America’s lead in advanced computational systems rests on more than chips and data centers. It depends on the high-speed networking fabric that connects those systems into coherent clusters capable of training frontier models and delivering inference at scale. Switches, routers, optical transceivers, and the underlying silicon and photonics form the circulatory system of the machine-intelligence era. Yet critical portions of that system—especially the optical modules that convert electrical signals into light for fiber—are heavily concentrated in foreign manufacturing, particularly China.
That concentration is a strategic vulnerability. The United States cannot remain dominant in machine intelligence while its infrastructure buildout depends on supply chains an adversary can constrain, delay, or compromise. The American MI Networking Initiative is our proposal to treat data networking equipment as critical infrastructure and to rebuild domestic manufacturing capacity at the scale required.
The core claim
Domestic manufacturing of high-performance networking gear—especially optical transceivers and related components—is essential for secure, scalable machine-intelligence infrastructure. The United States already has the network engineers and designers. What it lacks is sufficient high-volume production capacity and a clear public-benefit framework that earns community support for new data centers.
Why We Say Machine Intelligence (MI), Not Artificial Intelligence (AI)
This initiative is one concrete illustration of a larger point we make throughout our work: Machine Intelligence is the more accurate and useful term.
“Artificial Intelligence” carries baggage. “Artificial” suggests imitation, artifice, or something less real than human cognition. In practice the label has become a marketing umbrella that confuses narrow statistical tools, generative models, large-scale inference systems, and speculative future systems. Policy, investment, and public discussion suffer when the same two letters are asked to cover everything from a recommendation engine to systems that already shape critical infrastructure, national security, and economic capacity.
Machine Intelligence is clearer. These are engineered computational systems that perform intelligent functions—perception, prediction, planning, generation, optimization—at scales and speeds no biological system matches. They run on physical infrastructure. They consume power, move data, and create dependencies. They require governance, auditability, and resilient supply chains. Calling them Machine Intelligence keeps the focus on what they actually are: machines whose capabilities and risks grow with the quality of the hardware, the software stack, and the physical networks that bind them together.
The networking layer makes the distinction practical. AI rhetoric often stops at models and chips. MI forces attention to the full stack—including the switches, optics, and interconnects without which the models cannot operate at frontier scale. When we speak of an American MI Networking Initiative, we are naming the problem accurately: securing the physical foundation of machine intelligence systems, not merely “AI” as a brand.
This is only one reason among many. Terminology shapes what policymakers notice, what engineers design for, and what the public is asked to govern. We use MI deliberately, just as we propose a purpose-built Machine Intelligence Operating System (MIOS) rather than bolting safety onto general-purpose kernels after the fact. For a further example, see Self-Revising MI: Why Machine Intelligence That Rewrites Itself Needs a Plain-English Name.
Why Networking Gear Matters as Much as the Accelerators
Modern machine-intelligence clusters move enormous volumes of data between accelerators with extremely low latency. High-speed Ethernet fabrics, specialized interconnects, and optical links are not optional. They determine whether a large installation functions as a single coherent system or as fragmented, slower components. Optical transceivers sit at the end of nearly every fiber link inside these facilities. Without reliable, high-volume supply of these modules—and the switches and routers that use them—the physical expansion of U.S. capacity slows or stalls.
The United States already possesses world-class network engineers and designers. American companies and research groups lead in the architecture, silicon design, systems software, and operational expertise required for the highest-performance data-center fabrics. The talent that invents the protocols, programs the ASICs, designs the switches and routers, and runs the networks at hyperscale already exists here. What is missing is sufficient high-volume domestic manufacturing capacity—especially for optical modules and certain related components—currently concentrated abroad. Chinese firms have accounted for a large majority of global optical transceiver units and manufacturing capacity in recent assessments. Lead times for advanced switches already stretch many months amid broader component shortages. Power, cooling, and site constraints already limit data-center growth; networking bottlenecks compound them.
Security and supply-chain concerns are real. Untrusted hardware in the heart of the nation’s compute infrastructure creates risks of disruption, exposure, or leverage. The correct response is not permanent dependence. It is deliberate, accelerated domestic production of the networking gear that machine-intelligence systems require—leveraging the engineers and designers America already has.
The Political Reality: Voters, Data Centers, and Public Benefit
The 2026 midterm elections made something unmistakable: a significant number of voters and candidates expressed clear opposition to new data-center construction. On November 3, Ballotpedia counted 41 local data-center ballot measures across ten states, and voters favored limits in 40 of them, as we detail in Voters Didn’t Reject AI. They Rejected Building It Backward. Communities have raised legitimate concerns about electricity prices, water use, land, noise, traffic, and the sense that large projects primarily enrich a narrow set of companies while imposing costs on local residents and ratepayers. Ignoring those concerns is not a viable strategy. It breeds further resistance and undermines the social license required for the infrastructure the country needs.
The right response is not to abandon new capacity. It is to pair data-center expansion with tangible public benefits—beginning with domestic manufacturing that creates real jobs. When networking equipment, optics, and related components are manufactured in the United States, the economic footprint of the machine-intelligence buildout expands beyond a few hyperscale campuses. Local manufacturing jobs, supplier networks, workforce training, and tax bases become part of the story. Projects that deliver visible community benefit are far more likely to earn sustained public support than projects that appear purely extractive.
New data-center construction must serve more than private profit for a select few. It should strengthen national capability, create broadly shared economic opportunity, and operate under clearer public accountability. Domestic manufacturing of critical networking gear is one concrete way to deliver that broader benefit. Voters are not wrong to demand it. Policymakers who dismiss the midterm signal will find future projects harder, not easier, to site and approve.
Four realities that make this urgent
- National infrastructure. Machine-intelligence facilities support research, defense, and the broader digital economy. A foreign choke point on networking gear is a foreign choke point on American capacity.
- Single points of failure. Supply-chain concentration multiplies risk. Continuity of government and critical services require resilient, auditable foundations.
- Manufacturing compounds. Because design talent is already domestic, the primary task is scaling production. Delay locks in dependence.
- Political sustainability requires public benefit. The midterms showed that pure private-profit framing is insufficient. Pairing data-center growth with domestic manufacturing jobs is the practical path to sustained support.
What the American MI Networking Initiative Would Do
The Initiative treats high-performance data networking equipment as a strategic industrial priority. Core elements include:
- Targeted industrial capacity. Expand and incentivize U.S. manufacturing of optical transceivers, high-speed switches and routers, silicon photonics, and related components. Build on existing American and allied strengths—and on the network engineers and designers already working in the United States—while closing the volume gap that currently leaves large operators reliant on high-risk sources. The resulting jobs and supplier networks help address the very concerns voters expressed in 2026.
- Trusted supply-chain standards. Establish clear requirements so that operators of the largest machine-intelligence facilities can demonstrate secure, diversified, and increasingly domestic sourcing. Public reporting on material dependencies should become routine.
- R&D for the next generation. Accelerate work on co-packaged optics, lower-power architectures, and designs that improve both performance and security properties. Domestic manufacturing should enable the next leap, not merely replicate today’s modules. American engineers are already positioned to lead that work.
- Public-benefit requirements. New large-scale data-center projects should be evaluated not only on private returns but on measurable contributions to domestic manufacturing capacity, local employment, grid and water accountability, and long-term community benefit. Ignoring voter concerns is not a solution; designing projects that respond to them is. Our walk-through of how a data-center deal actually gets made shows where the public is left out today.
- Integration with broader governance. Networking cannot be separated from power, cooling, land use, workforce, and operating-system questions. Our Data Centers series examines these constraints. An elected technology leadership structure—what we call a Department of Technology—would be positioned to coordinate across agencies, track cross-sector risks, ensure public accountability, and help ensure that machine-intelligence infrastructure serves the broader public rather than a narrow set of interests. Such leaders need hands-on network experience, as we argue in Back to Basics: Why Our Technology Leaders Must Know Networks.
This is not a call for isolation. Trusted allies remain essential. It is a call to eliminate single-country choke points on the hardware that makes American machine intelligence possible at scale, while fully employing the design and engineering talent the country already possesses and delivering the public benefits voters have every right to expect.
Accountability and the Public Interest
When complex systems fail in public view, the public deserves clear answers. The same principle applies one layer deeper. When the physical networking layer that underpins machine-intelligence clusters depends on concentrated foreign production, citizens have a right to know the risk posture and the plan to reduce it. When communities are asked to host new data centers, they have a right to projects that create shared value rather than concentrated private gain.
A purpose-built Machine Intelligence Operating System can embed safety constraints, resource accounting, and auditability at the kernel level. Those properties mean little if the underlying interconnects are opaque or vulnerable to supply disruption, or if the facilities themselves lack public legitimacy. Openness and auditability at the software layer must be matched by resilience, transparency, and genuine public benefit at the hardware and siting layers.
The Path Forward
The United States still leads in model development, advanced accelerators, systems design, and network engineering. That lead is not self-sustaining. It requires the physical capacity to manufacture and wire the next generation of clusters without depending on an adversary’s factories—and the political capacity to build those facilities with the consent of the communities that host them.
The American MI Networking Initiative is a practical framework for closing both gaps: industrial policy focused on the specific components that currently constrain and expose the machine-intelligence buildout, paired with accountable technology leadership that can look across sectors, respond to legitimate voter concerns, and report to the public. Because the engineers and designers are already here, and because domestic manufacturing creates the jobs and broader benefits that can help earn public support, the opportunity is concrete and achievable.
Questions for policymakers
- What share of the optical modules and high-speed networking gear entering new U.S. facilities originates from high-risk sources?
- What is the realistic timeline and cost to establish sufficient trusted manufacturing capacity?
- How will new data-center projects demonstrate measurable public benefit—including domestic job creation—rather than private profit alone?
- Which office is responsible for tracking these dependencies and publishing unclassified assessments?
Precise language matters. Calling these systems Machine Intelligence rather than Artificial Intelligence keeps the focus on engineered systems, physical infrastructure, and the governance they require. The networking initiative is one demonstration of that principle in action. Responding to the clear signal from the 2026 midterms is another.
Readers who want the broader governance model can start with our explanation of what a Department of Technology is and our Data Centers series. The foundation of American machine-intelligence capacity is still being built. It should be built with American—and trusted allied—networking gear, designed and increasingly manufactured by the talent already working in the United States, and structured so that the public shares in the benefit.
We will refine this proposal with further technical and policy detail. The conversation is open.
Explore related work
New to our proposal?
Learn what a Department of Technology is, and how elected technology leaders would answer to voters when systems fail and infrastructure is built.
Read the basics
Leave a Reply