U.S. semiconductor policy is expanding beyond the buildout of domestic wafer manufacturing capacity to include data transmission, memory access, advanced packaging, materials, and supply chain security across AI computing systems. On July 29, 2026, the U.S. Department of Commerce announced that it had signed letters of intent with seven companies and planned to provide up to US$874 million in federal incentives under the CHIPS and Science Act to accelerate semiconductor research and development for next-generation computing and artificial intelligence.
The funding remains at the letter-of-intent stage. The seven companies must complete the Department of Commerce’s due diligence and formal review process before signing final agreements, and the amounts ultimately approved and disbursed could be lower than the stated ceilings. The Department of Commerce will also receive minority, non-controlling equity stakes in the companies, distinguishing these projects from conventional one-way subsidies.
The Three Largest Projects Receive Nearly 80% of the Funding, Directly Targeting AI System Bottlenecks
The seven projects cover integrated photonics, AI memory, advanced packaging, new computing architectures, dielectric materials, supply chain verification, and optical interconnect components. GlobalFoundries, Kepler Computing Inc. (hereafter referred to as Kepler), and Multibeam Corporation are expected to receive a combined US$685 million, accounting for approximately 78.4% of the total. This shows that policy resources are primarily concentrated on three major AI computing bottlenecks: data transmission, memory access, and chip integration.
| Company |
Proposed Incentive Ceiling |
Main R&D Focus |
Problem the Project Seeks to Address |
| GlobalFoundries |
US$300 million |
Silicon photonics, near-packaged optics, and co-packaged optics |
Increase AI chip interconnect bandwidth and reduce data transmission power consumption |
| Kepler Computing Inc. |
US$245 million |
New AI memory combining 3D integration and ferroelectric technology |
Improve memory bandwidth, access performance, and energy efficiency |
| Multibeam |
US$140 million |
Multi-chip assembly, stacking, and interconnection using thousands of wires |
Strengthen Chiplet and heterogeneous integration capabilities |
| Extropic |
US$75 million |
Thermodynamic sampling units |
Use less energy for probabilistic computing, simulation, and optimization |
| Thintronics |
US$50 million |
Ultra-low-loss interlayer dielectric materials |
Reduce signal loss in high-speed interconnects and advanced packaging |
| OBSIDIA Semiconductors |
US$34 million |
Non-invasive component verification technology |
Identify counterfeit or maliciously modified electronic components |
| Aeluma |
US$30 million |
Large-format, indium-phosphide-free substrate technology |
Support photodetectors, laser components, and AI optical interconnects |
Note: The amounts above are the maximum proposed amounts stated in the letters of intent and are not final approvals or disbursed funds.
Silicon Photonics and Optical Interconnects Become the Largest Single Investment Area
GlobalFoundries is expected to receive up to US$300 million, making it the largest individual project in the package. Through this funding, the U.S. Department of Commerce aims to accelerate the development of near-packaged optics (NPO) and co-packaged optics (CPO) in the United States by two to three years.
Traditional AI servers mainly use electrical signals to transmit data among processors, memory, and switches. As transmission distances, bandwidth requirements, and computing cluster sizes increase, copper interconnects face growing pressure from signal attenuation, power consumption, and heat dissipation. Silicon photonics instead uses optical signals to transmit data and places optical components close to computing chips, shortening the electrical signal path while improving bandwidth density and energy efficiency.
GlobalFoundries’ project covers next-generation silicon photonics wafers, new optical materials, 3D hybrid bonding, and advanced packaging. The related research and development will be conducted at the company’s facilities in Malta, New York, and Burlington, Vermont. The company stated that under a separate agreement, the Department of Commerce is expected to acquire an equity stake of approximately 1%, although the final ownership arrangement and incentive terms remain subject to confirmation in the final agreement.
Optical interconnects have therefore moved beyond being merely a communications component issue and have become a core technology affecting AI data center computing density, heat dissipation, and electricity costs. As the energy consumed by moving data continues to rise, overall system performance may remain constrained by interconnect bandwidth and power consumption even when processor computing power improves.
Memory and Advanced Packaging Determine Whether Computing Capacity Can Be Fully Utilized
The second-largest project is expected to go to Kepler, with proposed incentives of up to US$245 million. The company plans to use 3D integration and ferroelectric technology to develop a new type of high-performance AI memory.
The computing requirements of large AI models are increasing rapidly, but system performance does not depend solely on the theoretical computing power of GPUs or other accelerators. If memory capacity, bandwidth, and data access speeds fail to improve at the same pace, processors may be unable to operate at full capacity while waiting for data, creating the so-called “memory wall.” The U.S. government therefore hopes to cultivate new material and architectural approaches beyond existing high-bandwidth memory, reducing the time and energy costs associated with data movement.
Multibeam is expected to receive up to US$140 million to develop multi-chip assembly, stacking, and high-density wire interconnection technologies. As the cost of advanced process nodes continues to rise, AI processors are increasingly relying on Chiplet designs and heterogeneous integration. Computing, memory, communications, and input-output functions are manufactured separately and then combined into a complete system through advanced packaging. Interconnect density, manufacturing yield, and heat dissipation in packaging will therefore directly affect product performance and mass-production costs.
Materials, New Computing Architectures, and Supply Chain Verification Fill Upstream Gaps
The remaining four projects address materials, optical components, new computing architectures, and supply chain security. Thintronics will develop ultra-low-loss interlayer dielectric materials to reduce signal loss in high-speed interconnects and advanced packaging. Aeluma will develop large-format, indium-phosphide-free substrate technology to support the production of photodetectors and laser components.
Extropic plans to use natural thermal fluctuations to develop thermodynamic sampling units, with the aim of completing AI, simulation, and optimization tasks at lower power consumption. OBSIDIA Semiconductors will develop non-invasive component verification technology to identify counterfeit or maliciously modified electronic components. As AI chips are increasingly deployed in data centers, defense systems, and critical infrastructure, component origin, manufacturing history, and authenticity are becoming part of supply chain security.
The CHIPS Act Shifts From Fab Construction Subsidies Toward Investment in Critical Technologies
The initial policy focus of the CHIPS and Science Act was to attract companies to build or expand wafer fabs in the United States through subsidies, loans, and tax credits. Approximately US$39 billion was allocated to incentives for manufacturing facilities and equipment, while another roughly US$11 billion was directed toward the semiconductor R&D ecosystem.
The US$874 million package is not primarily intended to expand existing wafer production capacity. Instead, it supports technologies that are not yet fully mature but could shape the architecture of next-generation AI systems. This indicates that implementation of the CHIPS and Science Act is extending beyond manufacturing capacity expansion toward critical research and development capabilities.
The government’s acquisition of minority, non-controlling equity stakes also moves the role of public funding closer to strategic investment rather than one-way subsidies. If supported companies successfully commercialize their technologies, increase in valuation, or are acquired, the government’s equity holdings could generate returns for taxpayers. However, this model also makes policy implementation more complex. The government will need to manage equity valuation, exit mechanisms, and conflicts of interest while avoiding overlap among subsidy review, industry regulation, and shareholder interests.
Commercialization Progress Will Determine the Actual Impact of the Policy Investment
Most of the projects remain in the research or early commercialization stages. Their actual impact will depend on manufacturing yields, costs, system integration, customer validation, and mass-production capabilities. Subsequent reviews could also change the funding amounts, equity arrangements, or implementation terms of individual projects.
Compared with the construction of large wafer fabs, the US$874 million funding package is limited in scale, but its investment direction sends a clear policy signal. As the standalone performance of advanced processors continues to improve, the focus of competition is gradually shifting toward chip-to-chip communications, memory access, packaging integration, and energy efficiency. The United States hopes to use government capital to shorten the development timelines of these critical technologies and retain domestic capabilities in photonics, memory, packaging, materials, and verification.