AMD has announced that it has just acquired Toronto-based AI chip company Untether AI. This isn’t the usual corporate acquisition, though, where the bigger firm completely buys the smaller one. Instead, AMD only hired the entire engineering team of Untether, leaving everything else behind. Because of this, its speedAI AI inference processor and imAIgine SDK will no longer be supplied and supported, according to Untether AI’s announcement.
“AMD has entered into a strategic agreement to acquire a talented team of AI hardware and software engineers from Untether AI,” AMD said in a statement to CRN. “The transaction brings a world-class team of engineers to AMD, focusing on advancing the company’s AI compiler and kernel development capabilities as well as enhancing our digital and SoC design, design verification, and product integration capabilities. We are excited to welcome the team’s unique expertise to AMD.”
Untether AI specializes in building AI chips specifically designed for AI inference. While GPUs like Nvidia’s Blackwell Ultra or AMD’s Instinct MI350 excel at training AI models, speedAI performs better in inferencing and is far more energy efficient than these power-hungry GPUs. They achieve that by placing the processors right next to the memory, reducing latency and power consumption.
It’s unclear how many clients Untether AI has and how they will be affected by this change, especially as AMD did not purchase Untether AI’s assets. This meant that companies that bought the latter’s products were left holding the bag.
At the same time, this massive hiring is a sign that AMD is expanding its capabilities to challenge Nvidia in other AI-related fields, not just the raw computing horsepower of AI GPUs. Just one day before this announcement, Team Red also announced its acquisition of Brium, a startup that focuses on AI inference optimization.
These deals point to the company focusing on the next major step in AI development — inferencing. Justin Kinsey, President of semiconductor recruiting firm SBT Industries, said in a LinkedIn post, “AMD’s acquisition of Untether’s engineering group is proof that the GPU vendors know model training is over, and that a decline in GPU revenue is around the corner.” Although this is a bold prediction, Kinsey says that this has been an emerging pattern in the past six months.
As the AI power bill keeps increasing every year, companies will likely soon look for alternatives that are far more efficient. The current AI GPUs that require hundreds of watts are highly suitable for training, but they consume too much power for AI inference. If AMD can deliver a specialized chip just for this application, it could potentially challenge Nvidia’s dominance, especially as the AI industry continues to mature.
Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.
