AMD’s Bold Move Into Custom AI Silicon with Taalas Acquisition

AMD has agreed to acquire Taalas, a startup focused on AI inference chips. Taalas builds special chips designed for single AI models. These chips are “hard-wired” for specific tasks, unlike general GPUs.

The key advantage is speed. Taalas’ chips can output results thousands of times faster than traditional GPUs. They do this by etching AI models directly into silicon. This approach makes the chips less expensive and much faster at running targeted AI tasks.

Right now, Taalas’ chip runs a small version of Meta’s Llama 3.1 model. They are also developing chips for larger and more complex models. The company uses an older manufacturing process from Taiwan Semiconductor Manufacturing Co., combined with fast SRAM memory on the chips. This design supports quick data access and efficient processing.

Ljubisa Bajic, Taalas’ CEO, says the company has developed a platform that can turn any AI model into custom silicon. He added, “From the moment a previously unseen model is received, it can be realized in hardware in only two months.” This speed could change how AI hardware adapts to new model demands.

AMD’s AI Strategy Grows Stronger

This deal is part of AMD’s broader push into AI hardware. AMD plans to integrate Taalas’ chips and technology with its CPUs and Instinct GPUs. That means future AMD systems could run AI workloads more efficiently by mixing general and specialized processors.

In July 2026, AMD announced a partnership with Cerebras to add Cerebras’ AI chips into AMD systems later this year. This collaboration shows AMD’s interest in diverse AI chip designs, not just traditional GPUs.

AMD has also been busy acquiring AI companies. It bought Silo AI for $665 million in 2024 and ZT Systems for $4.9 billion. Last year, AMD acquired several smaller startups, including MK1. The Taalas deal strengthens AMD’s position in custom AI silicon.

Context in the AI Chip Race

AMD’s move comes just over seven months after Nvidia spent $20 billion buying assets from Groq, a startup also focused on AI inference chips. Nvidia’s dominance in GPUs has boosted its market value to over $5 trillion.

While Nvidia leads in general AI chips, startups like Taalas offer a different path. By customizing chips for specific AI models, they can reach much faster speeds at lower costs. Early demos show Taalas’ chips can process up to 17,000 tokens per second.

This kind of efficiency matters as AI models grow larger and more complex. Custom silicon can deliver faster results without the high power and cost of traditional GPUs. It also means AI hardware can keep up with new models as they appear.

AMD’s acquisition shows it wants to compete in this niche. By adding Taalas’ technology, AMD can offer a wider range of AI hardware options. This variety may help customers build better AI systems in the years ahead.

Stay connected via Google News

Follow us for the latest AI News