In a significant strategic shift that reflects the increasingly competitive landscape of AI hardware, Advanced Micro Devices (AMD) has acquired the engineering team of Untether AI, a Toronto-based startup specializing in AI inference acceleration. This move, which includes the absorption of Untether AI’s core staff but not its physical products or IP portfolio, signals a shift in AMD’s AI posture to deepen its in-house AI processing capabilities. The news, originally reported in Tom’s Hardware (2025), also confirmed that Untether AI will cease product support moving forward, sending ripples across the AI chip industry and raising questions about future competition, AI inference development, and market implications.
Why the Untether AI Acquisition Matters in 2025
Untether AI, founded in 2018, was a promising contender in the AI hardware landscape. The company developed “at-memory” computing technology, which drastically improved energy efficiency and throughput for AI inference workloads by reducing the need to shuttle data between processing and memory units. This neuromorphic approach aligned with low-latency AI applications, especially in edge computing and data centers. Though innovative, the company struggled to successfully commercialize at scale despite raising over $100 million USD in capital by 2022 (VentureBeat, 2022).
By 2024, industry insiders had begun hinting at Untether AI’s financial difficulties and its limited ability to compete with dominant players such as NVIDIA and Intel in the inference chip space (AI Trends, 2024). The inability to establish significant traction in cloud computing or enterprise sectors meant that its otherwise breakthrough technology couldn’t overcome commercialization and integration hurdles. AMD’s decision to bring Untether AI’s team into its fold while ending support for existing Untether products underscores a strategic pivot — focusing not on rebranding and selling Untether AI’s chips, but rather using their expertise to turbocharge AMD’s own AI roadmap.
Strategic Integration into AMD’s AI Roadmap
Unlike NVIDIA, which has led the AI space with its CUDA ecosystem and dominance in training workloads, AMD has historically lagged in building a robust AI hardware identity. However, under the architectural guidance of the CDNA and RDNA platforms and aggressive software modernization through ROCm, AMD is rapidly scaling. The integration of Untether AI’s engineers — many with cutting-edge R&D experience in memory-intensive inferencing — comes at a crucial time as AMD plans to integrate enhanced AI acceleration into its Instinct MI300X platform and future EPYC server chips.
The chiplet-based approach AMD uses allows modular integration of specialized processing units. This complements Untether AI’s approach to fine-grained, memory-centric inference computing. Combining AMD’s chiplet ecosystem with compact, data-local processing designs positions the company to challenge Intel’s Gaudi chips and NVIDIA’s Tensor Cores in low-power edge and on-premise inference markets.
Meanwhile, AMD CEO Dr. Lisa Su, during the company’s Q1 2025 earnings call, hinted at accelerating deployment of custom inference engines, emphasizing hardware-software co-design as a future proofing strategy (CNBC Markets, 2025). Merging Untether’s high-efficiency inference design mindset into that execution plan could enhance AMD’s offensive in AI infrastructures ranging from autonomous automotive to private data centers and sovereign AI deployments.
Implications for the AI Semiconductor Market
This acquisition aligns with a broader pattern observed in 2025 — tech giants are racing to acquire AI-specialized talent pools rather than full-scale startups with product lines. Recent moves by Intel (including a partial stake sale in its Mobileye unit) and the AI Alliances forming between AWS, Oracle, NVIDIA, and startups like Groq and SambaNova have emphasized vertical integration and talent absorption as strategic priorities (MIT Technology Review, 2025).
AMD’s decision not to assume the maintenance of Untether AI’s hardware has operational ramifications. Customers using those chips — concentrated in research clusters and specialized datacenters — will be forced to seek alternatives. That creates a vacuum for inference solutions under 10W TDP, a space previously ideal for Untether AI’s RunAI200 series accelerators. This could temporarily benefit startups like Tenstorrent or British AI chipmaker Graphcore, though survival has also proven tough for both (DeepMind Blog, 2025).
Table: Landscape of Major AI Chipmakers Post-Untether Acquisition
| Company | 2025 AI Focus | Recent Strategic Action |
|---|---|---|
| AMD | Inference accelerators, High-bandwidth memory AI | Acquired Untether AI team |
| NVIDIA | Training & Inferencing, Full-stack AI platform | Launched Blackwell GPU, Opened CUDA 14 (2025) |
| Intel | Multi-chip inference, CPU-integrated AI | Spinning off parts of Gaudi & Mobileye for strategic capital |
| Google TPU | AI SaaS and hardware acceleration | TPU v6 for Gemini 2 (early 2025) |
This market rationalization also reflects the capital-intensiveness of competing in AI hardware. According to McKinsey Global Institute (2025), developing and deploying a next-generation AI inference chip costs upwards of $450 million USD, making independent survival extremely rare. Consequently, the exit of Untether AI as a product entity validates McKinsey’s projection that less than 5% of AI processor startups founded between 2016 and 2022 will remain independent by the end of 2025.
What Users of Untether AI Products Should Do
One of the major concerns emerging from the transition is the sudden discontinuation of product support. For institutions and companies that have integrated Untether AI chips into their hardware stack, there’s no straightforward migration map. Given the specialized nature of Untether’s architecture — requiring bespoke software compilation and APIs — enterprises face difficulties replacing the chips without total reengineering of workloads (Kaggle Blog, 2025).
AMD has not signaled any intention to support legacy Untether chips or offer equivalent hardware-as-a-service tokens. Therefore, third-party integrators and academic institutions may need to pivot to alternatives such as NVIDIA’s small-form factor Jetson solutions or AMD’s own Versal Adaptive SoCs from its Xilinx acquisition.
The case exemplifies how fragile long-term infrastructure plans are in an era of rapid AI acquisitions. FTC statements as of mid-2025 show increasing scrutiny of tech mergers, particularly where hardware support tangibly impacts AI accessibility or usage rights for smaller players.
Conclusion: Talent Over Products in 2025’s AI Race
AMD’s acquisition of Untether AI’s team not only ends the Toronto company’s commercial journey but redefines how value is being created in the AI chip market. More than ever, the actual physical products — even those with innovative IP — are less important than the human expertise behind them. This is a marked divergence from past decades in semiconductors where hardware dominance was tightly linked to intellectual property ownership and proprietary fabrication capabilities.
As 2025 unfolds, companies are optimizing for advanced AI heuristics, inference, and training through human capital. This mirrors patterns OpenAI and Google DeepMind have observed — human breakthroughs in AI often outpace hardware adoption cycles (OpenAI Blog, 2025), suggesting engineering minds may now be more valuable than devices themselves.
by Alphonse G
Inspired by and based on the original reporting: Tom’s Hardware, 2025
APA References:
Tom’s Hardware. (2025). AMD scoops entire Untether AI chip team; Canada AI inference outfit will cease product support. https://www.tomshardware.com/tech-industry/amd-scoops-entire-untether-ai-chip-team-canada-ai-inference-outfit-will-cease-product-support
VentureBeat. (2022). Untether AI raises $100M for neuromorphic AI chips. https://venturebeat.com/ai/untether-ai-raises-100m-for-neuromorphic-ai-chips/
AI Trends. (2024). NVIDIA sets the pace in AI chips market. https://www.aitrends.com/machine-learning/nvidia-sets-pace-in-ai-chips-market/
MIT Technology Review. (2025). AI-focused tech battles. https://www.technologyreview.com/topic/artificial-intelligence/
DeepMind Blog. (2025). Consolidations in AI chip innovation. https://www.deepmind.com/blog
McKinsey Global Institute. (2025). Capturing value in next-gen inference computing. https://www.mckinsey.com/mgi
OpenAI. (2025). AI hardware still lags behind conceptual innovation. https://openai.com/blog/
Kaggle Blog. (2025). Architecture lock-in and AI migration strategies. https://www.kaggle.com/blog
FTC. (2025). FTC increases oversight of mergers in AI semiconductor space. https://www.ftc.gov/news-events/news/press-releases
CNBC Markets. (2025). AMD signals ramp-up of AI engine deployment. https://www.cnbc.com/markets/
Note that some references may no longer be available at the time of your reading due to page moves or expirations of source articles.