Intel has long stood at the center of semiconductor progress, and its chief executive Pat Gelsinger now believes the company can extend the gains once predicted by Moore’s Law through an entirely new approach centered on light rather than electricity. In a recent conversation reported by WIRED, Gelsinger outlined how silicon photonics and related optical technologies could keep performance climbing even as traditional transistor scaling faces physical limits. The discussion highlights both the technical promise and the commercial stakes involved in moving data with photons inside chips and across data centers.
Moore’s Law, the observation that the number of transistors on a chip roughly doubles every couple of years while costs fall, has guided the industry for decades. Yet the cadence has slowed. Shrinking transistors further demands extreme ultraviolet lithography, exotic materials, and enormous capital investment. Heat dissipation and signal delays also grow problematic at nanometer scales. Gelsinger argues that optics can relieve many of these pressures by handling data movement, a task that now consumes most of the energy and latency inside modern systems.
The core idea is to replace copper wires with light paths. Photons travel with minimal resistance, generate almost no heat, and can carry multiple signals at different wavelengths along the same fiber or waveguide. Intel has spent more than a decade developing silicon photonics components that integrate lasers, modulators, detectors, and waveguides directly onto the same silicon substrate used for conventional processors. This integration matters because it allows the chip industry to draw on existing manufacturing infrastructure rather than starting from scratch with exotic materials such as indium phosphide.
Gelsinger pointed to several practical demonstrations already in the market or nearing production. Intel’s silicon photonics transceivers have shipped in volume for data center interconnects, moving data at 100 gigabits per second and beyond between servers. These modules convert electrical signals to optical ones at one end and back again at the other, slashing power consumption compared with purely electrical cables. The next frontier, according to the Intel leader, lies in bringing the optical connection inside the package, directly beside or even beneath the compute die.
Such co-packaged optics could dramatically widen the bandwidth available to processors. Today a high-end CPU might connect to memory or accelerators through hundreds of electrical lanes, each limited in speed and distance. Optical links could supply terabits per second with far fewer pins and lower energy per bit. That improvement would matter enormously for artificial intelligence workloads, where models now demand constant shuffling of massive tensors between compute units, memory banks, and storage pools. Reducing the energy spent on data movement would free more power for actual calculations.
Intel’s strategy involves stacking optical interconnect layers using established 2.5D and 3D packaging methods. The company has demonstrated working prototypes that embed optical engines within the same substrate as the main processor. Light is generated off-chip by external lasers and coupled into the silicon waveguides, an approach that avoids the difficult task of building reliable lasers directly on silicon. Once inside the chip, the light is modulated, routed, and detected with components fabricated alongside ordinary transistors. Gelsinger noted that these optical engines can be tested and burned in separately before final assembly, preserving overall yield.
The technical challenges remain substantial. Precise alignment of optical fibers or free-space beams to sub-micron waveguides demands advanced assembly tools. Temperature fluctuations can shift wavelengths and degrade signal integrity, requiring active control systems. Packaging must protect delicate photonic structures while allowing efficient heat extraction from the compute die. Intel has invested in automated assembly lines and metrology equipment to address these issues at manufacturing scale. The company also collaborates with partners across the supply chain, from laser suppliers to contract manufacturers, to refine processes that were once confined to research laboratories.
Beyond the package, Gelsinger sees optical communication scaling out to entire racks and eventually across data center halls. Traditional electrical switches hit bandwidth walls as port counts and speeds increase. Optical switches that route entire wavelengths without converting back to electricity could reduce latency and power draw in large clusters. Intel has prototyped such switches and is working with hyperscale operators to test them in real networks. The long-term vision includes a disaggregated architecture where compute, memory, and storage live in separate pools connected by a high-speed optical fabric. Resources could then be allocated on demand, improving overall utilization and simplifying upgrades.
This optical future aligns with Intel’s broader technology roadmap. The company continues to advance its transistor processes, recently introducing RibbonFET gate-all-around devices and PowerVia backside power delivery. Those innovations address the front-end scaling challenges, while photonics tackles the back-end communication bottlenecks. Gelsinger emphasized that both paths must advance together. A faster transistor provides little value if the data cannot reach it quickly enough or if moving the data burns more energy than the calculation itself.
Competitive pressure adds urgency. TSMC, GlobalFoundries, and Samsung continue to push process nodes, while specialized AI accelerator companies such as Nvidia and AMD design ever-larger chiplet-based systems. Startups focused exclusively on silicon photonics, including Ayar Labs and Lightmatter, have raised significant funding to pursue similar optical interconnect ideas. Intel’s advantage lies in its ability to combine photonics with its own manufacturing, packaging, and high-volume assembly expertise. The company’s foundry business now offers silicon photonics as a service to external customers, aiming to create an industry-wide platform rather than a proprietary solution.
Gelsinger acknowledged that adoption will not happen overnight. System architects must redesign boards, firmware, and software stacks to take full advantage of optical links. Standards bodies are still defining common interfaces for co-packaged optics. Test and reliability practices need to evolve to cover both electrical and photonic failure modes. Nevertheless, he expressed confidence that the economic incentives are strong enough to drive change. Hyperscalers facing rising electricity bills and constrained power delivery in their facilities have clear motivation to embrace any technology that lowers energy per operation.
Research from universities and national laboratories has provided foundational knowledge that Intel has translated into products. Work on photonic integrated circuits at institutions such as MIT, Stanford, and the University of California system helped establish the device physics. Defense Advanced Research Projects Agency programs on optical computing and interconnects supplied early prototypes and talent. Intel itself has published hundreds of technical papers detailing its progress on modulators with low drive voltage, germanium detectors compatible with silicon CMOS, and low-loss waveguide routing.
Looking further ahead, Gelsinger sketched a future in which light handles not only communication but also certain computations. Optical neural networks could perform matrix multiplications at the speed of light using interference patterns, potentially accelerating inference tasks with minimal power. While such analog approaches face precision and training challenges, hybrid systems that combine digital electronics for control with optics for heavy linear algebra may emerge. Intel is investigating these possibilities through its research labs while keeping primary focus on interconnects that can be commercialized in the nearer term.
The financial commitment required is considerable. Building new fabs, developing packaging flows, and sustaining years of product qualification demand tens of billions of dollars. Intel has secured support through the CHIPS and Science Act in the United States and similar incentives in Europe. Those funds help offset the risk of pioneering technologies that may take longer to reach volume production than traditional roadmaps. Gelsinger stressed that the investment also positions the United States and its allies to maintain leadership in an area critical to both commercial competitiveness and national security.
Customers have begun to signal interest. Cloud providers have requested samples of optical interconnect prototypes for integration studies. Supercomputer projects exploring exascale systems have identified photonics as a potential solution for the massive all-to-all communication patterns required by scientific simulations. Even automotive and industrial markets could benefit if optical links prove more resistant to electromagnetic interference than copper in harsh environments.
Of course, photonics will not replace electronics entirely. Transistors remain essential for logic and memory. The most likely outcome is a heterogeneous integration where light and electrons each perform the tasks they do best. Electrical signaling will continue inside localized areas of a chip where distances are short and density is paramount. Optical channels will span longer distances inside packages, between packages, and across the data center floor. This division of labor could sustain performance gains for another decade or more, effectively extending the spirit of Moore’s Law even if the literal transistor count per square millimeter grows more slowly.
Gelsinger’s optimism rests on decades of engineering progress that turned what once seemed like laboratory curiosities into reliable, high-volume components. The same industry that figured out how to print billions of transistors on a fingernail-sized die has now learned to guide light across that same surface with precision measured in nanometers. The transition from copper to optics will not be instantaneous, yet each successive generation of silicon photonics products brings the vision closer to reality.
Intel continues to hire specialists in photonics, materials science, and high-speed packaging to accelerate development. University partnerships feed the talent pipeline, while internal research teams explore novel materials such as barium titanate for more efficient modulators. The company’s manufacturing scale provides a unique test bed: once a process is proven on a few wafers, it can be ramped across thousands. That capability differentiates Intel from pure-play photonics startups and gives system designers confidence that supply will meet demand when designs reach production.
As data hunger continues to grow, driven by large language models, scientific computing, and real-time analytics, the pressure to move bits more efficiently will only increase. Optical interconnects offer a path that sidesteps many limits of electrical signaling while building on the silicon platform the industry already understands. Pat Gelsinger’s message is clear: the laws of physics may constrain transistors, but they still allow light to travel fast and cool. Intel intends to harness that advantage to keep delivering faster, more efficient systems for the next generation of computing demands. The coming years will test whether the industry can convert these prototypes and pilot lines into the standard building blocks of data centers worldwide.
WebProNews is an iEntry Publication