AMD Networking Chief Jiandani: 'You Never Bet Against Ethernet'

AMD paid $1.9 billion for Pensando, the networking startup co-founded by Soni Jiandani, and that acquisition now anchors the chipmaker's argument that no single vendor can own AI infrastructure. Jiandani, who leads networking strategy at AMD, told SDxCentral that openness and programmable silicon — not a vertically integrated stack — will decide who builds the networks behind AI.

"There's no single company that owns the networking landscape in the AI arena," Jiandani said. "AI infrastructure is just too large and complex for one size or one vendor to make it fit in all forms and shapes."

Why does AMD's Ethernet stance matter?

The position is a deliberate contrast with Nvidia, which banks on tight integration with proprietary hardware, and with Cisco — where Jiandani spent two decades convincing the market that closed, vertically integrated stacks were the smart bet. The "green-colored rival" she declines to name markets itself as the largest networking company in the world; AMD is betting the opposite way.

Her career frames the argument. From Ungermann-Bass Networks through Cisco's multibillion-dollar modular switch business to Pensando, Ethernet has been the constant. What has changed is throughput, resilience and the standards built on top.

AMD's ecosystem work spans two domains:

• Scale-up, inside the rack: UALink over Ethernet (UALoE) adds resilience; Ethernet for scale-up networking (ESUN) improves throughput through targeted optimizations.
• Scale-out, rack to rack: multiprotocol transport supporting multipathing, which Jiandani said drives lower total cost of ownership for customers including OpenAI, Oracle and Microsoft.

"You never bet against Ethernet," she said. "The beauty about Ethernet is that the Ethernet from four decades compared to three decades to two decades to one decade to now are very different. It continues to evolve."

What is AMD's moat against Nvidia?

Cisco's moat is its install base and partner network. Nvidia's is proprietary integration. Jiandani was blunt about AMD's answer: "Programmability is a baseline architectural element. It's our moat, so to speak."

The capability comes from two acquisitions. Pensando supplied high-speed, processor-driven software programmability. Xilinx brought hardware-level adaptivity through FPGAs and later Versal chiplets.

The practical payoff, in her telling, is speed. Because the pipeline is fully programmable, AMD can adapt to evolving Ethernet standards in software rather than waiting for new silicon.

"Even when speeds are going up and not coinciding with the evolution of those protocols, I can just … write a P4 program, adapt to the evolving standard," she said. "I don't tell my customers, 'wait for the next silicon while I burn the new gates for that new enhancement in the protocol,' because the evolution of AI workloads is so fast."

Jiandani put a number on that acceleration: architectural change in the AI era is running "at least three- or four-times faster" than anything she saw before AI.

How does 'open' coexist with custom silicon?

The proof points landed within a day of the interview. AMD's role in helping Meta build its custom transport protocol, MetaRoCE, became public, and the same programmable pipeline underpinned the multipath reliable connection (MRC) work co-developed with OpenAI and others. Two hyperscalers, two bespoke transport layers, one underlying silicon — a quiet rebuttal to the idea that open and custom are opposites.

When does AMD go optical?

Optical interconnects may be the biggest change of her career. Copper is not going away, but silicon photonics will complement it, and AMD enters the market in 2027 through its Instinct MI500. Jiandani declined to get ahead of AMD's roadmap disclosure on specifics.

She framed the shift as a "phased transition" rather than a leap: "We don't think one day the world is electrical, then tomorrow it's going to be optical … stay tuned; more details to be revealed."

AMD is a founding member of the Optical Compute Interconnect Multi-Source Agreement consortium (OCI-MSA), which is developing interoperable open standards for optical scale-up interconnects, both across racks and within them.

She also stressed reliability at every bandwidth step, saying customers should not hear that a feature works at 800G but requires waiting for 1.6T. Asked whether optics, open scale-up standards or software-defined rack intelligence moves first, she declined to pick, pointing to collaboration as the constant.

"Collaboration across the industry is more important than ever before," she said. "It's moving so fast." Practitioners should watch the MI500 roadmap disclosures and OCI-MSA standardization work through 2026 as the concrete tests of that open-Ethernet thesis.