How long will AI-driven optical communication remain hot? Stock prices have surged tenfold, and Lumentum says orders are booked out five years.

On March 2, 2026, NVIDIA announced simultaneous $2 billion investments in Lumentum and Coherent, accompanied by multi-year procurement commitments and capacity priority rights worth several billion dollars. On March 31, NVIDIA invested another $2 billion in Marvell Technology, collaborating on silicon photonics and the NVLink Fusion platform. On May 6, NVIDIA partnered with Corning, investing $500 million and signing a multi-year commercial agreement. Corning will build three new fiber-optic–dedicated factories in North Carolina and Texas, expanding U.S.-based fiber production capacity tenfold. Within three months, NVIDIA’s direct investments in the photonics domain exceeded $6.5 billion. Including concurrent investments in data center operators such as CoreWeave and IREN, NVIDIA’s total strategic investment in 2026 surpassed $40 billion. These investments are not isolated events. Lumentum and Coherent sit upstream in the optical communications value chain, producing core components such as lasers that serve as light sources for high-speed optical modules; Marvell focuses on optical DSPs and silicon photonics platforms, handling signal processing and conversion between electrical and optical domains; Corning supplies foundational materials and transmission media—fiber optics and cables. Taken together, NVIDIA is proactively securing key links—from light source and interface to transmission medium—around high-speed interconnects for AI data centers. This move stems not merely from growing optical communication demand, but from preparing its supply chain to keep pace with AI cluster deployment. In the traditional telecom and cloud computing era, the optical components market grew steadily, allowing manufacturers to scale capacity at predictable rates. But AI data centers generate demand that is far more concentrated—and sudden. GPU cluster expansion, network bandwidth upgrades, and mass deployment of 800G and 1.6T optical modules are occurring nearly simultaneously. For the relatively small-scale, high-end optical supply chain, such a demand leap easily triggers bottlenecks at critical nodes. Through equity investments, multi-year purchase agreements, and capacity priority rights, NVIDIA has effectively pre-allocated the most scarce optical capacity over the coming years into its own supply system. Other AI chipmakers and cloud providers are adopting similar strategies. AMD participated in Ayar Labs’ $500 million Series E financing round and acquired startup Enosemi in 2025, while also making equity investments in Teramount and Celestial AI. The venture capital arms of Alphabet and Microsoft co-invested $80 million in optical company nEye’s Series C round in 2026. From Pluggables to CPO NVIDIA’s bet on the optical supply chain isn’t just about securing today’s optical module supply—it’s also an early positioning for the next-generation data center network architecture.

Industry Perspective | 27 Startups, $2 Billion Inflows: Global Competition Landscape and Development of Optical Computing

On April 28, 2026, Lightelligence officially listed on the Main Board of the Hong Kong Stock Exchange. Its market capitalization exceeded HK$80 billion (approximately USD $10.3 billion) on its first trading day, making it the world’s first publicly listed company in the photonic-electronic hybrid computing sector. Lightelligence’s core team originates from MIT; in April 2025, it published the PACE (Photonic Arithmetic Computing Engine) paper in Nature, demonstrating a large-scale photonic accelerator integrating over 16,000 photonic components. The PACE system adopts a photonic-electronic hybrid architecture: its photonic integrated circuit (PIC) is fabricated using commercial 65nm silicon photonics technology, while its electronic integrated circuit (EIC) uses 28nm CMOS technology; the two are densely interconnected via 2.5D flip-chip packaging. In 64×64 matrix operations, it achieves a 1 GHz operating frequency, 3 ns latency per cycle, 8.19 TOPS throughput, and 2.38 TOPS/W energy efficiency. In solving the Ising model, it completes each iteration nearly 500× faster than an NVIDIA A10 GPU (5 ns vs. 2300 ns). On March 25, 2025, Lightelligence launched its new photonic-electronic hybrid compute card, the ‘Lightelligence TianShu’. Its core processor comprises an Optical Processing Unit (OPU) and an Application-Specific Integrated Circuit (ASIC), built using advanced 3D packaging technology. It operates at 1 GHz with 8-bit output precision; its optical chip measures 600 mm² and integrates over 40,000 optical devices, supporting 128×128 matrix operations. TianShu successfully ran deep convolutional neural network ResNet50 and large language model LLaMA 2—marking the first commercial application of photonic-electronic hybrid computing. Other Notable Players LightOn (France): Focuses on optical computing for large-model inference, having already launched enterprise-grade optical acceleration solutions for AI inference. Optalysys (UK, founded 2013): Has raised $32.1 million to date, developing lensless optical processing prototypes targeting multi-PFLOPS-level optical computing. Lemurian Labs (Canada): Specializes in optical AI processor design. SpiNNcloud (Germany, founded 2021): Employs brain-inspired computing architecture incorporating optical computing concepts. Neurophos (USA, founded 2020): Core technology centers on an Optical Processing Unit (OPU) based on optical metasurfaces and silicon photonics, performing AI computations using light signals; plans mass production of its photonic AI chip by 2028. Guang Ben Wei (‘Optical Standard’): A post-95s team that successfully taped out the world’s first commercially viable standard optical computing chip. Its optical computing chip adopts the PCM + Crossbar approach, replacing electricity with light for AI inference computation. Guang Zi Xin Li (‘Photon Chip Power’): Core team members mainly hail from Tsinghua University; develops photonic-electronic fusion computing chips using a ‘full-wave metasurface’ technical route. Additionally, the three traditional chip giants—NVIDIA, Intel, and AMD—have all entered optical interconnect and optical computing through investments or in-house R&D. Intel’s years of accumulated expertise in silicon photonics transceivers gives it a first-mover advantage in CPO (co-packaged optics) technology.

Lightwave Logic Stock Jumps as AI-Photonics Bet Heads for a 2027 Production Test

May 13, 2026

Lightwave Logic shares surged 14% following talks for 2027 large-scale production driven by AI demand. The polymer-photonics firm reported a 27% revenue increase, with four Fortune 500 companies prototyping its electro-optic polymers.

TechStock²

2026-04-22

Lightwave Logic Shares Reach All-Time High Amid AI Polymer Optimism

2026-05-06

Lightwave Logic, Inc. Announces Timing of First Quarter 2026 Financial Results and Business Update Call

2026-05-13

Lightwave Logic Stock Jumps as AI-Photonics Bet Heads for a 2027 Production Test

2026-05-15

Lightwave Logic Inc. Reports Q1 2026 Earnings

2026-05-19

Lightwave Logic Shares Fall 19% After Wider Q1 Loss and Shelf Filing