AMD has put a price on its pending acquisition of Taalas: about $2.9 billion in cash and stock, according to the chipmaker's quarterly filing published November 4, 2026. Up to roughly $1.4 billion of that consideration is contingent, a large share for a startup that was founded in 2023.
The figure is new information about an existing agreement, not a new signing. AMD announced the definitive agreement to buy the Toronto-based inference-chip company in August without disclosing terms, and its November 3 earnings release mentioned Taalas but again gave no value.
The filing does not say what triggers the contingent payments or how the price splits between cash and AMD stock. AMD expects the deal to close by the end of 2026, subject to regulatory approvals and customary closing conditions.
Key facts
- 01AMD valued its pending Taalas acquisition at about $2.9B in a quarterly filing published November 4, 2026.
- 02Up to about $1.4B of the cash-and-stock consideration is contingent; the filing does not spell out the conditions.
- 03AMD expects the deal to close by the end of 2026, subject to regulatory approvals.
- 04Toronto-based Taalas, founded in 2023, hardwires AI model weights into custom chips; its HC1 runs Meta's Llama 3.1 8B.
- 05Taalas previously disclosed $50M at its 2024 stealth exit and $169M raised earlier in 2026.
Taalas builds what it calls model-specific integrated circuits, chips that bake a model's weights directly into silicon instead of running them on general-purpose GPUs. Its HC1 runs Meta's Llama 3.1 8B and nothing else, and the company has detailed an HC2 chip it expects to handle 20-billion-parameter models.
The pitch is speed and efficiency. Early Taalas demos reportedly generated up to around 17,000 tokens per second per user on certain workloads at 3-bit and 6-bit quantization, and the company says it can turn a previously unseen model into custom hardware in roughly two months, against an industry norm of one to two years. The current practical ceiling is about 8 billion parameters per chip.
“Taalas' technology and world-class engineering team strengthen our AI portfolio by delivering differentiated inference performance and efficiency”— Vamsi Boppana, AMD senior vice-president, AI group
AMD described Taalas as a pioneer in specialized AI inference silicon and said its technology reduces the compute and memory bottlenecks of general-purpose architectures. The plan is to fold Taalas accelerators into AMD's roadmap alongside Instinct GPUs in its Helios rack-scale systems, with ROCm software support.
Taalas was founded by three former AMD and Tenstorrent leaders: CEO Ljubisa Bajic, a Tenstorrent co-founder and its former CEO, along with COO Lejla Bajic and CTO Drago Ignjatovic. The company emerged from stealth in 2024 with $50 million from investors including Quiet Capital and Pierre Lamond, then raised $169 million earlier in 2026 from a group that included Fidelity.
“We founded Taalas to rethink AI inference from the ground up by building the hardware around the model”— Ljubisa Bajic, Taalas co-founder and CEO
The full Taalas team is joining AMD's AI organization and staying in Toronto, making the deal a genuine acquisition rather than an acquihire. It is AMD's fourth inference-focused deal since November 2025, following inference software firm MK1 and other startups.
The valuation lands as AMD's data center business is surging. The company reported Q3 2026 data center revenue of $8.5 billion, up 97% year over year, and ended the quarter with $15.8 billion in cash and short-term investments. AMD also has a separate pending all-stock acquisition of World Labs Technologies valued at about $8.2 billion, agreed in September.
AMD said in August that the Taalas deal reflects a continued commitment to retaining and growing Canadian talent, extending a Toronto presence that dates to its 2006 purchase of ATI Technologies.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
