GlobalFoundries' Optical Interconnects Scale to 70% of Data Centers by 2030

GlobalFoundries posted Q2 2026 revenue of $1.786 billion, up 6% year over year. The significant number: the communications infrastructure and data center segment rose 62% year over year and 20% sequentially. GF has raised its full-year forecast for that segment to 50–60% growth, up from a prior forecast in the high 30% range. Silicon photonics revenue is expected to more than double in 2026. The company holds a $300 million letter of intent with the U.S. Department of Commerce to fund the next-generation packaging technology needed to scale it.

At terabit-per-second data rates, copper links require aggressive equalization and retiming that consume power, add latency, and stop scaling as cluster sizes grow. GF CTO Gregg Bartlett frames the physics case simply: "If you believe in physics, then you should believe that photonics is ultimately going to succeed." CEO Tim Breen targets optical links accounting for at least 70% of data center connections by 2030, though he flagged that estimate may prove conservative.

GF's silicon photonics revenue is still dominated by pluggable transceivers, but the architecture is migrating in two steps. Near-packaged optics (NPO) ramps start in 2027; co-packaged optics (CPO) ramps follow in 2028. Many customers treat NPO as an intermediate step toward CPO. The target architecture eliminates the DSP retimer that pluggables require, cuts the electrical path between optical and electronic components to millimeters, and achieves bandwidth aggregation of 1.6 to 6.4 Tb/s per switch tile. NVIDIA's Quantum-X InfiniBand and Spectrum-X Ethernet switches have already deployed this architecture in production, reporting 5x power efficiency gain over pluggable alternatives and 5x improvement in sustained application runtime.

GF's SCALE platform — Silicon photonics Co-packaged Advanced Light Engine — combines a photonic integrated circuit (PIC), an electronic integrated circuit (EIC), drivers, transimpedance amplifiers, fiber attachment, bonding, and testing into a single known-good optical module. SCALE is the industry's first platform certified to the OCI MSA specification for AI scale-up interconnects; GF has demonstrated 8λ and 16λ bi-directional DWDM natively on its process. GF has seven active customer engagements on SCALE and is working with four of the five largest optical transceiver suppliers. One design win taped out in Q2 2026; another tape-out is expected in Q3.

Overall fab utilization sits in the high 80% range, but silicon germanium capacity is the pressure point. SiGe, which GF co-integrates with silicon photonics for the driver and TIA side of the link, is oversubscribed through 2027. GF is expanding SiGe on a 300 mm process in Singapore, adding output in Malta, NY, and Vermont. The November 2025 acquisition of Advanced Micro Foundry (AMF) in Singapore — making GF the largest pure-play silicon photonics foundry — added 200 mm capacity and a second geographic supply node. The $300M DOC award targets the next constraint: generation-2 SCALE moves from thermocompression to copper hybrid bonding, achieving sub-10-micron interconnect pitch between PIC and EIC without solder bumps. New modulator materials — thin-film lithium niobate, barium titanate, indium phosphide — target 400 Gb/s per wavelength as silicon alone approaches its modulator limits above 200 Gb/s.

For platform leads designing inference serving infrastructure, the signal is clear: SiGe-constrained capacity through 2027 means pluggable transceiver supply will stay tight for any design that relies on GF silicon germanium drivers. NPO and CPO won't be off-the-shelf options until 2027–2028 at earliest. Architects must decide now whether to design optical-module interfaces into next-generation switch and GPU board designs so that dropping in NPO in 2027 requires a swap, not a re-spin.