# AMD to Acquire Taalas, Bringing Model-Specific AI Silicon Into Its Accelerator Roadmap > Source: > Published: 2026-08-07 19:51:29+00:00 # AMD to Acquire Taalas, Bringing Model-Specific AI Silicon Into Its Accelerator Roadmap - AMD announced a definitive agreement to acquire Taalas on August 6, 2026. The announcement disclosed neither a purchase price nor an expected closing date, and said the transaction remains subject to customary closing conditions and regulatory approvals. [[1]](https://ir.amd.com/news-events/press-releases/detail/1296/amd-acquires-taalas-to-advance-compute-solutions-for-rapidly-growing-ai-inference-market) - Taalas’ HC1 Technology Demonstrator is built on TSMC’s 6-nanometer process, has an 815-square-millimeter die and 53 billion transistors, and runs Meta’s Llama 3.1 8B model. [[2]](https://taalas.com/products/) - The startup’s architecture places model weights in a mask-ROM recall fabric paired with SRAM, trading general-purpose flexibility for highly specialized inference. [[3]](https://www.datacenterdynamics.com/en/news/ai-chip-startup-taalas-raises-169m-unveils-hc1-processor-optimized-for-llama-31-8b/) - Taalas reported $169 million in its latest funding round, bringing total reported funding to about $219 million. The public materials reviewed for this article do not name commercial customers. [[4]](https://www.datacenterdynamics.com/en/news/ai-chip-startup-taalas-raises-169m-unveils-hc1-processor-optimized-for-llama-31-8b/) AMD has agreed to acquire Taalas, a Toronto-based startup that builds specialized chips by embedding AI models directly into silicon, as the chipmaker expands its inference portfolio beyond conventional GPU-based systems. AMD said on August 6, 2026, that it plans to integrate Taalas’ technology into its accelerator roadmap and develop system-level products that combine it with AMD Instinct GPUs. [[1]](https://ir.amd.com/news-events/press-releases/detail/1296/amd-acquires-taalas-to-advance-compute-solutions-for-rapidly-growing-ai-inference-market) The transaction is an agreement to acquire, not a completed acquisition. AMD’s announcement does not disclose financial terms or a target closing date. It says the deal remains subject to customary closing conditions and regulatory approvals. [[1]](https://ir.amd.com/news-events/press-releases/detail/1296/amd-acquires-taalas-to-advance-compute-solutions-for-rapidly-growing-ai-inference-market) ## A chip built around one model Taalas’ first public product is the HC1 Technology Demonstrator, a chip designed to run Meta’s Llama 3.1 8B language model. Taalas lists an 815-square-millimeter die fabricated on TSMC’s 6nm process, with 53 billion transistors and a 2.5-kilowatt server configuration. The company says HC1 delivers 17,000 tokens per second per user on Llama 3.1 8B. [[2]](https://taalas.com/products/) Taalas’ product page identifies the result as its own laboratory performance measurement and specifies a comparison setup using 1,000-token input and output sequences. The company also says the comparison data for other systems comes from a mix of vendor measurements and Artificial Analysis, so the figure should be treated as a company-reported demonstration rather than an independent benchmark. [[2]](https://taalas.com/products/) The architecture is designed around a known workload. Taalas has described a mask-ROM recall fabric that embeds model parameters and an SRAM recall fabric that stores model data and handles key-value-cache computations. The approach is intended to reduce memory movement and eliminate components that a general-purpose accelerator may not need for a fixed model. [[3]](https://www.datacenterdynamics.com/en/news/ai-chip-startup-taalas-raises-169m-unveils-hc1-processor-optimized-for-llama-31-8b/) The trade-off is flexibility. HC1 is built around Llama 3.1 8B rather than a broad range of models. Taalas has said the architecture supports adapters and customizations, including fine-tuning paths, but a substantially different model may require a new hardware configuration. [[3]](https://www.datacenterdynamics.com/en/news/ai-chip-startup-taalas-raises-169m-unveils-hc1-processor-optimized-for-llama-31-8b/)[[5]](https://api.taalas.com/) ## Funding and public product access Taalas emerged from stealth in March 2024 and reported a $169 million funding round in February 2026. Data Center Dynamics reported that the round brought total funding to roughly $219 million. Publicly identified backers include Quiet Capital, Fidelity and semiconductor investor Pierre Lamond. [[4]](https://www.datacenterdynamics.com/en/news/ai-chip-startup-taalas-raises-169m-unveils-hc1-processor-optimized-for-llama-31-8b/)[[6]](https://quiet.com/essays/ai-at-the-speed-of-silicon/) The startup has offered access to HC1 through a chatbot demonstration and an OpenAI-compatible inference API. Taalas’ API documentation identifies HC1 as the underlying hardware and lists a model endpoint called “three_bit_numerics.” [5] The public announcements and coverage reviewed for this article do not name commercial customers or disclose hardware revenue, inference revenue or production deployment volumes. Taalas was founded in 2023 by Ljubisa Bajic, a former AMD and Nvidia architect and co-founder of Tenstorrent, along with Drago Ignatovic and Lejla Bajic. AMD identified Bajic as Taalas’ co-founder and chief executive in its acquisition announcement. [[1]](https://ir.amd.com/news-events/press-releases/detail/1296/amd-acquires-taalas-to-advance-compute-solutions-for-rapidly-growing-ai-inference-market)[[4]](https://www.datacenterdynamics.com/en/news/ai-chip-startup-taalas-raises-169m-unveils-hc1-processor-optimized-for-llama-31-8b/) ## Where AMD says the technology fits AMD said Taalas’ technology will complement its Helios rack-scale systems, Instinct GPUs, EPYC CPUs and ROCm software. The company specifically described plans to integrate the technology into its accelerator roadmap and develop system-level solutions with Instinct GPUs. [[1]](https://ir.amd.com/news-events/press-releases/detail/1296/amd-acquires-taalas-to-advance-compute-solutions-for-rapidly-growing-ai-inference-market) That suggests a role alongside AMD’s programmable accelerators: specialized silicon could handle selected, high-volume inference paths while GPUs continue to serve workloads that require broader model and software flexibility. This is an inference from AMD’s stated integration plan and Taalas’ architecture, not a disclosed product schedule. [[1]](https://ir.amd.com/news-events/press-releases/detail/1296/amd-acquires-taalas-to-advance-compute-solutions-for-rapidly-growing-ai-inference-market)[[3]](https://www.datacenterdynamics.com/en/news/ai-chip-startup-taalas-raises-169m-unveils-hc1-processor-optimized-for-llama-31-8b/) The acquisition gives AMD a model-specific inference architecture at a time when infrastructure operators are looking for ways to reduce the cost and latency of serving deployed models. But the public record remains limited. AMD has announced the agreement, not a completed transaction or a shipping AMD product based on Taalas technology. Taalas’ current public demonstrator remains HC1, while the startup has described an HC2 program aimed at larger models, including a 20-billion-parameter version of Llama 3.1. [[2]](https://taalas.com/products/)[[4]](https://www.datacenterdynamics.com/en/news/ai-chip-startup-taalas-raises-169m-unveils-hc1-processor-optimized-for-llama-31-8b/) ## Companies mentioned ## Further sources [[1] AMD’s August 6, 2026 announcement of the definitive agreement, intended integra… ↗](https://ir.amd.com/news-events/press-releases/detail/1296/amd-acquires-taalas-to-advance-compute-solutions-for-rapidly-growing-ai-inference-market) [[2] Taalas’ product specifications and stated test conditions for HC1, including Ll… ↗](https://taalas.com/products/) [[3] Data Center Dynamics’ reporting on Taalas’ model-specific design, mask-ROM and … ↗](https://www.datacenterdynamics.com/en/news/ai-chip-startup-taalas-raises-169m-unveils-hc1-processor-optimized-for-llama-31-8b/) [[4] Data Center Dynamics’ reporting on Taalas’ February 2026 funding round, total c… ↗](https://www.datacenterdynamics.com/en/news/ai-chip-startup-taalas-raises-169m-unveils-hc1-processor-optimized-for-llama-31-8b/) [[5] Taalas API documentation describing the HC1 inference service, OpenAI-compatibl… ↗](https://api.taalas.com/) [[6] Quiet Capital’s account of its investment in Taalas and its participation in th… ↗](https://quiet.com/essays/ai-at-the-speed-of-silicon/) The stories that matter, in one email. Free — unsubscribe anytime.