At some point towards the end of last year, Nvidia discreetly redrew the map of its investment portfolio without any of the fanfare that typically accompanies a significant announcement. No leather-jacketed reveal on a stage in San Jose, no keynote slides at GTC. Just a February 13F filing that required analysts to double-check their reading by scrolling through the line items.
The majority of the story is revealed by the numbers themselves. With a stake in Intel that now makes up more than 60% of the entire book, Nvidia’s disclosed portfolio grew from about $3.8 billion to over $13.1 billion in just one quarter. On paper, this type of allocation appears to be almost reckless. It reads more like a thesis in actuality. Jensen Huang appears to be making a specific statement regarding who and where the next tier of AI infrastructure is constructed.
The most obvious surprise is Intel. For many years, Nvidia and Intel’s relationship was like that of two long-standing rivals stuck in the same trade show: courteous in public but more acerbic in private. Currently, Nvidia has about 215 million shares, which are worth almost $7.9 billion. Squinting is more necessary for the investment than creativity. When your product roadmap depends on someone, somewhere, producing chips outside of Taiwan, Intel’s foundry aspirations, U.S. factories, and CPU footprint inside hyperscaler racks suddenly become more important.
Then there is Synopsys, a chip-design software company that most people outside of the semiconductor industry couldn’t choose from a lineup. On December 1, 2025, Nvidia announced an expanded partnership and invested about $2 billion in it. It’s a subtle action with significant ramifications. AI-assisted chip design is emerging as a distinct field, and Nvidia is now financially dependent on the company that creates the tools that its rivals use. Reading between the lines gives the impression that returns aren’t the main focus of this.
Strangely, and perhaps most intriguingly, is the Nokia stake. A joint venture in AI-RAN—radio access networks based on Nvidia’s CUDA stack and a new Arc Aerial RAN Computer—is directly linked to a $1 billion purchase at about $6.01 per share. In the past, telecom infrastructure has proven to be a dead end for aspirational tech investors. However, the AI-RAN market, which is expected to reach $200 billion by 2030, is one of those projections that seems both exaggerated and underestimated. Cell towers serve as nodes for edge inference. Five years ago, this concept seemed absurd.
What was removed is nearly as instructive. Arm Holdings is no more. Applied Digital is no more. WeRide and Recursion Pharmaceuticals are both completely gone. These were hedges against unrealized futures, experimental positions. Eliminating them releases funds and, more significantly, makes it clear that Nvidia is done diversifying for the sake of optionality. Although Nvidia actually almost doubled its share count there in January, CoreWeave, the once-crown jewel of the portfolio, fell from over 86% of weight to about 13%. The dilution is not strategic, but rather mathematical.
The new portfolio feels so focused that it’s difficult to ignore. Five properties. One of them is much larger than the others. This would have been interpreted as overconfidence a few years ago. It now appears to be a business with enough cash flow to easily finance the ecosystem in which it wishes to operate. It’s still very much up in the air whether that wager will pay off, whether Intel executes, whether 6G truly happens, and whether AI-native networks are more than just a slide deck. However, the direction of travel is no longer subtle.
