Optical interconnects for AI set to hit $144B by 2030

Optical interconnects for AI to hit $144B by 2030 — AI-generated image

The global optical interconnect market for AI data centers is projected to reach $144.4 billion by 2030, up from $13.7 billion in 2024. Silicon photonics, cheaper and more scalable, will capture nearly two-thirds of this market, driven by the rise of co-packaged optics.

Tenfold growth in six years

According to projections from China Insights Consultancy (CIC), the optical interconnect market for data centers will grow from $13.7 billion in 2024 to $144.4 billion in 2030, representing a compound annual growth rate of 48.1%. This surge is driven by copper’s inability to meet the bandwidth and speed demands of AI-focused compute centers.

Copper, previously dominant, sees its performance collapse beyond a few hundred gigabits per lane: at 400 Gbps and above, the useful range drops to one or two meters, with prohibitive power consumption and signal loss. Optical interconnects, which transmit data as light, are thus becoming essential for ultra-high-speed links.

Silicon photonics leads the pack

Silicon photonics (SiPh) is expected to account for 63.7% of market revenue by 2030, up from just 16.6% in 2020. Its share will surpass other technologies by 2027, with an annual growth rate of 68.5%, compared to 32.6% for alternatives. This dominance stems from its integration into mature CMOS processes, used for conventional electronic chips, reducing costs and enabling mass production.

Silicon photonic integrated circuits (PICs), which form the core of optical interconnects, cannot generate light. Lasers, which are essential, are still made from III-V materials (such as indium phosphide), which are more expensive. Players like OpenLight and Tower Semiconductor are trying to bypass this limitation by directly integrating these materials into silicon production flows.

Co-packaged optics: a key accelerator

Co-packaged optics (CPO) place photonic components directly on the package of chips (GPUs, accelerators), reducing electrical distance to a few millimeters. This minimizes signal loss and power consumption, critical for extreme data rates. In 2024, 400 Gbps and 800 Gbps links still dominate the market, but by 2030, 1.6 Tbps and 3.2 Tbps technologies alone will account for $110 billion in revenue.

Investments are following this trend: over the past year, more than $15 billion has been poured into startups and CPO-related technologies. Nvidia invested $2 billion in Coherent and Lumentum, then another $2 billion in Marvell, while Microsoft, Meta, and OpenAI launched a working group (OCI MSA) to standardize optical interconnects at the scale of AI clusters.

Technical challenges remain

Despite these advances, obstacles persist. III-V lasers remain a bottleneck: Lumentum is currently the only supplier capable of mass-producing the 200 Gbps-per-lane emitters needed for 1.6 Tbps modules. Additionally, integrating photonic and electronic components poses thermal dissipation and precise alignment challenges, as highlighted by research from NHanced Semiconductors and the University of Florida at SMTA International 2026.

Their work focuses on hybrid bonding (copper-copper hybrid bonding) and silicon interposers, key techniques for 2.5D integration of photonic and electronic chips. These approaches enable connection densities exceeding 10,000 to 100,000 per mm² while reducing signal loss and power consumption.

If projections hold, optics will become the standard for AI data centers. But the race for performance still faces physical and industrial limits, where every gain in speed or density comes at the cost of complexity.