AI Could Push Data-Center Optical Interconnects to $144.4 Billion by 2030

The headline from the China Insights Consultancy (CIC) report is striking: global data-center optical interconnects are projected to grow from $13.7 billion in 2024 to $144.4 billion by 2030, equivalent to a 48.1% compound annual growth rate and more than a tenfold increase. The forecast was commissioned by Yuanjie Semiconductor Technology for its Hong Kong IPO filing, so it should be read as an issuer-linked market projection rather than an independently audited forecast.

The bigger story is not only the size of the estimate. It is the collision between rapidly rising AI-infrastructure requirements and a supply chain that still has limited capacity for many of the supporting chips and optical components needed to connect increasingly dense computing clusters.

Why AI is driving demand for optical interconnects

AI systems move enormous volumes of data between processors, memory and networking equipment. Optical interconnects use light to support the high-bandwidth, low-latency connections required by AI workloads, including links inside and between data-center systems.

That demand is shifting optical connectivity from a specialized networking component toward a core part of AI infrastructure. As clusters become larger, operators need links that can move more data without allowing power consumption, heat and physical density to become limiting factors. Industry efforts to develop open optical scale-up specifications reflect that pressure: the Optical Compute Interconnect Multi-Source Agreement group was formed by AMD, Broadcom, Meta, Microsoft, Nvidia and OpenAI to define an interoperable optical connectivity approach for large AI systems.

This does not mean copper connections disappear immediately. It does mean that optical links are increasingly being considered for parts of the system where electrical connections face reach, bandwidth or power constraints.

Silicon photonics and co-packaged optics are central to the thesis

CIC’s reported forecast sees silicon photonics becoming much more important in the market. The share attributed to silicon photonics is described as rising from 16.6% in 2020 to 63.7% in 2030, with the technology accounting for nearly two-thirds of future revenue.

Silicon photonics combines optical functions with silicon-based manufacturing approaches. Its appeal in AI infrastructure is closely tied to the need to produce higher-density optical connections at scale, although the supplied evidence does not independently verify every detailed CIC assumption behind the forecast.

The longer-term architectural direction is co-packaged optics (CPO), in which optical engines are placed closer to—or integrated into—the switching or compute package instead of relying only on conventional, detachable transceiver modules. Bringing optics closer to the source of computation can help address interconnect distance, power and density challenges, but it also introduces difficult manufacturing, cooling, testing and serviceability requirements. The OCI MSA’s work on an open optical scale-up specification is part of the broader industry push toward interoperable optical infrastructure.

What the forecast says about link speeds

The question surrounding the CIC report also includes projections that 1.6T links could become the largest segment at $65.6 billion by 2030, while 3.2T technology could reach $44.5 billion. Those figures are part of the reported forecast, but they are not independently substantiated by the supplied evidence and should not be presented as established market outcomes.

The direction of travel is clearer than the exact segment sizes. Optical infrastructure is moving toward higher per-link capacity as AI clusters scale, and industry roadmaps already discuss optical connectivity reaching the 1.6T-per-fiber level. The timing, adoption rate and eventual revenue split between 1.6T, 3.2T and other configurations remain forecast-dependent.

The market number needs context

The $13.7 billion-to-$144.4 billion estimate refers specifically to the global data-center optical-interconnect market in the CIC forecast reported by Tom’s Hardware. Other supplied market research uses a broader optical-interconnect definition that includes telecommunications and estimates the total market at $17.9 billion in 2024 and $151.4 billion in 2030, with a 42.8% CAGR.

Those figures are not necessarily contradictory: they cover different market boundaries and use different methodologies. The distinction matters because market-size comparisons can look inconsistent when one estimate covers data centers only and another includes telecom applications.

The same caution applies to the claim that more than $15 billion has been invested recently across the industry. That amount is not independently supported by the evidence provided here, so it is better treated as an unverified detail rather than a confirmed measure of the sector’s commitment.

Chinese foundries show the supply-side pressure

AI demand is affecting more than leading-edge accelerator production. Supporting systems also require mature-node logic, power-management and optical-related chips. SMIC said AI would continue to support robust foundry demand and that it was adjusting capacity and accelerating new production-line ramps to help ease industry-wide constraints.

SMIC reported 93.7% utilization in the second quarter of 2026, up from 93.1% in the first quarter. Its monthly production capacity also rose to the equivalent of about 1.1 million eight-inch wafers, yet utilization still increased.

Hua Hong was also reported to be operating under intense pressure. One report gave its utilization at 102.8%, alongside SMIC’s 93.7%; because utilization above 100% depends on the reporting methodology and effective capacity assumptions, the figure should be read as evidence of exceptionally tight conditions rather than as a directly comparable physical rate.

The available reporting also says that demand for some mature-node chips used alongside AI processors—particularly BCD power-management products—was visible through the end of 2027. That claim comes from secondary reporting and is not independently confirmed by the supplied primary evidence.

Expansion may not provide immediate relief. SMIC has discussed adding equipment at existing factories where space is available, but the supplied evidence does not establish firm dates for when meaningful new capacity will come online or how much of it will be qualified for particular optical and power-management processes.

Chinese foundries’ share of global 22–40 nm legacy-node capacity is projected to rise to 41% in 2027 from 32% in 2025. However, more aggregate capacity does not automatically eliminate shortages in a specific process technology, packaging flow or qualified production line.

What to take from the CIC projection

The forecast supports a strong strategic thesis: AI expansion is likely to make optical connectivity more important throughout the data-center stack. Silicon photonics, higher-speed links and co-packaged optics are positioned as potential answers to rising bandwidth, power and density requirements.

But the precise outcome is less certain than the headline suggests. The $144.4 billion figure is a commissioned projection; the detailed silicon-photonics share, 1.6T and 3.2T revenue figures, investment total and capacity-timing claims are not all independently verified in the supplied evidence. Meanwhile, high utilization at Chinese foundries shows that manufacturing capacity—not just network architecture—could determine how quickly the optical buildout scales.

The most defensible conclusion is therefore two-sided: AI is creating a powerful demand case for data-center optical interconnects, but converting that demand into a $144.4 billion market will depend on technology adoption, standardization and the industry’s ability to add qualified capacity.