MIPS takes Wave to the edge

Eyeing a move from training to endpoints, Wave Computing has acquired MIPS Tech, a pioneer in the development of RISC processors. The target, recently spun off of Imagination Technologies, provides the buyer, a designer of artificial intelligence (AI) semiconductors, the chance to sell its wares more broadly as organizations look to run AI algorithms directly on the endpoint.

Founded in 2010, Wave is one of multiple startups producing accelerators for AI workloads, and one of a smaller select group (along with Cambrian Systems, Cerebras Systems, Graphcore and Horizon Robotics) that have already raised over $100m in venture funding. Wave emerged from stealth in 2016 and made its compute appliance for training neural networks available to early-access customers in 2017. In March, Wave said it would be using the MIPS core as the integrated CPU within its next-generation Dataflow Processing Unit.

We noted in a previous report that MIPS would likely prove a valuable asset for AI applications. It’s been in business since the 1980s and has a significant embedded systems user base and a range of extensible cores. Once owned by SGI, MIPS ended up as part of UK-based GPU maker Imagination Technologies, but when Imagination was sold to China-based investor Canyon Bridge Capital Partners, US-based MIPS had to be divested separately to Tallwood Venture Capital for $65m. Tallwood is also an investor in Wave.

It’s likely that the power-efficient MIPS cores will be useful for the development of inference processors within edge devices, giving Wave an end-to-end story beyond its initial training focus, increasing its potential total addressable market significantly. MIPS will continue to be run as an independent unit, and will continue licensing its cores to third parties. 451 Research’s recent VotE: Internet of Things report shows that many companies are already making advanced calculations right on the endpoint – 40% of respondents claimed to run data analysis, cognitive computing or AI at the network edge or perimeter.

For more real-time information on tech M&A, follow us on Twitter @451TechMnA.