The episode reviews recent exits and acquisitions in the AI hardware startup space, highlighting that technical innovation alone doesn’t ensure market success, with factors like ecosystem fit and timing playing crucial roles. It showcases various company outcomes—from shutdowns and talent acquisitions to high-value buyouts—while emphasizing ongoing investor interest and the evolving, competitive nature of the AI hardware market.
The AI Hardware Show 2025 episode focuses on the recent exits and acquisitions of AI hardware startups, highlighting the challenges and successes in this competitive market. The hosts, Ian Cutras and Sal Foxton, discuss several companies that have either been acquired, shut down, or pivoted in the AI chip space. Despite initial skepticism about the survival of many startups, some have managed to persist, while others have exited through acquisitions or closures. The episode emphasizes that having a technically superior chip does not guarantee market success, as ecosystem fit, software support, and timing are critical factors.
Espiranto, founded in 2014, aimed to bring RISC-V architecture into AI inference with a focus on energy efficiency. Despite innovative designs and support for large language models, the company struggled to gain market traction and eventually shut down most operations by mid-2025, with its intellectual property up for sale. Untether, another startup with a bold at-memory compute design, achieved impressive power efficiency and partnered with ARM but failed to secure significant market adoption. Its engineering team and IP were absorbed by AMD in 2025, marking an exit more akin to a talent acquisition than a product success.
Graphcore, a British startup founded in 2016, developed the IPU chip designed to host entire machine learning models in scratchpad memory. Despite raising substantial funding and building a strong software stack, Graphcore faced challenges in ecosystem adoption, with developers and cloud providers favoring established GPU platforms. In 2024, SoftBank acquired Graphcore, likely as a strategic investment in AI hardware innovation rather than for immediate product traction. Similarly, Gray Matter, focusing on event-driven compute to reduce redundant processing, was acquired by Snap in 2023 to integrate its low-power vision and audio processing technology into AR hardware.
Other notable exits include Blaze, which went public via a SPAC in 2025 but has struggled with revenue generation, and Flex Logix, which was acquired by Analog Devices primarily for its embedded FPGA business rather than its AI chip efforts. Amper Computing stands out as a success story, acquired by SoftBank for $6.5 billion after developing power-efficient ARM-based CPUs for AI inference in data centers. Perceive, specializing in ultra-low power AI inference chips for edge devices, was acquired by Amazon in 2024 to enhance on-device AI capabilities in consumer products. South Korean startups Sapion and Rebellions merged to form a national champion, while GreenWaves Technologies, focused on ultra-low power edge AI chips, went into liquidation in early 2025.
The episode concludes with a discussion on the overall state of the AI hardware startup ecosystem, noting that while some companies have exited, many continue to operate and innovate. The trend of acquisitions often reflects the value of engineering talent and IP rather than immediate commercial success. Investor appetite remains strong, particularly for later-stage companies, and the market continues to evolve rapidly. The hosts hint at further analysis in their upcoming podcast, emphasizing the complex interplay between technology, market readiness, and strategic positioning in the AI hardware space.