<div id="readability-page-1"><div><div>BofA Securities on Thursday noted two key recently announced AI scale-up interconnect protocols: Nvidia Corp’s NVDA NVLink Fusion and the Advanced Micro Devices, Inc AMD, Broadcom Inc AVGO backed open-standard Ultra Accelerator Link (UALink). The analyst noted that both are critical scale-up techs connecting AI accelerators (GPUs, ASICs) via fast, low-latency, and shared-memory networking, enabling the scaling of accelerators into large rack-scale systems.The analyst said UALink is the industry’s latest effort to catch up to Nvidia’s proprietary NVLink technology, the chip maker’s key advantage in creating large rack-level systems.UALink is an open-standard initiative formed by a consortium of industry leaders including AMD, Broadcom, Intel Corp. INTC, Marvell Technology, Inc. MRVL, MediaTek, Alibaba BABA, Alphabet Inc.‘s GOOG GOOGL Google, Meta Platforms, Inc. META, Microsoft Corporation MSFT, and Arm Holdings plc ARM. Its primary goal is to provide an open alternative to Nvidia’s proprietary NVLink for AI scale-up interconnects. The UALink Consortium, founded in May 2024, now includes more than 70 member companies.Also Read: Nvidia Powers Big Gains For Quanta, Pegatron As AI Demand ExplodesWhile UALink generally remains 1-2 generations behind NVLink, the analyst noted UALink as a significant step forward for the non-Nvidia ecosystem in potentially tackling the chipmaker’s dominance in AI scale-up from 2027 onward. The analyst now views Nvidia as the chip, software, and system leader.Meanwhile, Nvidia’s NVLink Fusion partially opened up NVLink IPs to some of its ecosystem partners (excluding Broadcom and AMD), opening up ~$12 billion in networking TAM that Nvidia previously did not have access to, the analyst noted.The analyst said AMD is a founding member of the UALink Consortium and has since contributed to creating this industry standard and its surrounding ecosystem.With the MI400 generation targeted for the second half of 2026, AMD aims to begin designing rack-level AI systems with its ZT Systems tech. According to the analyst, UALink could provide its much-needed accelerator-to-accelerator and switching capabilities intra-rack.Still, the analyst highlighted that the latest UALink 1.0 standard only supports up to 800GB/s bandwidth per accelerator, less than half the rate of the latest NVLink 5.0, 1.8TB/s.Given that UALink compatible systems will likely ramp from 2027, Nvidia would ramp its NVLink 6.0-based Rubin systems, which implied that Nvidia would likely maintain its 1-2 generation advantage in AI scale-up networking, per the analyst.Meanwhile, NVLink Fusion could expand Nvidia’s TAM in networking that is attached to custom accelerators. The analyst assumed custom silicon represents ~15% of ~$500 billion AI compute TAM, or ~$75 billion, by the calendar year 2029-30, and AI networking represents 15% of that, or ~$11 billion.Similarly, NVLink Fusion opens up opportunities in non-Nvidia-based ARM CPU systems (such as from Qualcomm Inc QCOM), or ~$1 billion of networking TAM (15% of ~$5-6bn ARM CPU TAM), as per the analyst.The analyst added that with NVLink Fusion, every end customer (hyperscalers) will have a choice between pairing their silicon with NVLink and UALink, an incremental benefit for Nvidia.Nvidia’s NVLink, even with the partial openness of NVLink Fusion, is poised to remain the leading AI scale-up interconnect for the foreseeable future, driven by its technological lead and established ecosystem.However, the analyst asserted that UALink’s emergence as an open standard backed by major industry players is a crucial development. The analyst added that it provides a viable alternative that could challenge Nvidia’s long-term dominance by fostering greater innovation and competition in the AI interconnect space.The analyst stated that the coming years, particularly 2026 and beyond, will be critical in observing how UALink’s adoption accelerates and how Nvidia further evolves its proprietary technology.Price Action: NVDA stock is trading lower by 2.05% to $130.11 at last check Friday.<ul> <li>AMD Could See EPS Upside By 2027 Despite Nvidia’s Over 80% Market Share: Analyst</li> </ul>Image via Shutterstock</div></div></div>

NVDA

BofA Securities highlighted the emergence of UALink, an open-standard AI interconnect protocol backed by major industry players, as a potential challenger to Nvidia's NVLink technology. While UALink is currently 1-2 generations behind NVLink, it represents a significant step for the non-Nvidia ecosystem. The UALink Consortium, founded in May 2024, aims to provide an alternative for AI scale-up interconnects. Despite Nvidia's technological lead, the analyst believes UALink's development could foster competition and innovation in the AI interconnect space, particularly from 2027 onward.

- BofA Securities highlighted two key AI scale-up interconnect protocols: Nvidia's NVLink Fusion and AMD's UALink, aimed at enhancing AI accelerator connectivity.  
- UALink, backed by a consortium of over 70 companies, seeks to provide an open alternative to Nvidia's proprietary technology, although it currently lags behind in performance.  
- The emergence of UALink is seen as a significant development that could foster competition and innovation in the AI interconnect space, challenging Nvidia's dominance in the long term.

Emergence Of UALink As A Viable Alternative Could Challenge Nvidia's Dominance, Analyst Asserts