AMD has revealed the terms of the AMD Taalas deal: it will pay about $2.9 billion in cash and stock for Taalas, the Toronto startup building AI inference chips with a model's weights etched directly into silicon. The figure was disclosed in the company's quarterly report (10-Q) filed with the SEC on November 4, 2026, according to Canadian newspaper The Globe and Mail.
The deal itself is not new: AMD had already announced a definitive agreement to acquire Taalas on August 6, 2026, without disclosing the price at the time. What is new is the value — and the detail that up to $1.4 billion of the total depends on performance conditions that have not been publicly detailed. The acquisition is still expected to close by the end of 2026, subject to regulatory approval.
What AMD is buying
Founded in 2023 in Toronto by Ljubiša Bajić, former CEO of Tenstorrent, Lejla Bajić and Drago Ignjatovic, Taalas bets on a radical idea: instead of loading a language model's weights from memory at runtime, as GPUs do, it etches the model directly into silicon. The result is a chip that runs only that specific model — but does so far faster and more efficiently.
The company's first chip, the HC1, was unveiled in February 2026. Built by TSMC on its 6 nm process, it packs 53 billion transistors into an 815 mm² die and carries Meta's Llama 3.1 8B model etched into a mask-ROM memory fabric, with the KV cache held in SRAM — eliminating the HBM memory used by GPUs. According to Taalas, the HC1 delivers up to 17,000 tokens per second per user; independent measurements cited by specialist publications recorded around 14,300 tokens per second.
“AMD is building a full-stack AI platform that gives customers the flexibility to deploy the right compute solutions for every AI workload,” said Vamsi Boppana, senior vice president of AMD's AI Group, in the August announcement. “Taalas' technology strengthens our AI portfolio with differentiated inference performance and efficiency.”
Ljubiša Bajić, co-founder and CEO of Taalas, said at the time: “We founded Taalas to rethink AI inference from the ground up by building the hardware around the model.”
Why it matters
AMD plans to integrate Taalas' technology into its accelerator roadmap and develop system-level solutions alongside Instinct GPUs, including the Helios rack-scale platform. The bet is that, for stable, high-volume models — such as assistants and automations running the same model millions of times a day — dedicated silicon can cost less and use less power than general-purpose GPUs.
The move follows the industry's race for specialized inference silicon, a segment growing fast as AI shifts from training to large-scale deployment. With the AMD Taalas deal, AMD absorbs its second Toronto inference startup in just over a year. Taalas has raised more than $200 million in venture capital since its founding.
Image credit: Taalas, via NYU Shanghai RITS.
Sources: AMD — official acquisition press release; Tech Insider — value disclosed in the 10-Q; NYU Shanghai RITS — HC1 chip details.
Transparency: this article was produced with the help of artificial intelligence, cross-checked against public posts and online sources. Readers should consult the original sources to confirm the details.
Muse
