AMD Acquires Untether AI Team to Boost Energy-Efficient AI Chip Development and Challenge Nvidia

AMD Acquires Team Behind AI Chip Startup Untether AI

AMD has confirmed the acquisition of the engineering team from Untether AI, a startup specializing in AI inference chips. These chips are known for delivering faster performance and greater energy efficiency compared to competitors, targeting edge environments and enterprise data centers.

An AMD spokesperson stated, “AMD has entered into a strategic agreement to acquire a talented team of AI hardware and software engineers from Untether AI. This team will enhance AMD’s AI compiler and kernel development, digital and SoC design, verification, and product integration capabilities.” Financial details of the deal were not disclosed.

What This Means for Untether AI

Untether AI’s executive, Bob Beachler, announced that the startup will cease supplying and supporting its speedAI products and imAIgine software development kit following the acquisition. He expressed pride in the startup’s contributions to AI chip technology and gratitude toward their team, customers, partners, and investors. The team will continue their work within AMD.

Context of AMD’s AI Strategy

This acquisition is part of AMD’s broader effort to expand its AI computing capabilities and compete with Nvidia’s stronghold in the sector. Just a day prior, AMD acquired Brium, a compiler startup focused on optimizing AI performance for AMD’s Instinct data center GPUs.

Background on Untether AI

Founded in 2018 and based in Toronto, Untether AI developed AI inference chips with an “at-memory” architecture that effectively reduces data traffic within the chip. This design improves throughput and energy efficiency, critical for both edge and data center applications.

Untether AI’s speedAI240 Slim AI inference accelerator card, launched last October, demonstrated up to six times greater energy efficiency in closed edge environments and three times greater efficiency for data centers. It also delivered the fastest performance among single PCIe cards for the ResNet-50 image classification benchmark, according to peer-reviewed MLPerf results.

Partnerships and Market Adoption

The speedAI240 card, available in a 75-watt PCIe design, had been adopted by companies such as J-Squared Technologies in the U.S. and Ola-Krutrim in India. Untether AI also collaborated with several semiconductor and solution providers including Ampere Computing, Arm, NeuReality, Boston, Asa Computers, and Vertical Data.

According to Untether AI’s former CEO Chris Walker, who left the company earlier this year, there is strong demand for chips that offer high performance without the energy demands of traditional GPUs. Walker highlighted the need for efficient processors in power-constrained small data centers and industrial edge applications, where Nvidia’s GPUs can push racks to 120 kilowatts.

Looking Ahead

AMD’s acquisition of Untether AI’s engineering team strengthens its position in AI hardware development, particularly for energy-efficient inference chips suitable for edge and data center use cases. This move aligns with the company’s ongoing investments to enhance AI performance across its product lineup.

For those interested in building expertise in AI technologies and hardware, exploring relevant AI training courses can provide practical skills to keep up with advances in this fast-changing field.