Jensen Huang ties Vera CPUs to SK Hynix DRAM, signaling a big 2025 push

Nvidia CEO Jensen Huang just put a very specific partner bet on the record: Nvidia’s new Vera central processing units will use SK Hynix’s DRAM. Huang said this during a reporter stop outside a Seoul restaurant on Sunday, where he also had dinner with SK Group Chairman Chey Tae-won, SK Hynix CEO Kwak Noh-Jung, and executives from SK Telecom Co.

Huang didn’t just name the memory supplier. He also framed the timing as aggressive: “We had a very big year this year with SK Hynix, and we are preparing for a very, very large second half of the year and next year.” That line is a quiet signal to the whole AI hardware ecosystem that Nvidia is treating the next two quarters and beyond as more than business as usual. If you are an operator planning capex, a board reviewing vendor concentration risk, or an investor tracking AI infrastructure margins, the practical question becomes: which memory supplier relationship gets deepened first as demand tightens and production schedules get booked.

To understand why this is a big deal, it helps to remember what Vera actually is. Huang said, “We introduced Vera CPU, which is a revolutionary CPU.” Vera is Nvidia’s first standalone data center microprocessor, built to compete directly with Intel Corp.’s Xeon line and Advanced Micro Devices Inc.’s Epyc chips, as well as with in-house programs at large-scale operators like Amazon.com Inc.’s Graviton. In other words, Nvidia is not only shipping AI acceleration. It is building a broader compute stack, including the CPU layer that sits beside accelerators and helps determine system-level performance, cost, and deployment flexibility.

Then comes the memory piece. Huang said Vera CPU will “also use SK Hynix’s DRAM.” In data centers, memory availability and cost can become a bottleneck as systems scale. DRAM does not exist in a vacuum; it is part of a supply chain where lead times, yield, and capacity planning matter as much as chip design. When a company like Nvidia explicitly ties a flagship CPU roadmap to a named memory supplier, it is effectively stating that SK Hynix is in the critical path for future builds. That matters for competition because CPU platform transitions are hard. They require software readiness, validation, and hardware integration. Partner choices like this can speed execution by reducing uncertainty during ramp.

Huang’s trip underscores that this is not a one-off technical announcement. He arrived in South Korea on Friday to visit partners and suppliers and is scheduled to meet Samsung Electronics Co. Vice Chairman Jun Young-hyun. He is also expected to meet the heads of Hyundai Motor Group and LG Group among other business leaders on Monday. That schedule reads like a classic supplier-network play: secure components, coordinate manufacturing realities, and keep relationships close while the AI market demands more than just chips. For decision-makers, the message is that Nvidia is building a whole ecosystem, not simply negotiating individual purchase orders.

The story also hints at where the compute demand will travel next. Huang said he’s having discussions with telecommunication companies because “telco networks will be used for AI in the future.” This is important because it points to deployment patterns. Telcos historically think in terms of network infrastructure and managed services, which means AI workload placement will not only live in hyperscale cloud data centers. It can also spread into edge and regional computing environments where system design, memory supply, and vendor partnerships affect total project feasibility.

Finally, consider the competitive and governance implications inside the broader AI hardware race. Vera’s stated goal is head-to-head competition with Intel and AMD and with operator-built alternatives like Amazon’s Graviton. In that kind of race, memory partner depth can influence how quickly Nvidia can translate a “revolutionary CPU” into repeatable deployments at scale. Boards and procurement teams tend to worry about concentration risk, but they also need supply certainty. When Huang says Nvidia had a “very big year” with SK Hynix and is gearing up for “a very, very large second half” and “next year,” it suggests Nvidia sees value in deepening that supply relationship at the moment when AI compute demand is most likely to stress the supply chain.