As AI data centers push past the limits of traditional compute architecture, the industry’s attention is moving deeper into the physical layer. GPUs, accelerators, power systems and cooling platforms still dominate the headlines, but the network fabric that connects those systems is becoming just as critical.

A wave of recent announcements points to the same conclusion: future growth will depend not only on more compute, but on faster, denser, more efficient and more scalable optical connectivity.

A new multi-source agreement is bringing together major technology companies to standardize expanded beam optical connectivity for AI data centers. University of Arizona research is powering a new optical switching technology designed to reduce the energy consumed by data center networks. STL is planning to invest up to $100 million in U.S. manufacturing capacity to support AI data center and telecom customers with optical connectivity products.

Those developments are now being reinforced by a broader series of moves across the optical ecosystem: Corning’s major AI infrastructure partnerships with NVIDIA and Amazon, GlobalFoundries’ push into co-packaged optics, Sivers’ laser-array collaboration with GlobalFoundries, Wiwynn’s co-packaged optics demonstration at Computex, Credo’s acquisition of DustPhotonics, and emerging near-packaged optical interconnect designs from LightSpeed Photonics.

Taken together, these announcements highlight a maturing market around the optical layer of AI infrastructure. The value is not simply faster data movement. It is about reducing deployment complexity, lowering operating overhead, supporting higher-density clusters, improving energy efficiency and strengthening the domestic supply chain behind AI-ready networks. Let’s drill down into what these announcements mean.

Standards for the AI Optical Layer

The launch of a new coalition focused on expanded beam optical, or EBO, connectivity reflects a practical challenge facing AI deployments: as clusters grow larger and more bandwidth-intensive, physical connections become harder to deploy, maintain and scale.

3M announced that it has joined a group of technology leaders to create a multi-source agreement focused on open, interoperable specifications for EBO connectivity in AI infrastructure. The coalition includes a broad mix of companies across the networking, cloud, semiconductor, connector and optical ecosystem which, according to the group's website, as of June 29, 2026, included 44 member companies, among them many of the industry's best-known fiber optic and connectivity vendors.

The decision to build to a standard is important; data centers are not built by one vendor or around one component. They are assembled from servers, GPUs, switches, transceivers, fiber assemblies, connectors, cabinets, power distribution systems and software-defined network layers. As the number of optical links increases, the industry needs interoperability across that ecosystem.

Expanded beam optical technology is designed to improve the reliability and maintainability of optical links, particularly in high-density environments. Traditional multi-fiber physical contact connectors can be sensitive to contamination, inspection requirements and handling practices. In large AI clusters, even modest operational friction can multiply into significant deployment and maintenance overhead.

The EBO MSA group   (Expanded Beam Optical (EBO) Multi-Source Agreement)  is intended to address that issue by creating shared specifications for multiple connector configurations. The goal is not merely to create another connector type, but to establish a common framework that enables multiple vendors to build compatible products for hyperscale, cloud and enterprise AI deployments.

For data center operators, standardized EBO connectivity could reduce complexity in AI network builds, simplify procurement, improve supplier diversity and make optical cabling more resilient in dense environments. For hyperscalers, this can help speed cluster deployment. For enterprises building AI infrastructure, it can reduce the risk of being locked into a narrow ecosystem before the market has fully matured.

Optical Switching Targets Power and Heat

While the EBO coalition focuses on physical connectivity and standardization, the University of Arizona announcement addresses another pressing data center issue: the energy consumed by network switching.

The university said research developed by Pierre-Alexandre Blanche, a research professor at the University of Arizona, is powering Post Quantum Tek’s High-Speed Optical Switch, known as PQT-HOS. The announcement describes the technology as an optical switch that uses light diffraction to keep data in optical format throughout the switching process.

In many conventional data center network architectures, traffic may travel optically across fiber but then be converted into electrical signals for switching before being converted back into optical form.  Post Quantum Tek’s approach is described as optical-optical-optical switching, keeping the data in light form throughout the process.

According to Post Quantum Tek and the University of Arizona announcement, the switch is capable of operating up to 1,000 times faster while consuming approximately one-thousandth the energy of conventional switching approaches. Those performance figures remain developer-reported benchmarks that will ultimately need validation through commercial deployment and independent testing, but they point directly at one of AI infrastructure's defining challenges: ensuring the network does not become the energy bottleneck that limits compute performance.

The technology is not yet described as a fully commercialized data center product. The announcement says the PQT-HOS is patented, bench-proven and ready to be developed into a commercial prototype through the University of Arizona’s Tech Launch Arizona Institute. That places it earlier in the adoption curve than the EBO MSA or STL’s manufacturing investment.

The strategic relevance is substantial. The AI infrastructure market is now aggressively looking for ways to flatten network architectures, reduce electrical conversion points and move more data with less energy. Optical switching is one of the technologies that could reshape how future AI clusters are designed, particularly if it can be manufactured at scale, integrated with existing network architectures and proven under production workloads.

Corning and NVIDIA Put Optical Manufacturing at the Center of AI Factories

The most direct sign that optical connectivity has moved from supporting role to strategic infrastructure may be Corning’s long-term partnership with NVIDIA.

The companies announced a partnership to strengthen U.S. manufacturing for AI infrastructure, with Corning expanding U.S.-based optical connectivity manufacturing capacity by 10x and U.S. fiber production capacity by more than 50%. The expansion includes three new advanced manufacturing facilities in North Carolina and Texas and more than 3,000 new jobs.

NVIDIA has become the central supplier of accelerated computing systems for AI factories, but those systems depend on extraordinary volumes of fiber, connectivity and photonics to move data across clusters. As AI factories grow larger, the network becomes a constraint on how effectively GPUs can be used. Optical connectivity is now part of the scaling equation.

This means the value of the Corning-NVIDIA partnership is threefold:

1.      Supply. AI infrastructure buildouts are now so large that a shortage of optical connectivity components can delay deployment just as surely as a shortage of power equipment, transformers or GPUs.

2.     Domestic manufacturing. As hyperscale AI campuses become strategic assets, operators and chip companies are placing greater emphasis on resilient U.S.-based supply chains.

3.      Optics are becoming more tightly aligned with accelerated computing roadmaps. As NVIDIA systems move toward larger AI clusters and more demanding scale-up and scale-out architectures, the fiber and photonics ecosystem must evolve in parallel.

The announcement also reinforces the idea that AI infrastructure is not just a silicon story. The value of the GPU depends on the ability to connect thousands of GPUs into functioning systems. That makes optical connectivity a foundational component of AI factory design.

Amazon and Corning Reinforce the Data Center Fiber Supply Chain

Corning isn't limiting its AI infrastructure ambitions to NVIDIA. A separate multiyear, multibillion-dollar agreement with Amazon Web Services (AWS) extends the same strategy into the hyperscale cloud market. Under the agreement, Corning will supply optical fiber, cable and connectivity solutions supporting AWS's expanding U.S. data center footprint, while the partnership also includes manufacturing expansion, workforce development initiatives and new domestic jobs.

Amazon’s data centers support cloud computing, AI services, enterprise workloads and consumer applications. As those data centers grow denser and more distributed, the need for high-performance optical connectivity rises across multiple layers: inside buildings, between halls, across campuses and between regional facilities. It’s a straightforward fact: cloud data center expansion now depends on fiber capacity at enormous scale.

The workforce component is also important. AI infrastructure is frequently discussed in terms of chips, power and land, but deployment depends on skilled trades and manufacturing labor. Fiber optic production, splicing, testing and installation require trained workers. By tying the agreement to workforce development, Amazon and Corning are addressing one of the less visible constraints on scaling infrastructure: the availability of trained people to build and maintain the optical layer. This issue is one that seems to be hovering over all aspects of future data center development.

Large cloud and AI deployments depend on predictable delivery schedules. Domestic fiber and connectivity supply can help reduce lead times, simplify logistics and improve confidence that network builds will keep pace with compute deployments.

STL Builds U.S. Capacity for AI Data Highways

Sterlite Technologies Ltd. (STL)'s planned investment of up to $100 million in the United States fits directly into the same supply-chain narrative. The company said the investment will strengthen manufacturing capacity for customers, including AI data center and telecom operators, and support connectivity solutions such as terminated optical fiber cables.

The company described its optical solutions as high-density “AI Data Highways” designed to link data center campuses and support massive GPU-driven demand. Which, as we are seeing is a real market need. AI campuses are increasingly being planned at scales that require high-capacity connectivity between buildings and, in some cases, between geographically distributed sites. As power availability, land constraints and grid interconnection timelines shape data center location decisions, operators will need robust optical infrastructure to connect compute resources across larger footprints.

STL’s U.S. investment also comes at a time when domestic manufacturing and supply chain resilience are increasingly important considerations for infrastructure buyers. Predictable delivery, qualified suppliers and reduced exposure to supply disruptions are at top of mind for developers and investors. More manufacturing capability closer to U.S. demand will help reduce lead times and support future deployment schedules.

GlobalFoundries Pushes Co-Packaged Optics Toward Production

GlobalFoundries’ SCALE optical module solution adds another layer to the market story: the move from optical innovation to manufacturable photonics platforms.

The company introduced SCALE, a silicon photonics co-packaged optics solution aimed at advanced AI data centers. GF positioned the platform as an OCI MSA-capable (Optical Compute Interconnect Multi-Source Agreement) solution for modern AI scale-up architectures. As shown earlier, the plans for multi-vendor consortium building to a standard are making themselves felt. This platform uses silicon photonics, wavelength-division multiplexing and advanced packaging to improve bandwidth density and scalability compared with traditional copper interconnects.

Co-packaged optics is not just a component substitution. It changes where optical conversion happens. Instead of relying primarily on pluggable optics at the faceplate, CPO brings optical engines closer to switch or compute silicon. As with other optical standard updates, the goal is to reduce electrical trace lengths, improve bandwidth density and lower the energy penalty of moving data.

GF’s value proposition is also about ecosystem readiness. AI infrastructure suppliers need manufacturable, qualified platforms, not just lab demonstrations. Foundry-backed silicon photonics platforms could help move CPO from early adoption into broader commercial deployment by giving optical module and system companies a production path.

Sivers and GlobalFoundries Target the Optical Engine Layer

Just weeks after GF’s SCALE announcement, Sivers Semiconductors and GlobalFoundries announced a collaboration to develop advanced silicon photonics solutions for the AI infrastructure market. Sivers’ laser arrays will be integrated into reference designs built on GF’s silicon photonics platform and made available for GF’s SCALE optical engine solutions.

The collaboration supports multiple optical connectivity architectures, including co-packaged optics, linear pluggable optics and other emerging data center interconnect approaches. That flexibility is important because the industry is not moving toward one single optical architecture. Different workloads, reaches, switch designs and serviceability requirements may favor different solutions.

Wiwynn’s Computex Demonstration Shows CPO Ecosystem Formation

Wiwynn’s Computex 2026 announcement adds a systems-integration perspective. The company said it would showcase co-packaged optics (CPO) interconnect technologies for hyperscale AI data centers, working with ecosystem partners including Ayar Labs, Global Unichip Corp., Browave, Corning, FOCI, Molex, SENKO and TE Connectivity.

Demonstrating partners is the key. CPO cannot scale through isolated component innovation alone. It requires coordination among server designers, silicon photonics suppliers, connector vendors, cable providers, packaging specialists and cloud infrastructure companies.

The AI data center market needs optical technologies that can be integrated into real rack-level and system-level designs. Demonstrations that bring together multiple ecosystem players help move optical interconnects closer to deployable architectures.

Credo and DustPhotonics Bring Silicon Photonics Into the Connectivity Stack

Credo’s completed acquisition of DustPhotonics brings silicon photonics photonic integrated circuit technology to Credo, deepening the company’s optical interconnect portfolio across 800G, 1.6T and 3.2T near-packaged optics and co-packaged optics. Credo said the combination positions it with a connectivity stack spanning SerDes, DSP, silicon photonics and system integration for scale-out and scale-up networks.

The acquisition reflects the blurring line between electrical and optical interconnects. AI clusters will continue to use both, depending on distance, topology and performance requirements. Suppliers that can work across copper, optical DSPs, silicon photonics and system integration may be better positioned to serve the full AI connectivity stack.

Near-Packaged Optics Offers a Pragmatic Bridge

LightSpeed Photonics represents another path in the optical networking transition: near-packaged optics. Rather than placing optics directly inside the chip package, the company is pursuing a PCB-level optical interconnect approach that places modular optical light engines close to the processor. The concept aims to shorten copper paths without taking on the full packaging complexity of CPO.

LightSpeed’s reported 400G design is aimed at reducing power, heat and latency compared with conventional pluggable transceivers. Whether the approach becomes a major commercial category will depend on qualification, manufacturing scale and adoption by system vendors. But the idea fits a broader industry trend: optics are moving closer to compute.

For AI data centers, the potential value is practical. If near-packaged optics can reduce power-hungry retimers and DSPs while improving density and latency, it could help operators scale racks and clusters more efficiently without waiting for other technologies to mature fully.

A Shared Theme: AI Infrastructure Is Becoming Optical Infrastructure

Each of these announcements addresses a different layer of the same challenge.

The EBO MSA targets standardization and interoperability at the connector level. University of Arizona and Post Quantum Tek are targeting energy efficiency and speed at the switching layer. STL, Corning and Amazon are targeting manufacturing capacity and supply-chain readiness. Corning and NVIDIA are linking optical connectivity directly to the AI factory buildout. GlobalFoundries, Sivers, Wiwynn, Credo and LightSpeed are pushing optical technology closer to the compute and switching silicon that define next-generation AI clusters.

Together, they show that optical infrastructure is becoming a core pillar of AI data center design to address AI workloads that are network-intensive. Training and inference at scale require constant movement of data between accelerators, memory systems, storage platforms and distributed compute resources. As the size of GPU clusters grows, the network becomes both a performance determinant and an energy concern.

Copper still has a role inside data centers, particularly at short reaches and within tightly integrated systems. But the growth of AI clusters is pushing more traffic toward optical links. Higher bandwidth, longer reach, better density and lower signal degradation make optics increasingly important as AI systems scale across racks, rows, halls and campuses.

The next wave of AI data center value will come from reducing friction across the whole infrastructure stack. Power systems must deliver more capacity. Cooling systems must handle higher densities. Networks must move more data with less latency and less energy. Supply chains must deliver components fast enough to keep up with deployment timelines. 

The AI infrastructure race is increasingly becoming a race against latency. No technology will exceed the speed of light, but the winners will be those that engineer away every electrical conversion, connector, packaging constraint and supply-chain bottleneck that keeps data from moving as close to that fundamental limit as possible.

Keep pace with the fast-moving world of data centers and cloud computing by connecting with Data Center Frontier on LinkedIn, following us on X/Twitter and Facebook, as well as on BlueSky, and signing up for our weekly newsletters using the form below.