Capital

Nvidia and Intel Backed an Off-Lab Way to Train AI

Prime Intellect raised 130 million dollars at a 1 billion dollar valuation to pool idle GPUs worldwide so companies can train their own AI agents without a frontier lab. Why that reprices the build-versus-rent decision for European operators.

CapitalBy Servola Tech Desk2026-07-123 min read

AI-assisted, edited by humans. | Editor-in-Chief: Leon Soliman | Editorial standards

Nvidia and Intel Backed an Off-Lab Way to Train AI

The round that put chipmakers behind a rival route

Prime Intellect, founded in 2024 by Vincent Weisser and Johannes Hagemann, closed a 130 million dollar Series A this week at a 1 billion dollar valuation, led by Radical Ventures with Nvidia Ventures, Intel Capital and Dell Technologies Capital taking part. That lifts its total funding above 150 million dollars. The company builds what it calls an open stack for training: it aggregates idle datacenter GPUs from around the world, runs reinforcement-learning training across that scattered supply, and lets customers deploy the agentic models that come out the other end.

The notable detail is not the size of the round but who wrote the checks. Nvidia, Intel and Dell sell the picks and shovels of the AI build-out. Funding a company whose whole pitch is to squeeze more training out of borrowed, fragmented silicon is a bet on the same demand from a different direction, and a quiet acknowledgement that not every buyer wants to route through a frontier lab.

What distributed training changes for a buyer

Until now, training a capable model meant one of two things: paying a frontier lab for access to its model, or reserving a large contiguous GPU cluster from a hyperscaler. Prime Intellect aggregates idle compute worldwide and runs the post-training across that fragmented pool, then passes the cost gap to customers. A roughly 100 million dollar annualized revenue rate, with names like Ramp and Zapier on the list, says the appetite for a middle path is already real rather than theoretical.

The catch is that a global pool is not free sovereignty. Where your data and gradients travel, how reliable borrowed nodes are, and how the whole arrangement sits under the GDPR all become your problem to manage rather than the lab's. The gain is control over a model you own outright; the cost is that the operational and compliance burden moves onto your side of the table.

How a European operator should read it

For a European business the headline is not another AI unicorn. It is that a credible, chipmaker-backed alternative to renting from OpenAI, Anthropic or a United States hyperscaler now exists and is funded to scale. That widens the menu the next time you scope an agent project, and it changes the negotiating position even if you never switch.

Before you assume you must buy a frontier subscription, price the build path: what a model trained on your own data and owned outright would cost through a distributed stack, and whether your data-residency rules can accept GPUs you do not control. The option is genuinely new. Whether it fits is a question only your own numbers and your own regulator can answer.

Frequently asked questions

What did Prime Intellect raise and who backed it?

A 130 million dollar Series A at a 1 billion dollar valuation, led by Radical Ventures with Nvidia Ventures, Intel Capital and Dell Technologies Capital taking part, which lifts total funding above 150 million dollars. The company was founded in 2024 by Vincent Weisser and Johannes Hagemann.

How is distributed training different from renting a cloud cluster?

Instead of reserving one large contiguous GPU cluster, the platform aggregates idle compute from many datacenters worldwide and runs reinforcement-learning training across that fragmented supply, then passes the cost advantage to the customer.

Should a European company drop a frontier lab because of this?

Not automatically. It adds a credible option to build and own your model, but distributed training raises data-residency, reliability and GDPR questions. Price the build path against your frontier subscription using your own numbers before you move.

The number that will travel is the valuation, but the signal underneath it is who wrote the checks. When the companies that sell GPUs, CPUs and servers fund a startup built to wring training out of borrowed, scattered silicon, they are hedging their own customers. For an operator the takeaway is narrower and more useful: the choice is no longer only rent from a frontier lab or reserve a hyperscaler cluster. A third route now carries a billion-dollar backing and paying customers. Whether it fits your data and your risk rules is the question to run before your next model decision, not after it.

Prime IntellectVincent WeisserRadical VenturesNvidia Venturesdistributed trainingagentic AIGPU

SK Hynix Raised $26.5B in a Record Nasdaq Debut

SK Hynix priced its Nasdaq ADR debut at USD 149 on July 10, 2026, raising about USD 26.5 billion, the largest US listing ever by a foreign company, and the stock jumped 17 percent on day one. What it means for the memory market and compute buyers.

4 min read

Capital

A Carrier Just Bought the Glue Between Your Clouds

Lumen completed a 475 million dollar acquisition of Alkira, the network-as-a-service control plane that stitches multi-cloud traffic together. The neutral layer enterprises used to connect AWS, Azure and Google now sits inside a carrier's roadmap and pricing. Here is what changes for anyone whose fabric runs through it.

2 min read

Digital Sovereignty

The EU Council Revived the Rule That Scans Private Messages

The EU Council revived the lapsed rule that lets providers scan private messages with AI. For owners, confidential communication is now a governance choice.