Nvidia bets $2B on Marvell
Nvidia committed $2 billion to Marvell and is pushing NVLink Fusion as an "AI control layer," signaling a shift toward proprietary interconnects that could lock customers into mixed‑vendor hardware topologies. That makes hardware‑agnostic serving stacks and silicon abstraction approaches increasingly important for enterprise ML infra. (digitimes.com)
NVIDIA and Marvell issued matching press releases on March 31, 2026 announcing a strategic partnership to link Marvell into NVIDIA’s rack‑scale interconnect platform and to collaborate on silicon photonics technology. (investor.nvidia.com) Marvell’s release says it will provide custom XPUs and NVLink‑Fusion‑compatible scale‑up networking while NVIDIA listed Vera CPUs, ConnectX NICs, BlueField DPUs, NVLink interconnect and Spectrum‑X switches as components of the joint rack designs. (markets.businessinsider.com) Marvell completed its acquisition of Celestial AI on February 2, 2026, importing Celestial’s Photonic Fabric optical interconnect IP that Marvell expects to begin contributing revenue in the second half of fiscal 2028 and to reach a $500 million annualized run rate by the fourth quarter with a $1 billion run rate target by fiscal 2029’s fourth quarter. (investor.marvell.com) Market reaction was immediate: Marvell shares climbed into the single‑digit-to‑double‑digit percentage range intraday (reports show moves of ~7–13%), Bank of America raised its Marvell price target to $125, and NVIDIA shares traded up roughly 3% on the announcement. (247wallst.com) Technology outlets described the pact as positioning Marvell as a semi‑custom XPU and interconnect supplier inside NVIDIA’s rack‑scale ecosystem, with coverage noting the tie-up goes “beyond NVLink Fusion” into deeper systems integration. (datacenterknowledge.com) NVIDIA’s statement frames the collaboration as “fully compatible with NVIDIA AI infrastructure” and says the work will give customers “greater choice and flexibility” when developing next‑generation infrastructure on NVIDIA architectures. (investor.nvidia.com)
