AMD (AMD.US) acquires AI inference chip startup Taalas to improve data center product lineup

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

AMD did not disclose the transaction amount. According to available information, Taalas, founded in 2023, has raised about $219 million. This deal comes just about seven months after Nvidia (NVDA.US) spent $20 billion to acquire assets from high-performance AI chip design company Groq, highlighting that AI inference chips are becoming the new battleground in the sector.

With the rapid popularization of generative AI applications, the industry focus is gradually shifting from model training to model inference, which refers to the computational process where AI models respond to user requests in real-world business scenarios.

Unlike general-purpose computing chips such as GPUs, Taalas develops dedicated inference accelerators (ASICs) that are hardware-customized for specific AI models. These chips are “hardwired” for certain 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 only about two months from receiving a new AI model to completing hardware implementation.

Currently, Taalas’s products primarily support the Llama 3.1 small model developed by Meta (META.US), and they are developing new products for larger-scale models. The chips are manufactured using TSMC’s (TSM.US) mature processes and integrate high-speed SRAM storage to reduce inference latency.

Industry experts believe that this type of dedicated inference chip is especially suitable for low-latency scenarios, such as AI chatbots, real-time search, autonomous driving, and intelligent assistants—applications that require rapid response.

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

She pointed out that GPUs, with their high versatility, will continue to account for the majority of the AI chip market since they can support ever-evolving new-generation AI models. At the same time, specialized accelerators for different use cases are needed to build a complete AI computing ecosystem.

Currently, strong GPU demand has driven Nvidia's market cap to surpass $5 trillion, making it the world’s most valuable publicly traded company.

This acquisition also shows 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 individual chips to full rack-level systems. This year, AMD began delivering its first rack-scale AI system, Helios, to customers such as Meta and Microsoft (MSFT.US), directly competing with Nvidia’s DGX systems.

AMD stated that in the future, Taalas’s inference chips and related technologies will be integrated into the company’s product roadmap, working in tandem with its EPYC CPUs, Instinct GPUs, and complete system solutions to provide customers with a more comprehensive data center AI infrastructure.

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

Previously, AMD also completed several AI-related acquisitions, including the $665 million acquisition of AI model developer Silo AI in 2024, and the $4.9 billion takeover of server manufacturer ZT Systems to lay the technical foundation for its Helios rack products. In addition, last year the company also acquired several AI startups, including inference software developer MK1.

According to analysts, as the AI industry gradually shifts from model training to large-scale commercial deployment, demand for inference computing is growing rapidly. By acquiring Taalas to further enhance its dedicated inference chip capabilities, AMD is expected to improve its overall competitiveness with Nvidia in the enterprise AI infrastructure market.