Altrove, an exciting startup based in Paris, is using AI to shake things up in the materials science world.

Altrove was co-founded by Cambridge alum Dr. Joonatan Laulainen and second-time founder Thibaud Martin at the competitive international startup accelerator Entrepreneur First (backed by Reid Hoffman and founders/leadership of Stripe, Wordpress, Github, Twitch, Deepmind, and more). In early 2024, Altrove raised € 3.7 million in pre-seed funding from angels that include HuggingFace CTO Julien Chaumond.

Altrove develops alternatives to the scarce and increasingly fraught rare earth elements (REE) like cobalt and nickel that are essential ingredients in products like phones and electric vehicles. In materials science, like drug discovery, breakthroughs don’t come easy. Developing something new can take up to 100 iterations, and each one can drag on for weeks when decisions are made manually. They accomplish materials discovery 100x faster with a proprietary experimentation process that relies on AI.

They required GPUs to power these critical compute-heavy AI workloads that differentiated their business so they turned to Valdi’s on-demand GPUs for a solution that made sense for their startup: the right GPUs at the right price, quickly. And critically, enough expert support to get up-and-running quickly, and the flexibility to change everything in the future, if needed.

‍

Challenge.

Altrove got its start through the Entrepreneur First accelerator, which connected them to valuable resources—including a startup program with AWS that gave Altrove access to ~$50k in free GPU cloud credits, which they used to build their proof of concept.

“As a startup in the tech space, AWS and various cloud platforms basically throw money at you…in the hopes of snaring you as a long-term customer, I presume,” said Joonatan. “You can usually amass about $40-50k worth of credits and it's usually a pretty good place to do your initial testing.”

But as their journey progressed, those credits began to run out, and the team faced a big challenge: finding GPUs to keep their AI workflows running for fundraising and beyond.

“It was our first idea when we went to fundraise,” said Joonatan. “The fundraise plans were drawn up in late February/early March. We had to indicate what we needed the money for. And that's what... a month into the company existence, give or take? And the need for compute was there right from the beginning.”

Stick with hyperscalers?

So why didn’t they just stick with AWS or another traditional cloud provider once their credits ran out? According to Joonatan, it came down to cost:

“We needed to go bigger in terms of compute size…and compute through these institutional platforms…is really bloody expensive.”

Buy GPUs?

They explored their options and briefly considered buying their own GPUs. However, the upfront costs were high and the inflexibility of that route didn’t make sense for a small, growing startup with shifting needs.

“It would have been cheaper than renting in 3+ years. We were almost ready to pull the trigger on buying something. But in the end we chose not to because it was uncertain exactly what our compute needs would evolve to be. Early on we kind of knew what we needed but we didn't know the quantity. And there's a lot of architectural questions about how you connect these things. Because we weren't certain about those questions…It didn't make sense to make a commitment at that scale.”

Renting GPUs

That’s when they looked into renting for the flexibility it would give them. They started exploring their options based on clear priorities: securing H200 GPUs for a period of 1 yr, ensuring costs were transparent and reasonable, and partnering with someone they could trust.

Solution.

Here’s how their journey unfolded:

Genesis Cloud

Altrove’s search for GPU power started with Genesis Cloud, a respected leader in high performance computing for AI teams worldwide. The Genesis team carefully evaluated Altrove’s needs. Recognizing that Valdi’s unique approach—combining raw power with infrastructure expertise—would be an even better fit, they referred Altrove to Valdi.

Valdi

Before founding Valdi, Nikhil Jain and Mark Harvilla co-founded ObEN, an AI company building personal avatars for the metaverse. The vision was bold, but the biggest challenge wasn’t the AI—it was getting access to GPUs. Compute shortages slowed innovation and drove up costs, making scaling AI models a constant struggle.

Instead of accepting the problem, they set out to solve it. That’s how Valdi was born. With a platform spanning thousands of GPUs worldwide, Valdi lets AI teams spin up or reserve powerful clusters quickly without massive upfront costs or commitment. Unlike traditional cloud providers, Valdi operates like a customer-focused “SWAT team” for AI infrastructure. Teams don’t just get GPUs—they get expert guidance to design, optimize, and scale their workloads. This hands-on, problem-solving approach is why companies like Altrove choose Valdi—not just for raw compute, but for the expertise that comes with it.

Dedicated H200s

Altrove’s work demanded the power of NVIDIA H200 GPUs for their inference workloads, so they prioritized finding a provider who could deliver these specific resources. Dedicated, in this case, as opposed to “on-demand” because Altrove needed the hardware for a 1-year period.

"It was pretty clear we wanted H200. For the high memory tasks we do, H200 is considerably more performant than the H100 but about the same price,” said Laulainen. “So we looked to see who had that chip available right now, and can we talk to them? That was the strategy.”

Valdi checked this box immediately.

Transparent costs

One of the biggest frustrations with other providers was taking too long to make it clear what Altrove would need to pay for the precommitment, which Altrove knew was often half or more upfront, for startups. It can take quite a while to get to a stage where a provider is willing to commit to what your upfront costs will be. Valdi stood out by making their costs clear and straightforward from day one. Laulainen illustrated:

“People will say ‘it’ll depend, it'll depend’. And then finally you get a price and you're like… that's 10% more than I expected and that's a lot of money obviously, in this case. Valdi’s pricing was fast and it was transparent. I was pretty appreciative of the fact that it was clear what we were going to pay from an early stage.”

Trustworthy provider

For startups like Altrove, reliability and trust are non-negotiable when selecting a GPU provider. Their work relies on consistent performance and the assurance that their partner can deliver on promises. To ensure they were making the right choice, Altrove involved their investors in conducting due diligence on Valdi. After evaluation, they concluded that Valdi was a provider they could confidently work with.

“It's this complicated thing, between trying to figure out what providers are out there, who you're willing to give quite a lot of money to…knowing we don’t have a guarantee those guys will continue to exist. That’s why we did due diligence on you guys to determine if you were trustworthy enough, so to speak, to send significant money to.”

Results.

For Altrove, having access to high-performance GPUs was essential to their mission of accelerating materials science innovation. Their entire process depends on compute power. Without reliable GPUs, they can’t run their models, which means they can’t deliver results to their customers.

Valdi’s solution addressed these critical needs while offering significant advantages over the alternatives Altrove considered:

• Cost Savings: Compared to sticking with AWS or purchasing their own GPU hardware, Valdi’s pricing model saved Altrove a substantial amount of money. AWS and other traditional cloud providers were simply too expensive to use after free credits ran dry. And buying the hardware they needed would have a substantial upfront cost and a lengthy 3-yr. payback period.

• Flexibility: And unlike buying GPUs outright, Valdi’s cloud model allowed Altrove to change their compute resources as their architectural needs emerge or evolve, as new GPUs come on market, or simply because their needs have changed. This flexibility was critical for an early-stage startup still figuring things out. "We didn’t want to lock ourselves into specific hardware," said Laulainen. "Valdi let us stay agile."

• High Performance Hardware: Altrove needed high-performance H200 GPUs to handle their demanding AI workloads. Valdi’s extensive global inventory of the most sought-after GPUs ensured they could secure the exact hardware they required, wherever and whenever they needed it. This ensured Altrove could continue running their AI inference tasks with the performance their business depended on.

• Transparency & Trust: With Valdi, Altrove has a partner they can trust to deliver both the hardware and the dependable service needed to keep their business moving forward. “Valdi is reachable, transparent, and very good, technically speaking.”

In his own words:

“If someone came to me and said they needed compute for X reason, I would definitely direct them to Valdi. I’ve had a positive experience.”

‍