AMD Joins Nvidia and Others to Launch Optical Interconnect Alliance Tackling AI Data

The Optical Compute Interconnect Multi‑Source Agreement (OCI MSA) announced this week brings together AMD, Nvidia, Broadcom, OpenAI, Meta, Microsoft and several silicon‑photonic specialists to define a vendor‑agnostic PHY for optical links in AI data centers. According to Tom’s Hardware, the group’s charter is to replace copper’s “physical limits for optimal data‑transfer speeds and power consumption” with an open‑specification that can support “higher domain scale‑up sizes” and a multi‑vendor supply chain (Tom’s Hardware). The initiative targets a baseline throughput of 3.2 Tb/s per optical lane, a figure that would dwarf the 400‑600 Gb/s per‑lane limits of the latest copper‑based Ethernet standards used in today’s AI clusters.

Copper’s shortcomings are two‑fold: signal degradation at high frequencies and exponential power draw as distances increase. Tom’s Hardware notes that “pushing electrical signals to high speeds results in signal degradation and unsustainable levels of power consumption,” while the resistive nature of copper “necessitates huge amounts of power to send data over distances at high speeds.” By moving the physical layer to light, the OCI MSA aims to eliminate the resistive losses that force data‑center operators to over‑provision power and cooling. Optical fibers also allow “more systems to be concurrently connected over greater distances” without the attenuation penalties that copper imposes, a benefit that directly addresses the “ongoing AI infrastructure buildout” described by the alliance (Tom’s Hardware).

The consortium’s roadmap hinges on silicon‑photonic integration, a technology that already powers “pluggable transfer ecosystems” for inter‑switch connectivity, according to Xscape Photonics CEO Vivek Raghunathan in a Tom’s Hardware interview. He points to TSMC’s COUPE process as the “foundational bedrock for enabling optical and photonics in chips,” underscoring that the manufacturing ecosystem is already in place to produce the required optical PHYs at scale (Tom’s Hardware). By standardizing the interface, the OCI MSA hopes to unlock a competitive market for optical transceivers, reducing the current “supply‑chain constraints” that plague copper‑based interconnects and allowing data‑center builders to source components from multiple vendors rather than a single, bottlenecked supplier.

Despite the clear performance upside, the alliance acknowledges the nascent state of large‑scale optical deployment. Tom’s Hardware flags several technical and economic challenges: higher component costs, increased heat output, and “overall failure rates” that remain higher than mature copper solutions. The OCI MSA’s mandate therefore includes developing reliability metrics and cost‑reduction pathways alongside the raw bandwidth targets. By establishing a “protocol‑agnostic, scale‑up interconnect” specification, the group hopes to create a common testing and validation framework that will accelerate industry adoption and drive down prices through economies of scale.

If successful, the optical PHY could reshape the economics of AI training clusters, where power bills now account for a substantial share of total operating costs. The shift from copper to light would not only curb power consumption but also enable “larger scale‑up domains” that can host more GPUs and accelerators per rack without the thermal and electrical penalties that currently limit density (Tom’s Hardware). For AMD, joining the alliance signals a strategic pivot toward data‑center infrastructure, complementing its recent AI‑focused processor launches reported by TechCrunch and CNET. Nvidia’s participation reinforces its role as the de‑facto leader in AI hardware, while Broadcom’s involvement brings high‑speed Ethernet expertise to the optical domain. Collectively, the consortium’s effort represents the most coordinated industry response yet to the bandwidth bottleneck that threatens to throttle the next wave of generative‑AI models.