# GlobalFoundries: First Chipmaker to Support Open Standard for AI Scale Up > Source: > Published: 2026-06-15 22:00:00+00:00 GlobalFoundries expects to be the first with silicon to support the Optical Compute Interconnect Multi-Source Agreement (OCI MSA), an open standard that will let AI data center operators connect GPUs from different suppliers, GlobalFoundries senior fellow Vikas Gupta told EE Times. GlobalFoundries has set a milestone toward a ramp up of silicon that is likely to start as early as 2027, according to Tantra Analyst founder Prakash Sangam. OCI MSA, announced in March, includes top AI players AMD, Broadcom, Meta, Microsoft, Nvidia, and OpenAI. The consortium is supporting a hyperscalers-driven ecosystem to build a multi-vendor supply chain for optical scale-up interconnects. The agreement is in part a hedge against potential GPU shortages. “OCI MSA was the first attempt by industry to create a standard for scale-up networking, which traditionally has sort of been the wild, wild West, where all of these companies were coming up with their own protocols,” Gupta told EE Times. “Our silicon photonics technology is ready for this [OCI MSA] protocol. We have all the packaging systems available for customers to build OCI MSA-based systems. That’s a huge step forward for the industry in terms of getting time to market.” [View All](https://www.eetimes.com/category/sponsored-content/) Scale-up networks connect multiple GPUs inside a data center rack to directly access each other’s memory, transforming multiple processors into one virtual GPU. OCI MSA enables hyperscalers to disaggregate top-tier processor unit (XPU) engines and top-tier scale-up switches through a common optical physical layer, coupling compute more closely with optics, the alliance members noted in a press statement. OCI will enable migration from copper-based to optical-based scale-up architectures, helping eliminate copper interconnect speed and power bottlenecks, the statement said. “We do have customers that are working on the OCI MSA,” Gupta said. “This year, we’re expecting silicon out.” The agreement is in part a bet to hedge against potential shortages of GPUs, Gupta observed. “You don’t want vendor lock in,” he said. “It’s more making sure you’re not stuck with one vendor, and if they cannot supply, there is a supply chain issue.” Deployment by hyperscalers such as Amazon or Google for their own services, where they mix and match products from different vendors, is the initial business opportunity for the new standard, according to industry analyst Sangam. The standard has broad support from both the demand and supply sides, so the potential business opportunity is “huge,” Sangam noted. Other foundries are likely to offer their own products supporting OCI MSA, according to Sangam. “Samsung and Intel will be compelled to support it,” Sangam said. “TSMC has its own COUPE solution. But if its major customers, Nvidia and Broadcom, who are founding [alliance] members, insist on OCI MSA, it would be relatively easy for TSMC to comply with that spec as well.” OCI MSA will compete with established proprietary standards such as NVLink from Nvidia and UALink from AMD. **Scale out, scale up** GlobalFoundries is working toward optical scale up and scale out, Gupta said. “We’re working on high-speed modulators that are rated up to 400 gig per lane,” he said. Scale out is the addition of servers to a distributed system to handle increased workloads. “One of the innovations on scale out is that the conventional wisdom is after 200 gigabits per second per lane, silicon will run out of steam as a modulator material,” Gupta said. “There’s a lot of buzz in the industry about using thin-film lithium niobate (TFLN) or barium titanate. As we explore these materials, we’re also starting to see some good results on silicon-based modulators.” TFLN and barium titanate aren’t easily integrated onto a silicon photonics platform, Gupta noted. “If we can squeeze out silicon for one more generation, that pushes out this complexity for yet another generation,” he said. **Exceeding the standard** GlobalFoundries claimed its Scale (silicon photonics co-packaged advanced light engine) tech exceeds the OCI MSA standard in at least two ways. OCI relies on dense wavelength division multiplexing (DWDM). The OCI Gen1 specification uses four wavelengths (lambda) per fiber. “We have customers that have demonstrated 16 lambdas using our technology already,” Gupta said. “That’s more lambdas than are required by the MSA. All our modulators are already rated 100 gigabytes per second. The second part of it is our fiber-coupling systems are already broadband. The insertion loss of our fiber-coupling system is flat over the entire CWDM [coarse wavelength division multiplexing bandwidth].” While CWDM and DWDM boost fiber optic capacity by sending multiple data streams of light wavelengths along a single strand of fiber, CWDM is best for shorter-range networks and DWDM better fits long-haul, high-capacity needs. GlobalFoundries supports the CWDM and DWDM standards. (Source: GlobalFoundries) “Our competition has this parabolic insertion loss performance, where the insertion loss is lowest only at a particular wavelength, and they’ll need multiple grating couplers or multiple fiber couplers,” Gupta said. “The system that we have developed has a flat insertion across the full CWDM band. So even as you add wavelengths, you don’t have to change the fiber-coupling system.” **Advanced packaging** GlobalFoundries has been developing advanced packaging to support its Scale tech. “We will have customers that are doing die stacking,” Gupta said. “The technology was developed ground up, accounting for the fact that our customers will require advanced packaging. We had already developed pads to be able to put the EIC through silicon vias to the photonic IC. Whether they [customers] want to attach it to an organic laminate or whether they want to put it on a silicon interposer, the pitches of bumps that we have can address both these applications.” With interconnect links starting to go beyond 200 gigs, moving from copper to optical is a must, according to Sangam. “’27, ’28 is where we’ll see real CPO deployment,” he said. “CPO is gaining traction, and it’s the preferred option for scale up. OCI MSA is a major milestone toward achieving that.” ##### See also: [GlobalFoundries Champions ‘Essential Technologies’ over Moore’s Law](https://www.eetimes.com/globalfoundries-champions-essential-technologies-over-moores-law/) [How GlobalFoundries Takes AI from Pilot to Global Scale](https://www.eetimes.com/how-globalfoundries-takes-ai-from-pilot-to-global-scale/) [MIPS, GlobalFoundries Bet on Physical AI](https://www.eetimes.com/mips-globalfoundries-bet-on-physical-ai/)