AMD Acquires AI Chip Startup to Embed Models Directly in Silicon
AMD snaps up a startup that bakes AI models straight into chip hardware, a move that could reshape how AI inference gets done.
AMD just made a bold bet on the future of AI hardware by acquiring a chip startup called Taalas, which has figured out how to hardwire AI models directly into silicon. Instead of running AI software on top of general-purpose chips — the way most systems work today — Taalas takes a fundamentally different approach: the model lives inside the chip itself. That's a pretty big deal if you care about speed and efficiency.
Right now, Taalas' chip runs a compact version of Meta's Llama 3.1, one of the more popular open-source large language models out there. Think of it like burning a book into the walls of a library rather than just shelving it — the information is always right there, no searching required. The company is also working on chips capable of handling larger, more sophisticated models, which suggests AMD is playing a long game here.
Read more OpenAI Moves to Dismiss Apple Trade Secret Theft Lawsuit →
For AMD, this acquisition slots into a broader push to compete more aggressively with Nvidia in the AI chip space. Embedding AI models directly into hardware could dramatically cut latency and power consumption, two pain points that data centers and edge devices constantly wrestle with. If Taalas' approach scales well, it could give AMD a genuinely differentiated product rather than just another GPU alternative.
For everyday tech watchers, the takeaway is simple: the AI chip wars are heating up fast, and companies are exploring every angle — including literally building the AI into the chip itself. Whether this hardwired approach becomes mainstream or remains a niche solution depends a lot on how flexible and upgradable these chips turn out to be as AI models keep evolving at breakneck speed.
Continue reading at US Top News and Analysis
