Karsten Wenzlaff, Advisor
August 26th, 2025
Karsten Wenzlaff is a Crowdfunding and Fintech Expert from Germany. He is Secretary-General of Digit[...]
AMD Buys Toronto AI Chip Startup Taalas
August 12, 2026 | NCFA Market Activity | Artificial Intelligence And Data, Capital Markets And Market Infrastructure, Competition And Market Structure
Toronto Chip Design Brings A More Specialized Inference Option
On August 6, 2026, AMD agreed to acquire Taalas, a Toronto AI chip company that designs specialized chips to run individual AI models more efficiently. The price wasn't disclosed and the transaction hasn't closed yet, as it's subject to customary closing conditions and regulatory approvals.
AMD plans to bring Taalas technology into its accelerator portfolio alongside Instinct GPUs. That's a useful clue to the strategy. GPUs remain valuable because they can handle many models and workloads. Taalas gives AMD another option for cases where a model is used often enough that more specialized hardware could lower the cost of running it.
Taalas Reduces The Work Between Memory And Compute
Running an AI model requires processors to work through huge numbers of stored parameters. On conventional accelerators, much of that data has to travel between memory and compute hardware. The transfers take time, consume power and make high bandwidth memory an expensive part of an AI system. Taalas brings more storage and computation onto the same silicon and tailors the hardware to the model being run. The design can reduce the external memory, advanced packaging and data movement required by conventional accelerator systems.
Ljubisa Bajic, Founder and CEO, Taalas:
“The production of optimal silicon for each individual model.”
Its first HC1 demonstrator runs Meta's Llama 3.1 8B model on a TSMC 6nm chip with 53 billion transistors. Taalas packages the system in a 2.5 kW server and currently provides access through a beta inference service and API.
The trade-off is straightforward. A GPU can be reprogrammed for many models. HC1 is largely built around one. It supports different context lengths and LoRA fine tuning, but a substantially different model requires another hardware implementation.
Taalas says it can turn a new model into silicon in about two months. If that process works economically at larger scale, AMD could use specialized chips for mature, heavily used models while keeping programmable accelerators for workloads that change more often.
17,000 Token Benchmark
Taalas says HC1 can generate about 17,000 output tokens per second per user on Llama 3.1 8B using a 1K input and 1K output sequence.
That's a company benchmark. Taalas ran its own HC1 result and measured the Nvidia B200 comparison itself, while several other comparison figures came from Artificial Analysis. It shows what the architecture can do on this model and configuration. It doesn't establish that Taalas hardware is faster than Nvidia or other accelerators across AI inference.
HC1 also uses a custom format combining 3-bit and 6-bit parameters. Taalas acknowledges some loss in model quality compared with GPU benchmarks and says its next generation will use standard 4-bit floating point formats.
The company also reports large advantages in power use and system cost in its comparison. Those claims haven't been independently demonstrated across a wide range of models or production environments. It makes the acquisition a calculated bet. AI companies are spending enormous amounts to train models, but every model that reaches widespread use can create an equally serious inference problem: how to serve millions of requests quickly enough and cheaply enough.
That pressure is already changing how AI infrastructure companies compete on cost and performance. Taalas gives AMD a way to explore much deeper specialization without abandoning the flexibility of Instinct.
Toronto Built A Chip Company AMD Chose To Buy
Taalas was founded in Toronto in 2023 by Ljubisa Bajic, Drago Ignjatovic and Lejla Bajic. The team brought processor experience from AMD, Nvidia and Tenstorrent, which Ljubisa Bajic previously founded.
The company raised US$50 million before emerging from stealth in 2024 and another US$169 million in February 2026, bringing reported funding to about US$219 million. Investors included Quiet Capital, Fidelity and semiconductor investor Pierre Lamond.
Taalas says 24 people developed HC1, which shows how concentrated the engineering effort behind the first chip was.
The deal adds another company to Canada's growing AI hardware record. Toronto and Waterloo based Astrus is working on automated chip design, while Tenstorrent has built a much larger processor business from Toronto.
It also follows another major Canadian semiconductor transaction. U.S.-based Qualcomm agreed in 2025 to acquire Toronto-founded Alphawave Semi for US$2.4 billion, putting another Canadian-founded chip company under foreign ownership.
AMD will control Taalas if this acquisition closes, but it has also said it plans to retain and grow Canadian talent. There is no disclosed commitment to a specific Toronto headcount or to keeping Taalas as a separate company.
That makes the Canadian issue less about whether foreign capital is inherently good or bad and more about how much ownership, intellectual property and future economic value Canada retains as its AI companies scale. Canada's own AI strategy debate has put sovereign capital and domestic IP retention directly on the table.
Talking Point
Taalas raised more than US$200 million, built working silicon and attracted a strategic buyer in about three years. Canada can clearly produce teams and technology that global semiconductor companies want. The difficult question is whether enough domestic capital, procurement and infrastructure exist for more of those companies to scale further before selling.
The National Crowdfunding & Fintech Association (NCFA Canada) is a financial innovation ecosystem that provides education, market intelligence, industry stewardship, networking and funding opportunities and services to thousands of community members and works closely with industry, government, partners and affiliates to create a vibrant and innovative fintech and funding industry in Canada. Decentralized and distributed, NCFA is engaged with global stakeholders and helps incubate projects and investment in fintech, alternative finance, crowdfunding, peer-to-peer finance, payments, digital assets and tokens, artificial intelligence, blockchain, cryptocurrency, regtech, and insurtech sectors. Join Canada's Fintech & Funding Community today FREE! Or become a contributing member and get perks. For more information, please visit: www.ncfacanada.org
 |  |  |
|---|---|---|
 |  |  |
Leave a Reply