\n<\/p>\n
\n\t\tOpinions expressed by Entrepreneur contributors are their own.\t<\/p>\n
Every founder I talk to has an AI application idea. Almost none of them are thinking about what those applications actually run on. <\/p>\n
That\u2019s the gap I\u2019ve been investing in for the past two years \u2014 and 2026 has made the case louder than I expected.<\/p>\n
The early AI chip narrative was all about training \u2014 who could build the biggest cluster to teach the biggest model. That race increasingly has one obvious winner, and it isn\u2019t a startup. The far more interesting, and far less crowded, race is inference: running trained models fast, cheaply and at scale, in production, for actual paying customers. That\u2019s an efficiency problem, not a brute-force one, and efficiency problems are where specialized architecture beats general-purpose hardware.<\/p>\n
That\u2019s the thesis behind my largest position, Cerebras Systems. I backed Andrew Feldman\u2019s team in a very early round, when the company was valued at around $2 billion \u2014 well before its Series F pushed past $4 billion, before the $1 billion Series H that valued it at roughly $23 billion in February, and long before what came next. <\/p>\n
That conviction was validated in the biggest way possible in May, when Cerebras went public on the Nasdaq under the ticker CBRS, pricing its IPO at $185 a share, raising $5.55 billion \u2014 the largest U.S. tech IPO in years \u2014 and popping 68% on debut to a market cap near $95 billion<\/a>. Wafer-scale compute solved the inference latency problem in a way incremental GPU improvements couldn\u2019t, and the public market has now agreed, emphatically.<\/p>\n Groq is the clearest evidence that inference, not training, is where the value is settling. Seven months after signing a $20 billion chip licensing deal<\/a> with Nvidia, Groq raised another $650 million \u2014 not to keep competing on training hardware, but to double down on its inference-focused AI cloud business. A company that could have taken Nvidia\u2019s money and exited chose instead to re-commit to inference. That\u2019s a signal worth paying attention to.<\/p>\n Positron is the newest proof point. After raising $230 million<\/a> in a Series B backed by Arm and the Qatar Investment Authority in February, the company\u2019s valuation \u201cskyrocketed\u201d in a follow-on round announced in September, with $875 million raised to keep building inference-focused chiplet architecture aimed squarely at Nvidia\u2019s core business. Capital is moving fast into this category because the opportunity window is real, not theoretical.<\/p>\n d-Matrix rounds out my thesis at the earlier stage: A $275 million<\/a> Series C last November valued the company at $2 billion, built specifically around digital in-memory compute for inference workloads.<\/p>\n Tenstorrent<\/a>, the AI chip company built by legendary architect Jim Keller, just delivered the loudest proof point yet. After raising $693 million<\/a> in a Series D at roughly a $2.6 billion valuation in December 2024, the company closed two Series E tranches in August 2026 totaling about $1.4 billion, pushing its valuation to $5.76 billion<\/a>. That kind of markup, this fast, only happens when the market is convinced the specialized-silicon thesis is right \u2014 and I\u2019m glad to have been in early.<\/p>\n I didn\u2019t get into these positions because I\u2019m a semiconductor expert. I got into them because I run an operating business that increasingly depends on AI tools, and I asked a founder\u2019s question: Who actually captures the value when every company on earth needs cheaper, faster inference? <\/p>\n The answer wasn\u2019t not only the app layer, but also the infrastructure underneath it \u2014 built by teams willing to bet on specialized architecture instead of chasing the incumbent\u2019s playbook.<\/p>\n If you\u2019re building anything AI-adjacent right now, don\u2019t just ask which model to use. Ask who\u2019s building the rails those models will run on for the next decade. That\u2019s usually where the durable value \u2014 and the durable investment opportunity \u2014 actually sits. <\/p>\n The application layer will keep churning through winners and losers as models commoditize, but the infrastructure underneath it \u2014 the wafers, the interconnects, the inference stacks \u2014 compounds quietly in the background. I\u2019d rather own the toll roads than bet on which car wins the race.<\/p>\n Disclosure: I hold investments in Cerebras, Groq, Positron, d-Matrix, and Tenstorrent, companies discussed in this article, and therefore have a financial interest in their performance. These interests represent a potential conflict of interest. The views expressed are my own and are provided for informational purposes only, not as investment advice or a recommendation to buy or sell any security.<\/em><\/p>\n<\/p><\/div>\n Every founder I talk to has an AI application idea. Almost none of them are thinking about what those applications actually run on. <\/p>\n That\u2019s the gap I\u2019ve been investing in for the past two years \u2014 and 2026 has made the case louder than I expected.<\/p>\n The early AI chip narrative was all about training \u2014 who could build the biggest cluster to teach the biggest model. That race increasingly has one obvious winner, and it isn\u2019t a startup. The far more interesting, and far less crowded, race is inference: running trained models fast, cheaply and at scale, in production, for actual paying customers. That\u2019s an efficiency problem, not a brute-force one, and efficiency problems are where specialized architecture beats general-purpose hardware.<\/p>\n<\/p><\/div>\nThe pivot that proved the thesis<\/h2>\n
The lesson for founders, not just investors<\/h2>\n
Key Takeaways<\/h2>\n
\n
Training was the first race. Inference is the real one.<\/h2>\n