AMD still lags behind Navidia in the race for AI performance dominance.
AMD is actively recruiting to close the performance gap between Instinct and Blackwell GPUs
Brium's compiler expertise can help AMD accelerate inference without relying on hardware-specific dependencies
Untether AI team joins AMD, but existing customers are left without product support
AMD's recent moves in AI have all been centered around strategic acquisitions designed to strengthen its position in a market largely dominated by Nvidia.
These include the acquisitions of Brium, Silo AI, Nod.ai, and the engineering teams at Untether AI, each designed to bolster AMD's AI software, inference optimization, and chip design capabilities.
The goal is clear: close the performance and ecosystem gap between AMD Instinct GPUs and Nvidia's Blackwell line.
Smart acquisitions in a competitive ecosystem
Acquisition of Brium to optimize AI
AMD described the acquisition of Brium as a key step in strengthening its AI software capabilities.
The company said Brium brings advanced software capabilities that improve the ability to deliver highly optimized AI solutions. Brium's strengths lie in compiler technology and end-to-end AI inference optimization, areas that are critical to achieving better out-of-the-box performance and reducing dependence on specific hardware configurations. Integrating Brium will affect multiple projects, including OpenAI Triton and SHARK/IREE, which are seen as improving AMD's inference and training capabilities. The use of precision formats such as MX FP4 and FP6 suggests that the strategy is to squeeze more performance out of existing hardware. But Nvidia has made similar moves and maintained a lead in raw processing power and software maturity, highlighting that AMD is still playing catch-up rather than leading in the AI software ecosystem.
Acquisition of the Untether Team
Another notable move was AMD's acquisition of the entire engineering team at Canadian startup Untether AI. Untether AI is known for its energy-efficient inference processors, and AMD's acquisition of talent rather than the company itself left Untether products unsupported. The company confirmed that the move is focused on compiler and kernel development, as well as SoC design, marking a strong push into inference-specific technologies. These technologies become increasingly important as training-based GPU revenue faces a potential decline. Justin Kinsey, president of SBT Industries, said this proves that GPU vendors are aware of the end of model training and the downward trend in revenue. Although it may be exaggerated, it reflects the industry sentiment: energy efficiency and inference performance are the next frontier. AMD is optimistic about opening up its AI software platform, but there are doubts about whether it can match Nvidia's hardware and CUDA software integration. Ultimately, Nvidia maintains a lead in hardware efficiency and software ecosystem, and these acquisitions may bring AMD closer to the goal, but Nvidia's Blackwell remains the benchmark GPU for AI workloads.
Source: Content from techradar
