Intel Poised for Agentic AI Surge, Nvidia Also Benefits
I wrote on $INTC on March 29th as a new idea based on Agentic AI requiring more server microprocessors per gigawatt versus chat-based AI. Though the stock is up over 50% already since then, I think there is more upside over the course of the year in Intel based on new advanced packaging deals with hyper-scalers later this year and foundry agreements early next year. In addition, there is the optimistic scenario of them being more intimately involved in the buildout of TeraFab with Elon Musk. But I believe there is another idea that might be an overlooked beneficiary as well. Custom ASICs seem to be taking share from GPUs with each new announcement by $GOOGL with their TPUs, $AMZN with Tranium and $MSFT with Maia, not to mention companies like OpenAI and Anthropic looking to potentially design their own ASICs. But I believe, $NVDA is also a beneficiary as Agentic AI ramps. When training a LLM, the algorithm is doing the same thing over and over again which was perfect for an ASIC in 2023 and 2024 when that was the majority of AI workloads. As chat-based AI interactions ramped in 2025 overtaking training as the main workload, this was also fairly good for ASICs. Interactions were mostly staying within the LLM being engaged. However, with Agentic AI taking off with the formalization of OpenClaw in late January of 2026, multiple LLM models might be getting accessed for different tasks and many external tools or APIs might be engaged. In this case, with the server microprocessor orchestrating these different actions, it is hard to beat Nvidia for the complex reasoning needed compared to a specialized ASIC. I think the recent deal between Anthropic and Coreweave which runs on Nvidia GPUs is a tangible proof point in this regard. Google and then Amazon invested in Anthropic in 2023. As a result, Anthropic’s primary training and inference workloads have historically been driven by custom silicon. Anthropic is the fastest growing at scale AI company in the world having gone from $9B in run-rate revenues at the end of 2025 to $19B by early March of 2026 to $30B by early April. The deal signed on April 10, 2026, between Anthropic and Coreweave represents a strategic pivot. It isn't Anthropic's first use of GPUs, but it is their first major move to treat NVIDIA GPUs as a third production pillar outside of the two big tech hyperscalers. The CoreWeave agreement captures the first volume shipments of Nvidia’s Vera Rubin GPUs in the second half of 2026. To be fair, this does follow Nvidia’s $2B investment into Anthropic earlier this year. But Nvidia’s ownership of Anthropic is still in the low single digits compared with Amazon and Google which are both estimated to be in the low double digit to mid-teens range including warrants. As for Nvidia’s stock, it has roughly traded sideways since the end of July last year and has slightly underperformed the S&P and Nasdaq. But reported quarterly earnings since then have gone from $44.1B in revenue with EPS of $0.96 to $68.1B (+54%) with EPS of $1.62 (+69%) most recently. As a result of this earnings ramp, Nvidia’s current PE is an undemanding 24x CY26 versus the S&P of 21x with the agentic ramp having just been kick started by OpenClaw earlier this year. I see upcoming earnings as a result being strong. While the multiple could always continue to compress given legitimate concerns around circular financing and OpenAIs spending plans among other things, Agentic AI token generation is 10-100x that of Chat-based AI. I see Nvidia as an under-appreciated way (and yes it sounds ridiculous to say that about Nvidia to some degree) to play that theme.