The News
Nvidia unveiled the RTX Spark laptop superchip at Computex on May 31, then announced the Vera data-center CPU at GTC Taipei on June 1, per Tom's Hardware and Nvidia's own brief. The same week, the company finished folding $15 billion of supplier and customer equity into its position through the Anthropic round.
The View
Drop the words "chip company" from any sentence about Nvidia and you start to see what is happening.
Spark and Vera are not two product launches. They are one move from two angles, and the move is to own the socket Nvidia GPUs plug into. For thirty years Intel and AMD sold the host CPU and Nvidia sold the discrete graphics card. The unwritten rule of the industry was that nobody crossed the line. Last week Nvidia crossed it twice, on consumer and on enterprise, in roughly seventy hours.
The instinct of every CPU vendor watching this is to assume Nvidia will eventually retreat, because the CPU market is too mature, the margins too thin, the supply too commoditized. That instinct misses what the new architecture is for. Spark and Vera are bandwidth-bound, accelerator-adjacent silicon designed specifically to feed Nvidia's own GPUs. The benchmarks against Xeon, Epyc, and Snapdragon are sideshows. The real benchmark is how much Nvidia GPU revenue Nvidia loses today because someone else's CPU bottlenecks the system. That gap is what Vera and Spark are eating.
Add the equity moves to this. Anthropic's $65 billion round last week wrapped in $5 billion from Amazon, plus strategic checks from Micron, Samsung, and SK hynix on the supply side, plus cloud and AI partners on the demand side. Nvidia is doing the same trick in adjacent rounds. The cap table is becoming the contract. Hold equity in your customer, sell them silicon, and the line between "I'm shipping you compute" and "I'm running your business" gets blurry on purpose. That blurry place is where every prior platform monopoly was built. Microsoft on Windows OEMs. Apple on accessory makers. Amazon on third-party sellers.
The throughline is that Nvidia spent the last decade buying its way to the GPU monopoly, and is now spending the next decade taking the dependencies. CPU. CPU again. Memory through investment. Networking through Mellanox already. Storage will be next. Software is already inside the moat; anyone who shipped CUDA into production in the last five years can attest to that. The end state of this strategy is "Nvidia is the platform underneath the AI economy and Intel rents space," not "Nvidia sells better chips than Intel." Those are very different end states for valuation and for antitrust.
Room for Disagreement
The honest counter is that platform plays create their own gravity wells. The hyperscalers (Google with TPUs, Amazon with Trainium, Microsoft with Maia, Meta with MTIA) are all building their own accelerators precisely to avoid being captive to Nvidia. If Nvidia squeezes the socket too hard, the response is a sharper push by the same hyperscalers to develop full in-house silicon stacks, which is the one scenario that breaks the moat. The platform thesis is real, but it has a customer-revolt ceiling Nvidia has to manage.
Notable
- Spark debuts in Microsoft's new Surface Ultra. The first Microsoft flagship to ship on non-Microsoft Arm silicon.
- Vera's early adopters include OpenAI, Anthropic, SpaceX, ByteDance, CoreWeave, and Oracle Cloud Infrastructure.
- Nvidia's data-center revenue hit $75.25 billion last quarter, up 92% year over year.
Bottom Line
This was the week Nvidia stopped being a vendor and started being a platform. If you sell silicon, software, or services into AI workloads, your competitive map needs a redraw.
— Hank