Every GPU.
On demand.
H100, H200, B200 and RTX 5090 by the hour or as dedicated clusters. Capacity from Tier III datacenters across four regions, one account and one invoice.
H100
Hopper · SXM5From a single card to a full cluster.
On-demand rates per GPU-hour. Reserve a cluster for 3 months or longer and the rate drops.
Strata is one account for GPU capacity across many datacenters. We reserve compute from Tier III operators and give it to you by the minute, by the month or as a dedicated cluster.
You run applications and models. We take care of the layer underneath: finding capacity, keeping it available and moving you to newer GPUs when they arrive.
Pick the job. We'll match the GPU.
Starting points for common workloads. Our team will size the cluster with you.
Pretraining and large runs
Multi-node clusters with NVLink and fast interconnect for long training jobs.
Adapt open models
LoRA and full fine-tunes of Llama, Qwen, DeepSeek and Mistral models.
Serve models in production
Steady capacity for LLM APIs, embeddings and agents, scaled to traffic.
Image and video models
Flux, SDXL, Wan and Hunyuan Video pipelines with fast turnaround.
Commit longer, pay less.
Dedicated GPUs allocated to you alone for the whole term. Lock today's rate and stop worrying about availability.
Indicative quote. Final pricing depends on region and availability.
| Term | Discount | Payment | Best for |
|---|---|---|---|
| On demand | — | Per minute from balance | Experiments, bursts |
| 3 months | −6% | Monthly in advance | Projects with a deadline |
| 6 months | −12% | Monthly in advance | Steady inference |
| 12 months | −20% | Quarterly in advance | Training roadmaps |
| 24–36 months | −28% | Quarterly, 15% deposit | Production platforms |
Run close to your users.
Pay where it's cheaper.
Pick a region for latency or data residency, or let us place your workload where capacity is available and priced best.
Don't buy hardware that ages.
A GPU bought today is a two-year-old GPU in two years, and its market value swings. Renting keeps you on current hardware and turns capex into a monthly bill.
H100 rental price, $ per GPU-hour
Owners carried this swing. Renters locked rates when it suited them. Sources: Silicon Data, SemiAnalysis.
Buying servers
- $250–300k up front per 8×H100 server
- Weeks of procurement, racking and setup
- Value falls with every new generation
- Idle hardware still costs power and space
Renting from Strata
- Start in minutes, scale up or down any time
- Pay only for the minutes or months you use
- Move to new GPU generations as they arrive
- No datacenter, power or hardware to manage
From request to running cluster.
On-demand GPUs start in minutes. Dedicated clusters are allocated within days.
Pick GPUs and region
On demand from your dashboard, or tell us the size and term of the cluster you need.
Top up or sign a term
Card, bank transfer or stablecoins. Reserved clusters get a fixed monthly invoice.
Connect and build
SSH, Jupyter or your own images. Our engineers help with the first deployment.
From our blog.
Questions buyers ask first.
Whose hardware will I run on?
Servers in Tier III or better datacenters run by our operator partners. For every reservation we tell you the region and the datacenter tier.
How fast can I get a cluster?
Single on-demand GPUs start within minutes. Dedicated clusters usually take from a few days to two weeks, depending on size and region.
How is billing done?
On demand is billed per minute from a prepaid balance. Reserved clusters get a fixed invoice monthly or quarterly in advance.
Which payment methods do you accept?
Card, bank transfer, USDT and USDC. Invoices are issued for every payment.
What happens if a node fails?
We replace it from the same pool and credit the downtime. Reserved clusters come with an uptime commitment in the contract.
Is there a minimum commitment?
None for on demand beyond the first top-up. Reserved clusters start at one 8-GPU node for three months.
Tell us what you need to run.
GPU type, cluster size and start date. We'll come back with available capacity and a price.
Three ways to back
the compute behind AI.
Strata is raising capital to contract GPU capacity, finish the platform and build the sales team. Choose how you want to take part: a share of the company, a position in the venture round, or a prepaid block of GPU hours.
Pick the instrument that fits you.
Equity and venture fund the company. Capacity partnerships fund specific wholesale contracts and pay back in compute.
Equity
Own a share of Strata Compute and grow with the company.
Venture
Lead or join the round with institutional rights and reporting.
Capacity partners
Prepay once and receive a locked volume of GPU hours at a discount.
Demand for compute keeps outrunning supply.
Public figures from the companies and analysts that track GPU rental. Each number links to its source.
We contract datacenter capacity wholesale on fixed terms and resell it by the minute and by the cluster. The spread between wholesale and retail rates, at high utilization, is the business.
Owners carry depreciation and price swings. Contracts end on a fixed date and roll into newer GPUs, so capital goes into capacity customers are already asking for.
Platform launch
Self-serve rental, per-minute billing, automatic provisioning.
First wholesale contracts
Reserved H100 and H200 blocks with two or more operators.
Capacity partner program
Prepaid GPU-hour agreements with companies and AI teams.
Multi-region scale
Contracts across Europe, North America and the Middle East.
A share of the company that sells the capacity.
For angels, operators and strategic investors who want long-term exposure to AI compute without owning hardware.
Shares in Strata Compute, or a SAFE that converts into shares at the next priced round.
The platform, the first wholesale capacity contracts and the sales team. See the use of funds below.
Through growth in the company's value, realised at a later round, acquisition or dividend. Nothing is paid out on a schedule.
Quarterly update: revenue, gross margin, utilization, capacity under contract, cash runway and progress against the roadmap.
Datacenter operators, AI companies and infrastructure investors can combine equity with a capacity or supply agreement.
Shared in the data room after eligibility checks.
For funds that lead or join the round.
We look for funds with infrastructure, AI or marketplace experience who can help with operator relationships and enterprise sales.
Pre-seed round with a lead investor setting terms. Co-investors join on the same terms.
Pro-rata in future rounds, information rights and a board seat for the lead.
Monthly KPI dashboard and quarterly board meetings.
GPU-hours sold, revenue, gross margin per GPU-hour, utilization of contracted capacity, customer count and concentration, capacity partner prepayments.
Financial model, operator quotes, customer pipeline, cap table and company documents.
From first call to term sheet in two to four weeks.
Put money into capacity. Get capacity back.
You prepay for a block of GPU time. We use the prepayment to sign a wholesale contract, and you receive a fixed volume of GPU-hours at a locked, discounted rate for the whole term.
Discount = volume tier (10% from $10k, 15% from $50k, 20% from $200k) plus term bonus (0% for 6 months, 3% for 12, 6% for 24).
Run your own workloads, give hours to your teams or portfolio companies, or switch between GPU types at their relative value.
A capacity partnership is a prepaid service agreement. You receive compute, not cash returns, and it is not an investment in Strata's shares.
Your prepayment is matched to a wholesale contract of the same length, so the hours you bought exist on our side for the whole term.
Hours roll over month to month within the term. Terms for transfer and extension are set in the agreement.
Side by side.
Different instruments, different risks. Many partners combine more than one.
| Equity | Venture | Capacity partner | |
|---|---|---|---|
| Who it's for | Angels, operators, strategics | Venture funds | Companies and AI teams that use GPUs |
| What you hold | Shares or a SAFE | Shares with institutional rights | A prepaid block of GPU-hours |
| What you get back | Growth in company value | Growth in company value | Compute at a locked discounted rate |
| Horizon | 5–7 years | 5–7 years | 6, 12 or 24 months |
| Liquidity | At exit or later rounds | At exit or later rounds | Used as you go |
| Main risk | The company does not grow | The company does not grow | Strata cannot deliver the hours |
| Eligibility | Eligible investors only | Professional investors | Any business customer |
Most of the money goes into capacity.
Planned allocation of the equity round. Capacity partner prepayments are separate and go only into matching wholesale contracts.
A live self-serve platform, wholesale contracts with at least two operators, the first capacity partners and a repeatable sales process for reserved clusters.
Each quarter investors see spend by category against this plan.
One cluster, one number that matters.
A hypothetical 64-GPU cluster on a 12-month wholesale contract. Move the inputs to see how utilization turns into gross profit or loss. An illustration of the mechanics, not a forecast.
Break-even utilization at these prices: —
What can go wrong.
Investing in an early-stage company can lose you all of your money. Read these before anything else.
Wholesale contracts are paid in full whether or not capacity is sold.
GPU rental prices can fall sharply, as they did by about 40% from mid-2024 to late 2025.
A datacenter partner can fail, go offline or breach its contract.
Large clouds and marketplaces can undercut prices or lock up supply.
New GPU generations can reduce demand for contracted hardware.
Shares in a private company are hard to sell and may never pay out.
Strata has a short operating history and a small team.
Rules on investment, sanctions, export controls and data can change.
From first call to closing.
Eligibility is checked before any terms or financials are shared.
- Enquiry
Tell us who you are and which path interests you.
- Intro call
30 minutes with the founder on the model and the round.
- Eligibility and KYC
Investor classification, identity and source of funds.
- Data room
Model, operator quotes, pipeline and documents under NDA.
- Documents and closing
Agreements prepared with legal counsel; funds and first report.