Kopin Analysis Finds Optical Interconnect Architecture Could Break the Data Center “Copper Wall”

New white paper shows how moving beyond copper to massively parallel MicroLED-based optical channels could improve GPU utilization and reshape future AI data centers. Join Kopin’s live “Ask Me Anything” discussion on October 20, 2026, for insights into the findings and the future of AI.

WESTBOROUGH, Mass.--(BUSINESS WIRE)--Kopin Corporation (Nasdaq: KOPN), a leading provider of application-specific optical systems and high-performance microdisplays, today released a white paper with Fabric.AI examining the potential of Neural I/o™ to address constraints of future AI systems. The analysis models that this approach could increase useful training throughput by as much as 1.6 times, allowing the same number of GPUs to complete the same workload in up to 38% less time.

“AI infrastructure’s ability to connect compute efficiently is becoming as important as the compute itself,” said Michael Murray, Chief Executive Officer of Kopin Corporation. “The industry has invested heavily in increasing the performance and scale of AI processors, but those processors ultimately have to operate as a system. When the physical limitations of today’s interconnect no longer dictate that system’s architecture, our analysis shows the potential for better utilization of existing compute and much greater freedom in how future AI infrastructure is designed.”

A new white paper, “The Data Center in a SerDes-Free World,” authored by Kopin CEO Michael Murray, Kopin Sr. VP of Engineering Dr. John Callahan and Fabric.AI Senior Strategic Advisor James Altucher, examines the potential of Neural I/o at the scale of an entire AI system. The analysis moves beyond link-level performance to explore how optical interconnects could improve GPU utilization and change the physical architecture of future AI data centers.

A central finding is that moving beyond the rack as the defining unit of AI compute could change the physical constraints that determine how switching and cooling infrastructure are organized. Neural I/o™ provides the technical basis for that shift, replacing a small number of high-speed serialized electrical links with massively parallel MicroLED-based optical channels. By extending high-bandwidth connectivity from approximately one meter to tens of meters, the architecture could expand a tightly connected compute domain from roughly 72 GPUs today to row-scale configurations spanning hundreds or more than 1,000 GPUs.

Key findings include:

• More useful training throughput from the same GPUs. For a 10,000-GPU system, the architecture is modeled to deliver 1.3 to 1.6 times greater useful training throughput per GPU, equivalent to approximately 23% to 38% less training time for the same workload. The modeled gains come from reducing communication, synchronization and reliability-related losses rather than increasing GPU compute performance.
• Greater flexibility in how AI infrastructure is built. Larger high-bandwidth compute domains could reduce the number of boundaries across which data must move and allow switching infrastructure to be separated from compute racks, creating more flexibility in how compute, switching and cooling are distributed across the data center.
• Parallel optical capacity for system coordination and reliability. Neural I/o’s modeled architecture provides 3.2 terabits per second of payload bandwidth and allocates additional optical channels to redundancy, synchronization, timing, telemetry and control. That capacity could help coordinate GPUs, absorb individual channel failures and monitor the health of the interconnect without competing with primary AI workloads.
• A path toward building-scale AI systems. The paper’s longer-term “Cube” concept models approximately 100,000 GPUs organized into roughly 1,000-GPU rows connected through an optical spine, exploring how future data centers could operate as more closely coordinated computing systems rather than collections of smaller clusters.

The findings broaden the potential role of Kopin’s MicroLED and optical technologies from individual interconnect links to the architecture of larger AI systems. Kopin and Fabric.AI have begun initial transmission-link testing of Neural I/o, with a public technology demonstration targeted for January 2027 and first-generation production silicon targeted for 2028. These milestones represent the next steps toward evaluating how the technology can scale from individual links to increasingly complex AI infrastructure.

Performance figures for systems that do not yet exist, including training throughput, utilization, latency and future system architectures, are modeled results based on assumptions detailed in the white paper and are not measured performance results from production systems.

The full white paper, “The Data Center in a SerDes-Free World: From the Copper Wall to the Row — and From the Row to One Machine the Size of a Building,” is available here: https://www.kopin.com/the-data-center-in-a-serdes-free-world-lander/

Forward Looking Statements

This press release includes statements that constitute forward looking statements made pursuant to the safe harbor provisions of the Private Securities Litigation Reform Act of 1995. Forward looking statements may be identified by the use of words such as “forecast,” “guidance,” “plan,” “estimate,” “will,” “would,” “project,” “maintain,” “intend,” “expect,” “anticipate,” “prospect,” “strategy,” “future,” “likely,” “may,” “should,” “believe,” “continue,” “opportunity,” “potential,” and other similar expressions that predict or indicate future events or trends or that are not statements of historical matters.

Forward looking statements are only predictions and are subject to risks, uncertainties, and assumptions that are difficult to predict and, in many cases, outside of our control. Actual results may differ materially from those expressed or implied in any forward looking statements. All forward looking statements are qualified in their entirety by these cautionary statements. We undertake no obligation to update forward looking statements except as required by law. Important factors that could cause actual results to differ materially from those described in forward looking statements are detailed in the Company’s most recent form 10-K and other filings with the Securities and Exchange Commission.

In addition to the risks described in the Company's most recent Form 10-K and subsequent SEC filings, factors that could cause actual results to differ materially include, but are not limited to: the Company's ability to successfully develop, test and commercialize Neural I/o™ and related optical interconnect technologies; the timing and success of research and development activities; the ability to achieve projected technical performance, scalability, reliability and manufacturing objectives; uncertainties associated with modeled performance estimates and assumptions; the results of ongoing transmission-link testing and future demonstrations; the ability to develop production silicon and bring products to market on anticipated schedules, if at all; customer adoption of new data center architectures and optical interconnect solutions; competitive technologies and industry standards; supply chain, manufacturing and cost challenges; the availability of strategic partners and customers; rapid changes in artificial intelligence infrastructure requirements; and general economic, market and industry conditions.

About Kopin

Kopin Corporation (Nasdaq: KOPN) is a leading developer and provider of innovative display, Optical Interconnect devices and application-specific optical solutions sold as critical components and subassemblies for defense, enterprise, professional and consumer products. Kopin’s portfolio includes microdisplays, display modules, eyepiece assemblies, image projection modules and vehicle mounted and head-mounted display systems that incorporate ultra-small high-resolution Active Matrix Liquid Crystal displays (AMLCD), Ferroelectric Liquid Crystal on Silicon (FLCoS) displays, MicroLED displays (µLED) and Organic Light Emitting Diode (OLED) displays, a variety of optics and low-power ASICs and optical interconnect devices for data centers.

For more information, please visit Kopin’s website at www.kopin.com.

Kopin is a trademark of Kopin Corporation.

Follow us on LinkedIn, X and Facebook.