Wells Fargo Keeps $185 PT on Nvidia
Wells Fargo analyst Aaron Rakers reiterated on May 20, 2025, a ‘Overweight’ rating on Nvidia. The price target of $185 was reaffirmed.
“NVDA’s NVLink Fusion: A Deeper Look-Escalating the Competitive Battleground in Scale-Up Interconnect (vs UALink)
NVIDIA’s NVLink Fusion announcement coming out of this week’s Computex 2025 has been considered the most significant new news.
We thought it would be useful to take a quick / deeper look.
Our Call
While at the surface, NVLink Fusion has left some to consider NVDA’s move as a defensive competitive response vs UALink, we think this technology/announcement could extend the reach of NVIDIA’s NVLink technology into the increasingly important custom AI ASIC market, and in the process, support growth in NVIDIA Networking for scale-up capabilities.
UALink efforts should continue to be a competitive focus; however, we still think catching up to NVLink’s lead/scale will take time.
What Was Announced?
Earlier this week, NVDA announced NVLink Fusion, which allows 3rd-party custom silicon (ASICS; most notably, internally developed hyperscale silicon) and CPUs to integrate NVIDIA’s memory coherent NVLink for chip-to-chip interconnect-NVLink CZC.
Some industry reports have highlighted that the CPU connectivity isn’t necessarily new, as NVDA had announced NVLink-C2C interconnect support for custom dies at GTC 2022 (see here).
NVLink-A Foundational Technology for Al Infrastructure
While NVDA’s data center GPU intros/product cycles (Ampere-to-Hopper-to-Blackwell-to-Rubin) garner the most attention, we have long believed NVLink interconnect capabilities could be considered as important, as we see the proliferation of rack-scale / full-stack integration, e.g., see our Jun 24 report: A Deep Dive into ‘Scale-Up’ Al Interconnects NVDA’s NVLink Differentiation; An Evolving Competitive Battleground (UALink).
NVLink interconnect provides significant bandwidth advantages vs. industry standard PCle connectivity, eg., NVLink v5 provides 1.8TB/s bandwidth (18x lanes @ 100GB/s); >14x vs.
PCIe Gen5 bandwidth.
NVLink v6 is expected to release in 2H26 w/ Rubin R200 GPUs @ 3.6TB/s.
NVDA’s push to externalize NVLink brings us back to when Broadcom emphasized the need to accelerate the roadmap of PCle.
NVLink was the driving force behind industry open-source efforts such as Ultra Accelerator Link (UALink).
Opening Up NVLink via Chiplets
NVLink Fusion (diagram below) enables NVDA’s Grace / Vera (Arm CPUs) to connect to custom accelerators (initial partners-Marvell, Alchip.
MediaTek, & Astera Labs; Synopsys, & Cadence for IP blocks & design services) or a custom CPU (Fujitsu & Qualcomm initial partners) to NVDA Blackwell / Rubin GPUs.
For NVDA CPU-to-custom ASICS NVDA will sell an NVLink Chiplet, while the NVLink functionality runs in NVDA GPUs for the NVDA GPU-to-custom CPU topology.”
This research note is reproduced verbatim from the issuing firm. Price Target never edits, paraphrases or alters analysts’ words — we only republish them in one place.
