BigGo Finance reported on Wednesday that AMD completed its acquisition of Toronto-based AI chip designer Taalas on Tuesday. The cash-and-stock deal was worth about $2.9 billion. According to the report, the deal closed about three months after it was announced in August.
The purchase gives AMD chip technology specialized for AI inference, the stage in which a trained model answers user questions or performs tasks. Taalas's design hardwires the information a specific model needs for computation directly into semiconductor circuits, reducing the repeated exchange of data between compute units and memory, according to the report. That approach is intended to improve response speed and power efficiency at a time when inference workloads and their power consumption are rising quickly.
The transaction price includes up to about $1.4 billion in contingent consideration payable only if certain conditions are met. BigGo Finance calculated that figure from AMD's quarterly report filed this month. Taalas was founded in 2023 by co-founders including Ljubisa Bajic, who had earlier started AI chip firm Tenstorrent. The startup is valued at approximately 3.9 trillion won.
AMD plans to integrate Taalas's technology into its AI accelerator roadmap and to build system solutions with its Instinct graphics processing units, EPYC central processing units, ROCm software platform, and rack-scale Helios system. AMD sees the deal as a chance to expand its engineering presence in Canada and to attract top technical talent. "As the AI inference market evolves rapidly, customers need a broad range of computing solutions optimized for their individual workloads," said Vamsi Boppana, who leads AMD's AI Group as senior vice president. He added: "With Taalas joining AMD, we will expand our ability to deliver differentiated inference performance and efficiency while further solidifying the competitiveness of our full-stack AI platform".
BigGo Finance described the purchase as a move to address an industry shift in which the cost of serving AI models depends less on training, which requires massive GPU clusters, and more on everyday inference, where fast, low-power chips determine competitiveness. Nvidia dominates both training and inference with general-purpose GPUs, and AMD's acquisition is seen as targeting custom inference silicon to gain share. Some market participants see the deal as evidence that AMD is widening its AI offerings and continuing to invest in startups focused on inference-specific chips. Others note that commercializing Taalas's technology and winning customers will require considerable integration work. The fact that roughly half the price is contingent on performance conditions underscores that uncertainty.
Sources: BigGo Finance
To report an error in this article, write to corrections@venturistnews.com. Corrections log.
Sam Okafor covers capital markets, public listings and corporate finance for Venturist. He writes from filings, earnings releases and published reporting.
