Nvidia’s next generation of AI processors, expected under the internal code name Feynman, could become early candidates for a packaging technology that replaces traditional organic substrates with glass. The shift depends on work now underway at TSMC and on a parallel U.S. government investment in domestic glass substrate manufacturing through the CHIPS for America program. At the center of that domestic effort is Absolics Inc., which is building capacity in Covington, Georgia, specifically to produce glass substrates designed for artificial intelligence and high-performance computing chips.
Why glass packaging for AI chips matters right now
The core tension behind this story is straightforward: AI accelerators are growing so large and so power-hungry that the materials holding their silicon together have become a performance bottleneck. Organic substrates, the standard for decades, warp at high temperatures and limit how tightly chip components can be packed together. They also constrain the number of power and signal connections that can be routed from the chip to the rest of the system, which in turn caps memory bandwidth and total power delivery.
Glass, by contrast, stays dimensionally stable at higher temperatures and can carry finer wiring traces with more consistent electrical properties. That translates into shorter signal paths, tighter via spacing, and potentially lower energy consumption per operation. For AI accelerators that already draw hundreds of watts per package and rely on extremely wide memory interfaces, even modest reductions in resistance and inductance across the substrate can unlock higher clock speeds or improved efficiency at the same performance level.
That physical advantage is why glass substrates are now an active, funded research and manufacturing focus tied directly to AI packaging. The U.S. government has backed this direction explicitly. Absolics’ Covington facility is part of the CHIPS program, and its stated purpose is to build domestic supply chain capacity for glass substrate technology intended for advanced packaging that increases performance and reduces power consumption for AI, high-performance computing, and data center applications.
The hypothesis that U.S.-funded glass substrate lines will reach volume production at least 12 months ahead of equivalent Asian capacity, giving early-adopter AI chips a measurable power-efficiency edge in 2026 system benchmarks, is plausible but not yet confirmed by any public timeline or production data. Absolics’ project establishes a clear intent to build that capacity on American soil, and TSMC has separately signaled interest in glass packaging for future nodes. But no primary source available today pins down when either effort will hit volume scale, or whether the U.S. facility will lead or follow Asian counterparts.
Absolics, CHIPS funding, and the SMART Packaging Program
Two distinct but related government records anchor the factual case for this shift. The first is the CHIPS for America project page for Absolics in Covington, Georgia, which identifies the facility as a supply chain investment focused on glass substrates for advanced packaging. The project description ties the technology directly to improved performance and reduced power draw in AI, high-performance computing, and data center systems. This is not a general semiconductor plant; it is purpose-built for a specific material transition that policymakers view as strategically important.
The second record comes from the SMART Packaging Program, also administered through NIST, which references Absolics and describes an effort to develop glass core substrate panel manufacturing capabilities. Panel-level manufacturing matters because it allows substrates to be produced in larger formats than traditional wafer-based processes, which can reduce per-unit costs and increase throughput once the technology matures. It also demands tight control of flatness and defect density across large areas, two characteristics that play to glass’s strengths.
Together, these two government-backed programs show that glass-core substrates have moved beyond laboratory curiosity into an active industrial development phase with federal financial support. The investment is not speculative research funding; it targets manufacturing capability, which signals that the U.S. government views glass packaging as a near-term commercial technology rather than a distant possibility. In policy terms, the projects are framed as supply chain resilience measures as much as performance bets, aiming to ensure that critical packaging materials for AI infrastructure are available from domestic sources.
For Nvidia, the relevance is direct. Each new generation of its data center GPUs pushes the limits of what organic substrates can support. The company’s current Blackwell architecture already uses enormous multi-die packages that stress existing materials, combining logic dies with stacks of high-bandwidth memory and aggressive power delivery requirements. A successor architecture, widely expected to carry the Feynman name based on Nvidia’s pattern of naming GPU families after physicists, would arrive at a point where glass substrates could offer tangible gains in signal integrity, power distribution, and thermal management.
If glass cores allow more I/O connections and cleaner high-speed signaling, Nvidia could widen memory interfaces or increase chiplet counts without proportionally increasing package size. That, in turn, could raise effective performance per accelerator or reduce the number of accelerators needed to reach a given cluster capability. From a data center operator’s perspective, either outcome-higher performance per rack unit or lower power per training run-would be meaningful.
What Nvidia and TSMC have not confirmed about Feynman
The gap between government investment records and confirmed product plans remains wide. No primary document from Nvidia or TSMC confirms that the Feynman architecture will use glass substrates, or that either company has qualified glass packaging for volume production. The Absolics project pages establish facility purpose and federal backing but do not disclose production capacity figures, yield targets, or customer commitments. The SMART Packaging Program documentation describes manufacturing development goals without publishing technical performance data or benchmarks for glass-core substrates.
Several specific questions remain open. First, glass substrates must prove they can be manufactured at yields comparable to organic alternatives before any chip company will commit to them for a high-volume product. Yield data for glass-core panels has not appeared in any publicly available government or corporate filing, leaving outside observers to infer maturity only from the scale of announced investments.
Second, TSMC’s own glass packaging research timeline is not public, so the relationship between TSMC’s internal work and the Absolics supply chain is unclear. It is not known whether Absolics aims to serve foundries like TSMC directly, to supply intermediate packaging houses, or to focus initially on domestic chipmakers and system integrators. Without that clarity, it is difficult to map a straight line from the Covington facility to any specific Nvidia product.
Third, the competitive picture in Asia is opaque. Samsung, Intel, and several Japanese materials companies have all explored glass substrates, and no available source establishes which effort is furthest along. Some may prioritize internal use for their own processors, while others could position themselves as merchant substrate suppliers. The presence of multiple parallel programs suggests that glass packaging is widely seen as strategically important, but it also means that any temporary U.S. lead could be short-lived.
The absence of confirmed timelines means the hypothesis about a U.S. production lead cannot be validated with current evidence. What can be said with confidence is that the U.S. government has placed a clear bet on domestic glass substrate manufacturing through the Absolics project and the SMART Packaging Program, and that the technology’s intended applications-AI accelerators and data center processors-align precisely with Nvidia’s product roadmap. Whether Feynman becomes the first major GPU family to ride that wave will depend on how quickly glass substrates clear the remaining hurdles of yield, integration, and cost.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
