[OFC 2026] Part 2 of 5: CPO and the AI Interconnect Challenge

Key Takeaways

• AI racks now exceed 1 MW power per rack
• Copper interconnects hit bandwidth and routing limits
• Meta demands CPO match cost, power, reliability
• Five papers address lasers, SOA, BiDi, reliability, detectors
• Operational viability remains the biggest CPO hurdle

Pulse Analysis

The relentless growth of AI workloads is forcing data‑center architects to rethink traditional copper interconnects. Meta’s recent rack upgrades illustrate a tipping point: a single rack now consumes more than a megawatt, a level at which copper’s electrical losses, heat dissipation, and signal‑integrity challenges become prohibitive. Engineers are scrambling for alternatives that can sustain terabit‑per‑second links without exploding power budgets, and the industry is watching closely as these constraints shape the next wave of hardware design.

Co‑Packaged Optics (CPO) has emerged as the leading candidate to replace copper, but Meta’s public baseline at OFC 2026 raises the bar. Optical modules must not only deliver higher bandwidth; they also need to match or beat copper in cost per gigabit, power efficiency, and long‑term reliability. This three‑pronged requirement forces vendors to innovate across the entire optical stack, from laser sources to photodetectors, while keeping manufacturing expenses in check. The emphasis on operational viability signals that the market will only adopt CPO once it proves itself in real‑world, high‑density deployments.

The five papers spotlighted at OFC provide a cross‑section of the challenges ahead. Furukawa’s work on external lasers promises tighter wavelength control, while Lightmatter’s quantum‑dot SOA and bidirectional transceiver designs aim to boost link margins and simplify cabling. Meta and Broadcom’s field‑reliability study, encompassing 36 million operational hours, offers rare empirical data on failure modes. Finally, Marvell and McGill’s photodetector advances target lower noise and higher sensitivity. Collectively, these efforts underscore a critical gap: while performance metrics are improving, proving sustained, cost‑effective operation at scale remains the decisive hurdle for CPO adoption.