In this episode of the AI Hardware Podcast, hosts Sally Ward Foxton and Ian Cutress discuss various AI hardware companies that have recently exited the market, either through acquisitions, IPOs, or closures. They reflect on the expectations around exits in the AI hardware space, noting that while many anticipated more mergers and acquisitions, the reality has been a mix of company failures, team acquisitions, and a few positive exits. The episode covers a range of companies, from early-stage startups to more established players, highlighting the challenges and dynamics of the AI hardware industry.
The discussion begins with companies like Espiranto and Untether, both of which faced difficulties leading to their closures or partial acquisitions. Espiranto, a RISC-V startup focused on data center chips, shut down after losing most of its engineering team and struggling to find product-market fit, particularly as hyperscalers developed their own chips. Untether, which had promising technology and a roadmap including chiplets for edge devices, ran out of funding and was acquired by AMD in a deal that mainly involved hiring the engineering team rather than acquiring the technology itself. These cases illustrate the tough market conditions and the importance of timing and funding in the AI hardware space.
The episode also covers more successful exits, such as Graphcore’s acquisition by SoftBank and Snap’s acquisition of the French neuromorphic startup Gray Matter. Graphcore, a UK-based AI chip company, was acquired for roughly the amount it had raised, allowing investors to break even but leaving employees with little gain. Despite some office closures, Graphcore is still hiring and developing new chips under SoftBank’s ownership. Gray Matter’s technology fits well with Snap’s augmented reality glasses, highlighting how specialized AI hardware can find a home in consumer devices. Other companies like Blaze (formerly ThinkEye) went public via a SPAC, though with modest hardware sales, reflecting the challenges of scaling in niche markets like security cameras and drones.
Further, the hosts discuss Flex Logix’s acquisition by Analog Devices, which was a natural fit given Analog Devices was an early customer of Flex Logix’s embedded FPGA technology. Ampere Computing, an ARM-based data center CPU company, was acquired by SoftBank at a valuation significantly above its total funding, marking a relatively successful exit. The episode touches on some industry speculation and conspiracy theories around these acquisitions, particularly regarding SoftBank’s strategic interests and ARM’s royalty revenues. Additionally, Amazon’s acquisition of an ultra-low-power edge AI chip company and the merger of two South Korean startups, Sapion and Rebellions, demonstrate ongoing consolidation and specialization in the edge AI market.
Finally, the podcast covers recent developments such as Ambiq’s IPO, the liquidation of European RISC-V edge chip company GreenWaves, and Meta’s acquisition of stealthy AI chip startup Rivos. Rivos, which had been secretive but growing rapidly, was acquired by Meta reportedly for around $2 billion, signaling Meta’s continued investment in custom AI hardware for its massive recommendation and AI workloads. The hosts conclude by reflecting on the future of AI hardware startups, noting that once companies reach a certain size, acquisitions become less likely, and IPOs or independent growth are the main paths forward. They also hint at upcoming episodes focusing on automotive AI chips and other emerging trends in the industry.
