Voltage Park Is Now Lightning AI: GPU Pricing in 2026

Voltage Park's H100 rate hasn't moved. That's the headline most people missed after the January news cycle moved on. The merger with Lightning AI closed on January 21, 2026, and the combined company now trades under the Lightning AI name, but the pricing page you'd actually buy from still lists H100 SXM at $1.99/hr per GPU, no contract, same as before the deal (Voltage Park pricing). If you're comparison shopping this quarter, the merger headline is old news. What matters now is what the numbers say today.

This post separates what actually changed in the deal from what a buyer needs to check before signing anything, and lines up current H100 and B200 rates against Spheron and the rest of the neocloud market.

What Actually Changed When Voltage Park Became Lightning AI

Lightning AI and Voltage Park completed their merger on January 21, 2026. The combined entity operates under the Lightning AI name and is valued at over $2.5 billion, with combined annual recurring revenue exceeding $500 million, up from $18 million in 2024 (Forbes; Voltage Park). That revenue jump reflects Voltage Park's GPU rental business getting folded into Lightning AI's existing platform revenue, not organic growth at either company alone.

The Deal in Plain Terms

William Falcon, Lightning AI's founder, remains CEO of the combined company. Ozan Kaya, Voltage Park's former CEO, becomes President. Saurabh Giri, who came from AWS Bedrock, joins as Chief Product and Technology Officer (Voltage Park). Voltage Park was originally funded by a $900 million grant from Jed McCaleb's Navigation Fund. The Navigation Fund holds a significant equity stake in the merged company, but McCaleb himself has no direct ownership and only advises occasionally (Forbes).

The combined company runs 60 megawatts of active data center capacity across six facilities in four US states. Forbes ranks it the third-largest neocloud operator by this measure, behind CoreWeave and Nebius. Depending on which release you read, the fleet includes 35,000+ (Forbes) or 36,000+ (Voltage Park's own numbers) owned and operated H100, B200, and GB300 GPUs. Neither source reconciles the discrepancy, so treat it as roughly 35,000-36,000 rather than picking one figure as gospel.

One line from Falcon is worth sitting with if you're evaluating neocloud counterparty risk: "It was the only neocloud without debt" (Forbes). That's a direct jab at CoreWeave, which has taken on more than $14 billion in debt to fund its buildout. Whether debt-free growth was worth trading independence for is a separate question, but it tells you how Lightning AI is positioning the acquisition to the market.

What Stays the Same for Existing Voltage Park Customers

If you already have a Voltage Park deployment, nothing about your contract changed on day one. Existing customers keep their current terms with no deployment disruption (Voltage Park). That's the practical answer to the question most existing customers actually have: will my running workload get touched. It won't, at least not as a direct consequence of the merger itself.

What's New

Voltage Park customers now get access to Lightning AI's bundled software stack, inference, model serving, team management, and observability, at no additional cost on top of their existing GPU contract. Saurabh Giri put it bluntly: "Customers spend hundreds of millions on inference platforms that they now get bundled for free on Lightning AI" (Voltage Park). Lightning AI's own platform, PyTorch Lightning plus the Lightning cloud, already serves over 400,000 individual developers, startups, and enterprises, so the bundled tooling isn't a bolt-on prototype. Ozan Kaya framed the long-term thesis this way: "The next generation of cloud platforms won't be built by stitching together single-purpose tools. They'll be AI-native from day one, just as AWS was foundational for the internet era, we are building ground up for the AI era" (Voltage Park).

H100 and B200 Pricing Under the New Lightning AI Umbrella

Is the $1.99/hr H100 On-Demand Rate Still Live

Yes. Voltage Park's public pricing page, as pulled in July 2026, still advertises H100 SXM on-demand rentals at $1.99/hr per GPU, no contract, deployable within 15 minutes and scaling from 1 to 1,016 GPUs on self-serve (Voltage Park pricing). That's the number that made Voltage Park a name in the first place, back when it launched with 24,000 H100s funded by the Navigation Fund grant, and the merger hasn't touched it. If you were quoted $1.99/hr before January and you're re-checking whether that survived the ownership change, it did.

Two caveats sit next to that headline number on the same page. InfiniBand-networked H100 configurations (3200 Gbps) are listed as "contact for pricing" rather than a published rate, and long-term reserve contracts (6+ months, 32 to 8,000+ GPUs) also require a sales conversation. So the $1.99/hr figure applies specifically to unnetworked, self-serve, no-minimum-term H100 rentals. If your workload needs multi-node InfiniBand throughput, you're back to a quote, merger or no merger.

B200 and GB300 Access

There's no public on-demand rate for any Blackwell-class GPU on Voltage Park's pages. B200 and B300 are listed as available "with a long-term contract" only, and GB200/GB300 route straight to a contact-sales form (Voltage Park Blackwells). If you're pricing out a B200 cluster and hoping the merger opened up self-serve Blackwell access the way it did for tooling, it hasn't, at least not publicly. You'll get a number from a sales rep, not a page.

What the Merger Does and Doesn't Tell You About Future Pricing

A merger announcement is not a pricing signal, and it's worth being precise about that distinction. Nothing in the public merger materials commits Lightning AI to holding the $1.99/hr H100 rate indefinitely, and nothing suggests an imminent increase either. The stated capital structure argument (no debt, versus CoreWeave's leverage) is a signal about the company's ability to hold prices steady without needing to raise rates to service debt, but it's not a price guarantee. If you're locking in a contract based on projected pricing, get the rate in writing rather than extrapolating from a January press release that's now half a year old.

Where Spheron Fits If You're Rebuilding Your GPU Vendor List

A vendor ownership change, even one where pricing hasn't moved yet, is a reasonable trigger to re-check your GPU sourcing rather than assume the status quo holds. Here's where things stand today across providers that publish real numbers.

On-Demand H100 and B200 Pricing Comparison Table

Spheron pricing below comes from the live pricing API as of July 2026, split strictly between on-demand (dedicated) and spot offers.

Voltage Park's unnetworked $1.99/hr H100 rate is still the lowest on-demand H100 number in this table, and it's worth taking seriously as a baseline if your workload genuinely doesn't need InfiniBand. But it comes with a structural asterisk that a straight price comparison hides: no published rate covers B200, B300, GB200, or GB300, so a team standardizing on Blackwell can't build a budget off Voltage Park's own pricing page the way they can with a full on-demand catalog. Spheron's B200 SXM6 on-demand pricing starts at $9.30/hr, with spot as low as $5.34/hr, both published rates you can check before you talk to anyone. For the full multi-provider view including AWS, GCP, Azure, and OCI rates, see the GPU cloud pricing comparison 2026.

Pricing fluctuates based on GPU availability. The prices above are based on 05 Jul 2026 and may have changed. Check current GPU pricing → for live rates.

Why a Still-Independent Aggregator Matters When a Vendor's Ownership Just Changed

The Voltage Park merger is a reminder that single-provider bets carry ownership risk on top of pricing risk. Nothing here suggests Lightning AI is a bad steward of the business, existing contracts held and the software bundle is a real add, but any team that had Voltage Park as its sole GPU vendor spent January finding out about a change to their supplier's ownership structure from a press release rather than a contract amendment. An aggregator that pulls live rates across 5+ providers doesn't eliminate that kind of event happening at any single one of them, but it means your sourcing decision isn't locked to one company's cap table. If Voltage Park's terms change under Lightning AI down the line, or a different provider's do, you're comparing against current numbers instead of re-litigating your entire vendor relationship from scratch.

Migration Checklist for Teams Re-Evaluating GPU Vendors This Quarter

If the merger is prompting you to actually re-price your GPU spend rather than just read about it, here's what to check before you commit to anything:

• Re-confirm the rate in writing. Don't quote a vendor's blog post or a news article from January. Pull the live pricing page or an API response, screenshot it, and get the rate in your contract terms.
• Check whether your workload needs InfiniBand. Voltage Park's $1.99/hr applies to unnetworked H100 only; InfiniBand-connected configurations go to a sales quote. Confirm which tier your actual workload requires before comparing headline rates.
• Separate on-demand from spot/preemptible in every comparison. A provider's marketing page will sometimes lead with its cheapest tier. Confirm instanceType (dedicated vs spot) before you put a number in a budget spreadsheet.
• Ask for the Blackwell rate explicitly if you need B200 or newer. If a provider's pricing page doesn't list an on-demand B200/B300/GB200/GB300 rate, assume it's contract-only and budget time for a sales cycle, not a self-serve signup.
• Confirm contract minimums and egress fees separately from the headline hourly rate. These are the line items that turn a competitive quote into an uncompetitive bill six months in.
• Re-check pricing again closer to signing. Rates move. The table above is a snapshot; treat any number older than a few weeks as a starting point for a fresh check, not a final answer.

For a broader look at where displaced Voltage Park buyers can look next, see our CoreWeave alternatives roundup. We also cover more options in the top GPU cloud providers guide. When you're ready to deploy, the Spheron quickstart guide walks through connecting your first instance.

FAQ

Is Voltage Park still a separate company from Lightning AI?

No. The merger completed on January 21, 2026, and the combined company operates under the Lightning AI name. Voltage Park's brand, site, and pricing page are still live and existing customer contracts are unaffected, but new deals and product direction now run through Lightning AI leadership.

Is the $1.99/hr H100 rate still available after the merger?

Yes. As of July 2026, Voltage Park's pricing page still lists H100 SXM at $1.99/hr per GPU on-demand with no contract required, unchanged since before the merger closed.

Can I get B200 or GB300 pricing from Voltage Park/Lightning AI without a sales call?

No. B200 and B300 are listed as long-term-contract only, and GB200/GB300 require contacting sales directly. There is no published on-demand hourly rate for any Blackwell-class GPU on the combined company's pricing pages as of July 2026.

Who runs the combined Lightning AI and Voltage Park company?

William Falcon, Lightning AI's founder, is CEO of the combined company. Ozan Kaya, formerly Voltage Park's CEO, is now President. Saurabh Giri, who previously worked on AWS Bedrock, joined as Chief Product and Technology Officer.

Re-checking your GPU vendor list after the Voltage Park/Lightning AI merger? Spheron's H100 PCIe starts at $2.01/hr on-demand and B200 SXM6 at $9.30/hr, both published rates with no contract minimum and no sales call required.

Spheron H100 instances → | B200 GPU pricing → | Get started →