AMD is set to acquire a Finnish artificial intelligence startup called Silo AI for $665 million in one of the most significant takeovers in Europe, as the US chipmaker aims to enhance its AI offerings to compete with the leading player, Nvidia.

Headquartered in California, AMD announced that Silo’s team of 300 members will utilize its software tools to develop customized large language models (LLMs), the type of AI technology that serves as the foundation for chatbots like OpenAI’s ChatGPT and Google’s Gemini. The all-cash transaction is anticipated to be finalized in the latter part of this year, pending regulatory approval.

“This deal will enable us to expedite our interactions with customers and implementations, while also accelerating the development of our own AI technology stack,” stated Vamsi Boppana, who serves as the senior vice president of AMD’s artificial intelligence division, in an interview with the Financial Times.

This acquisition marks the biggest purchase of a privately owned AI startup in Europe since Google’s acquisition of the UK-based company DeepMind for approximately 400 million pounds in 2014, as reported by data from Dealroom.

This agreement comes at a time when acquisitions by tech companies in Silicon Valley are facing heightened scrutiny from regulators in Brussels and the UK. AI startups in Europe, including Mistral, DeepL, and Helsing, have garnered significant investments this year, as investors look for a local contender to challenge US-based entities like OpenAI and Anthropic.

Located in Helsinki, Silo AI is one of the largest private AI laboratories in Europe, providing tailored AI models and platforms to corporate clients. The Finnish startup launched an initiative last year to create LLMs in European languages, including Swedish, Icelandic, and Danish.

AMD’s AI technology stands in competition with that of Nvidia, which currently dominates the high-performance chip market. Nvidia’s success has driven its valuation past $3 trillion this year, as tech corporations strive to establish the computing infrastructure required to support the most advanced AI models. AMD rolled out its MI300 chips towards the end of last year as a direct challenge to Nvidia’s “Hopper” chip line.

Peter Sarlin, the co-founder and chief executive of Silo AI, described the acquisition as the “logical progression” for the Finnish group as they aim to become a premier AI firm.

Silo AI is dedicated to offering “open source” AI models that are freely accessible and customizable by any individual. This sets it apart from companies like OpenAI and Google, which prefer utilizing their own proprietary or “closed” models.

In the past, the startup has introduced its array of open models named “Poro” as a crucial step towards “enhancing European digital sovereignty” and democratizing the availability of LLMs.

The increasing centralization of powerful LLMs in the hands of a select few Big Tech companies based in the US has attracted scrutiny from antitrust regulators in Washington and Brussels.

The acquisition of Silo illustrates AMD’s efforts to swiftly expand its business and boost customer engagement with its proprietary offerings. AMD perceives Silo, which crafts customized models for clients, as a bridge between its core AI software and the practical applications of the technology.

Software has emerged as a new battleground for semiconductor companies as they endeavor to lock in customers to their hardware and create more stable revenues, beyond the erratic chip sales cycles.

Nvidia’s success in the AI domain can be traced back to its substantial investment in Cuda, its proprietary software that enables chips originally developed for graphics processing and gaming to handle a broader range of applications.

Since the inception of Cuda in 2006, Nvidia has extended its software platform to encompass a variety of applications and services, predominantly tailored to corporate customers lacking the internal resources and expertise that Big Tech corporations possess for building on its technology.

Nvidia currently offers over 600 “pre-trained” models, simplifying deployment for customers. The company from Santa Clara, California, recently introduced a “microservices” platform called NIM, designed to facilitate rapid development of chatbots and AI “co-pilot” services.

While historically providing its software at no cost to buyers of its chips, Nvidia announced plans this year to introduce charges for products such as NIM.

AMD, alongside other companies, is contributing to the development of Triton, an alternative to Cuda led by OpenAI, which promises to ease the transition for AI developers between different chip providers. Meta, Microsoft, and Intel have also collaborated on Triton.

© 2024 The Financial Times Ltd. All rights reserved. Please do not copy and paste FT articles and redistribute by email or post to the web.