On June 5th local time, AMD announced the "acquisition" of Untether AI, a Toronto-based AI chip company. However, this is not a regular acquisition. AMD simply acquired the entire engineering team of Untether AI and left everything else behind.
Untether AI also issued an announcement through its official website stating that it has reached a deal with AMD. As a result of the transaction, the speedAI AI Inference Processor and the imAIgine Software Development Kit (SDK) will no longer be provided and supported. “While today marks the end of the Untether AI journey, we are proud of this groundbreaking research that has laid the foundation for our work in advancing state-of-the-art AI chip technology. We thank our team for their dedication and support from our customers, partners and investors. We look forward to our world-class team’s contribution to AMD.”
AMD In a statement to the media, it said: "AMD has reached a strategic agreement to acquire a talented team of AI hardware and software engineers from Untether AI. The deal brings AMD a world-class team of engineers focused on improving the company's AI compiler and kernel development capabilities and enhancing our digital and SoC design, design verification and product integration capabilities. We are pleased to welcome the team's unique expertise to AMD."
Data shows that Untether AI can provide energy-centric AI inference acceleration chips from the edge to the cloud, supporting various types of neural network models. While GPUs like Nvidia’s Blackwell Ultra or AMD’s Instinct MI350 perform well in training AI models, speedAI performs better in reasoning and is more energy-efficient than these power-consuming GPUs. They achieve this by placing the processor next to memory, reducing latency and power consumption.
It is unclear how many customers Untether AI has and how they will be affected by this change, especially if AMD has not purchased assets for Untether AI. This means that companies that buy the latter products can only stand by and watch.
At the same time, AMD's acquisition also shows that it is accelerating its expansion of its AI capabilities in order to better challenge Nvidia in AI-related fields, rather than just the original computing power of AI GPUs.
At present, AI GPUs with power consumption up to hundreds of watts are mainly suitable for training AI models, but using them for AI inference will bring a lot of waste of power consumption. Especially as AI applications enter the terminal side from the cloud, many companies are beginning to look for more efficient AI chips to reason.
Justin Kinsey, president of SBT Industries, a semiconductor recruitment company, said in a LinkedIn post: "AMD's engineering team acquired Untether proves that GPU vendors know that model training is over, GPU A decline in revenue is just around the corner. "While it's a bold prediction, Kinsey said it's a pattern that has emerged in the past six months.
It is worth noting that the day before the disclosure of this acquisition, AMD announced on June 4 local time that it had acquired open source software company Brium to further enhance its AI strength.
AMD said: "The acquisition of Brium strengthens the foundation of our long-term innovation. This reflects our strategic commitment to artificial intelligence, especially to developers who build the future of intelligent applications. This is also the follower of acquisitions of Silo AI, Nod.ai and Mipsology After that, the latest in a series of targeted investments. Together, these investments have enhanced our ability to support the open source software ecosystem and provide optimized performance on AMD hardware. ”
Editor: Xinzhixun-Longkejian
