AMD acquires AI inference chip startup Taalas to enhance its data center product matrix

According to Zhitong Finance APP, American chip manufacturer AMD (AMD.US) is further expanding its artificial intelligence (AI) chip layout. As the generative AI wave enters its fourth year, GPUs remain the core hardware for AI training and inference, but the industry is gradually realizing that a single GPU cannot meet all AI application scenarios. On Thursday, AMD announced that it has reached an agreement to acquire Taalas, an AI inference chip startup based in Toronto, Canada, further strengthening its competitiveness in the inference chip field and improving its data center AI product matrix.

AMD did not disclose the transaction amount. According to information, Taalas, founded in 2023, has raised approximately $219 million in total financing. This transaction comes just about seven months after NVIDIA (NVDA.US) spent $20 billion to acquire the assets of high-performance AI chip design company Groq, reflecting that AI inference chips are becoming a new focus of industry competition.

As generative AI applications rapidly proliferate, the industry's focus is gradually shifting from model training to model inference, which is the computational process of AI models responding to user requests in actual business scenarios.

Unlike general-purpose computing chips like GPUs, Taalas develops dedicated inference accelerators (ASICs) that are hardware-customized for specific AI models. The chips are optimized for "hard-wired" connections to specific models during design, so while they have lower flexibility, they can significantly improve inference efficiency and reduce costs.

According to Taalas, its technology can quickly convert any AI model into a dedicated silicon chip, taking about two months from receiving a brand new AI model to completing hardware implementation.

Currently, Taalas's products mainly support the small model Llama 3.1 under Meta (META.US) and are developing new products suitable for larger-scale models. Its chips are manufactured using TSMC's (TSM.US) mature process and integrate high-speed SRAM storage to reduce inference latency.

Industry insiders believe that such dedicated inference chips are particularly suitable for low-latency scenarios, such as AI chatbots, real-time search, autonomous driving, and smart assistants that require quick responses.

AMD CEO Lisa Su previously stated that she has always believed that there is no "one chip fits all applications" situation in the AI chip market.

She pointed out that GPUs, with their high versatility, will still occupy a large share of the AI chip market because they can support the continuously evolving new generation of AI models. However, at the same time, different application scenarios also require more specialized accelerators to jointly build a complete AI computing ecosystem.

Currently, GPU demand remains strong, driving NVIDIA's market value to exceed $5 trillion, making it the highest-valued publicly traded company in the world.

This acquisition also reflects that AMD is shifting from simply selling GPUs to providing complete AI computing system solutions.

In recent years, competition in AI data centers has expanded from single chips to whole cabinet systems. This year, AMD began delivering its first cabinet-level AI system, Helios, to customers such as Meta and Microsoft (MSFT.US), competing with NVIDIA's DGX systems AMD stated that it will integrate Taalas's inference chips and related technologies into the company's product roadmap, developing in synergy with EPYC CPUs, Instinct GPUs, and complete systems to provide customers with a more comprehensive data center AI infrastructure.

In fact, AMD has been continuously enhancing its AI ecosystem through acquisitions recently. In July of this year, the company announced a partnership with AI chip company Cerebras to integrate its AI accelerators into AMD systems.

Previously, AMD has completed several AI-related acquisitions, including spending $665 million in 2024 to acquire AI model developer Silo AI and acquiring server manufacturer ZT Systems for $4.9 billion, laying the technical foundation for the Helios cabinet products. Additionally, the company acquired several AI startups last year, including inference software developer MK1.

Analysts believe that as the AI industry gradually shifts from model training to large-scale commercial deployment, the demand for inference computing is rapidly growing. By acquiring Taalas, AMD further enhances its capabilities in dedicated inference chips, which is expected to improve its overall competitiveness against NVIDIA in the enterprise AI infrastructure market