondary-market resale platform is planned for the future. Finding a buyer, the sale price, timing, and liquidity are not guaranteed.
Hardware can remain commercially useful even as new generations arrive, but the resale value and rental economics of older hardware can fall. See the Value over time section for historical rental-price trends and modeled hardware-value scenarios. Neither predicts your future proceeds or asset value.
Fractual is an initiative by David and Daniil Liberman, technology entrepreneurs working on new forms of ownership and access to AI infrastructure.
Choose an amount, select among the use options available for the relevant pool, review the final agreement, and confirm the purchase. The order remains subject to hardware supply, distributor terms, buyer eligibility, applicable export controls, and law. This static page is a design prototype; it does not accept orders.