Finland’s Silo AI is being acquired by AMD for $665 million. As Europe’s “largest private AI lab,” it already provides services to the likes of Rolls Royce and Philips. However, the acquisition is primarily a step for AMD to enter the AI battle with Nvidia, for which it has increasingly better papers.
Silo AI has a multifaceted offering and with it a diverse customer base. Use cases vary from AI assistance for cancer diagnoses in Philips medical devices to driving aids inside Rolls Royce, Honda and Mitsubishi cars. Co-founder and CEO Peter Sarlin sees a vast range of applicabilities in a variety of other sectors, from retail to finance.
As a European player, it has sought to make AI a core element for companies – in a European manner. Which is to say: a deployment of AI infrastructure on this continent, with European languages supported and an open-source base to build value, Sarlin has outlined. It will be music to many a Brussels policymaker’s ears, especially since Europe is very much behind when it comes to AI infrastructure. Enter AMD, which is shelling out a similar amount for Silo AI as Google paid for the famous DeepMind in 2014.
AMD picks up European players
On Wednesday afternoon, AMD unveiled that it would acquire Silo AI for $665 million, roughly 613 million euros. This arms the chipmaker with an AI startup that not only has a mature customer base, but also 300 scientists and other experts, more than half of whom have a PhD in the AI field. In effect, AMD is gaining an additional AI division, something made more apparent by the fact that Peter Sarlin is staying on board. He will report to Vamsi Boppana, SVP of the Artificial Intelligence Group at AMD, as soon as the deal closes later this year.
The Silo AI acquisition amount is nowhere near what AMD paid for FPGA maker Xilinx in 2022 (49 billion). Nor is it an expansion of its own portfolio like the Radeon division was, which was formed out of the $5.6 billion acquisition of ATI in 2006, making AMD not only a CPU maker but also a GPU vendor. Still, this expansion of its own AI offering is a big step.
AMD lags behind in the AI battlefield no matter which way you slice it. Nvidia, with nearly 98 percent market share in datacenters, is supreme when it comes to GPUs. While CPUs certainly play a role in AI acceleration, GPUs are currently considered the real workhorse to enable GenAI with. There’d be no ChatGPT, GitHub Copilot or Google Vertex AI without a fleet of Nvidia chips that emanate through IT infrastructures worldwide. The fact that just about every IT vendor worth its salt is now an Nvidia partner says enough about the huge impact the company has had on the tech landscape. That extends beyond hardware, as we discussed earlier, and cries out for a mature competitor.
Opportunities in Europe
There is also work to be done for Europe. Mistral AI complained a little over a month ago about a shortage of datacenter capacity, and with it, AI training capacity on the continent. One is much more likely to run into an Nvidia chip in America or Asia than in Europe. Despite the fact that there are plenty of promising AI players located within the EU, it seems they will eventually be forced to move out to America or East Asia if the situation fails to change over time. However, there is an opportunity within Europe to do AI, “EU style”. That requires a preconceived plan, even if it may require the aid of a North American company like AMD.
Why is this important in light of the Silo AI acquisition by AMD? Because it reflects a clear mindset from AMD, from which real opportunities in Europe will emerge. The whole vision behind the EU AI Act has its detractors, chief among them being Mistral AI once more, but it is based on sovereignty, security and due diligence. This is in contrast to a “move fast and break things” credo that has taken hold in the U.S. AI approach for now, where concerns about copyrights, privacy concerns and the societal impact of AI are fought out in court rather than defined by law.
AMD is steadily building an ecosystem that supports the European AI vision. That makes it attractive for infrastructure projects to bet heavily on AMD GPUs. In terms of performance, the latest offering of MI300 chips sit between Nvidia’s older H100 and the up-and-coming Blackwell series, but are reportedly significantly cheaper. Like Intel Gaudi 3, it still counts as an AI alternative when no Nvidia offerings are affordable or available. That may change if AMD becomes prominent as a European-minded player. Given Europe’s need to catch up to provide enough AI capability, the continent may well find a supplier of note in Silo AI-enhanced AMD.
Also read: Intel feels hot breath AMD Epyc 5 with Xeon 6 chips