The AI boom isn’t slowing in Europe. VCs invested more than $7bn into AI in Europe in 2023, according to Dealroom. Startups are also deploying AI in ever-more niche areas, like Greek startup Langaware’s plans to predict Alzheimer’s disease or Hortya’s idea that farmers could use AI to talk to plants.
Whatever the solution, these data-intensive models all rely on graphics processing units (GPUs) capable of parsing vast swathes of data. Without reliable access to them, it won’t be possible for startups to unleash their novel and world-changing AI solutions.
The problem is, as more companies get in on the AI boom, access is becoming harder to obtain.
“There is a huge demand. Everyone wants to use, invest in or acquire AI,” said Rosanne Kincaid-Smith, group chief operating officer of Northern Data Group, a high-performance computing solutions provider, in an interview with Sifted’s AI editor Tim Smith. “But a lot of that is driven by hyperscalers, who have access to large amounts of capital. What that means for smaller businesses is that access is limited and the opportunities to experiment and scale are shrinking.”
So how do we strike the balance between unrelenting innovation, the demand for computing power and the competition to be the first, the biggest and the fastest?
Startups struggle to access GPUs
AI startups access “compute” — the industry term for GPU processing — by buying chips or paying to rent them from cloud companies and data centre providers.
In 2023, GPU shortages made it difficult for companies to purchase or access compute reliably, or at affordable prices. In Europe, there are also concerns over whether there are enough data centres to match the demand for training AI models.
It’s something Cambridge-based startup BeyondMath, which is training AI models on physics equations, is currently navigating.
It currently rents GPUs from cloud providers but plans to cut costs by buying a small amount — less than a dozen — of its own GPUs in the near future. It will be enough for the startup’s day-to-day activities, meaning it will only need to use cloud providers for less frequent tasks that require more computing.
If this generation of AI-driven startups and scaleups are to thrive, we mustn’t let them fall at the first hurdle.
“We’re honest with the cloud providers. We’d rather use [them], but they’re so in demand,” Darren Garvey, BeyondMath’s cofounder, says.
NUA Software, creators of AI-powered food recognition technology for the food services industry, has felt for themselves the step changes possible with the benefit of access. They have used Northern Data Group’s high-performance GPUs for training and inferencing AI models, leading to a 99% accuracy rate in food recognition and positive customer feedback thanks to shorter wait times.
This is why Kincaid-Smith said that companies like Northern Data Group, who provide access, must do what they can to democratise it: “If we don’t, it does mean that [AI] becomes the purview of just a few to drive innovation.”
To make this work in practice, Northern Data Group has recently launched an AI Accelerator that will provide selected startups with access to its NVIDIA HGX H100 GPU servers, alongside support from the company’s expert engineers.
Not only do you have the opportunity to scale your concept at rapid speed, but you also get the support that you need to make that business a real prospect.
“The AI Accelerator is our response to market dynamics. If this generation of AI-driven startups and scaleups are to thrive, we mustn’t let them fall at the first hurdle. Our Accelerator is essentially a platform from which to secure access to our H100s and benefit from our experience,” said Kincaid-Smith. “In the spirit of democratisation, the application system is global and open to any startup with an innovative solution that, with support, can change the world for the better.”
The AI Accelerator will also provide these startups with mentorship from industry leaders — including HPE and Supermicro — and access to other learning materials, such as NVIDIA’s Deep Learning Institute and sessions on ESG supported by Deloitte. The Accelerator’s aim is to help startups that are ready to scale up their proof of concepts or products.
“Not only do you have the opportunity to scale your concept at rapid speed, but you also get the support that you need to make that business a real prospect,” Kincaid-Smith added.
Sustainability and AI compute
These solutions not only reduce AI workloads, but also address another problem facing the AI industry — its environmental impact.
“There are different solutions to the same problem and ultimately, if all of them get funded both from venture capital into smaller companies and the larger giants into their research teams, I’m positive about the direction of travel,” says Flavia Levi, a deeptech investor at Octopus Ventures.
However, Levi points out that challenges around energy management may be slower to address, given they are about changing hardware.
“When ChatGPT came online, the energy requirements that came off the back of that spiked,” she says. “That will feed into the data centres.”
Companies with limited access to capital [are] often looking at price, not the impact of that innovation.
Other solutions include rethinking the chips themselves. Fractile, for example, is a UK startup building chips optimised for specific AI tasks, and which can run up to 100 times faster. Intrinsic is another, also based in the UK, making chips with integrated low-power memory, which use far less energy than other types of chips.
As Europe’s largest and cleanest GenAI cloud service provider, Northern Data Group is also addressing AI’s environmental impact through its Accelerator. As well as sessions that cover topics like how to approach investors, workshops on ESG will also be provided, while the GPU servers that startups will have access to are — in line with all of Northern Data Group’s cloud platform — also powered by 100% clean energy.
The Accelerator is open to all geographies and sectors, with startups working in fields like finance, logistics, security and healthcare particularly encouraged to apply.
“Companies with limited access to capital [are] often looking at price, not the impact of that innovation,” said Kincaid-Smith. “We’ve taken steps to place [our data centres] in locations that operate predominantly off renewable sources, and we also use direct-to-chip liquid cooling to ensure there’s less residual heat pushed out into the environment.
“For companies like us we take sustainability incredibly seriously, we consider it part of our ethos to power innovation without compromising our planet.”
Applications to Northern Data Group’s AI Accelerator close July 28. For more information, to check your eligibility or to apply, click here.