Jensen Huang’s Texas Experiment: Can an “AI Factory” Hire a Town?
On a warm morning in Sherman, Texas, Nvidia’s Jensen Huang didn’t talk about GPUs as much as he talked about people. He stood in front of an expansion that, if it delivers, will add roughly a thousand jobs to a region that has learned to live on global supply chains but rarely gets to anchor them. The stage was not a gleaming data center but a supplier’s site—Coherent’s facility north of Dallas—where indium phosphide will be shaped into the lasers that let many chips behave, in Huang’s phrasing, like a single “AI factory.” The promise was not abstract. Nvidia disclosed a $2 billion partnership with Coherent tied to this expansion, and the headcount math was specific: about 1,000 roles including construction, roughly 550 of them in advanced manufacturing, engineering, and technical work. For a news cycle that has turned “AI” into shorthand for layoffs, the Texas bet felt like a hard pivot back to the factory floor.
The bet behind the ribbon
Huang’s claim is bracingly simple: the buildout of AI infrastructure will create, not displace, skilled manufacturing jobs in the United States. It’s not a hand-wavy forecast; it’s an experiment you can audit. Coherent plans to double its floorspace and quadruple output. Either those hires show up on payrolls and in parking lots—or they don’t. For the world’s most valuable chipmaker to pin part of its narrative to the fortunes of a single plant is unusual. But that’s the point. The company is reframing AI as industrial infrastructure, not merely cloud software with a marketing budget, and trying to prove it with shippable parts, forklift beeps, and W-2s.
The invisible bottleneck is light
Coherent’s specialty—indium phosphide for chip-to-chip optical links—sits at a quiet chokepoint in AI economics. Today’s frontier models need clusters so large that copper can’t move data fast enough without drowning in heat. Light can. Executives say these optical interconnects can cut power draw by up to half while making many GPUs behave as one contiguous machine. You can taste the second-order effects: less energy per inference, lower latency across the cluster, and, crucially, a cheaper cost per token. As the price curve drops, workloads that were a CFO’s thought experiment become viable line items. That cascade doesn’t stay in the cloud. Every point shaved off energy and interconnect overhead justifies more deployments in sectors where the work is physical—inspection lines, field service, local integration—work that can’t be offshored to a hyperscaler.
From tokens to steel-toed boots
Cheaper AI is often discussed as a threat to desk jobs. In Sherman, it’s being pitched as oxygen for factories. If your inference bill falls and throughput rises, the map of “economical to automate” expands. But automation in the physical economy rarely arrives as a pink slip delivered by a robot; it arrives as a reconfiguration: new stations, new sensors, new maintenance routines, new upstream components. Those require operators, technicians, machinists, safety specialists, and process engineers—precisely the 550 higher-skill roles Coherent says it will hire. Nvidia is arguing that AI’s appetite for bandwidth and power doesn’t just enrich data landlords; it pulls real work into towns that build the plumbing. If the plant hits its targets, you will be able to point to a place on I‑75 and say: that’s what lower token costs look like in the real world.
Industrial policy as co-author
This is not the invisible hand acting alone. The expansion blends corporate capital with political will: $33 million from the CHIPS and Science Act and an additional $17 million grant from the Trump administration. That bipartisan through-line is pragmatic. Washington wants the AI era’s supply chains—and the payrolls that come with them—rooted in the U.S. The federal posture toward AI is not uniformly permissive; we’ve watched new controls ripple through labs, including decisions that led Anthropic to shut off some public model access. At the same time, the political courtship of Nvidia is out in the open, and energy anxiety is very real. The bargain, then, is conditional: public money and policy cover in exchange for domestic capacity and headcount. If Sherman delivers, appropriators get a talking point stronger than any white paper.
What success would actually prove
If the promised thousand jobs appear as the facility doubles and output quadruples, it won’t prove that AI creates jobs everywhere. It will show something more specific and more important: that the supply chain for scaling AI has labor-intensive nodes that pay well, and that those nodes can be sited in the U.S. with intentional policy and concentrated corporate demand. It will also show that “AI factory” is more than a metaphor. A factory is a place where unit economics are disciplined by physics and logistics. If optical interconnects let clusters act as a single organism, the discipline transfers: lower variance in compute, fewer thermal penalties, more predictable costs per task. That predictability is the hinge on which real capital budgets swing.
There is also a cultural proof point at stake. Much of the past year’s AI story has been disembodied—models floating atop rented racks, features turned on and off by a blog post. Sherman makes the case that AI has a body: furnaces for wafers, cleanrooms for assembly, shipping docks for boxes that fail under human fingers if you drop them. It’s harder to dismiss that as ephemeral disruption when a town’s tax base and training programs are being reshaped to support it.
The fragilities in the promise
No narrative should outrun the meter. Construction jobs end. Ramp schedules slip. Global demand can turn on a geopolitical headline. Energy constraints and public pushback over electricity use can hard-cap throughput. And it’s easy for a flagship to become an outlier: a photogenic plant with ribbon-cuttings that don’t generalize. The test here is scale and repeatability. Do other suppliers follow with expansions of their own? Do community colleges spin up programs that feed not just this site but a regional lattice of employers? Are the wages defensible when subsidies sunset? These are the interrogatories that matter more than a single quote from a CEO, even one as quotable as Huang calling these systems “the infrastructure of the new industrial revolution.”
The scoreboard we’ll watch
Over the next 12 to 24 months, the numbers will either backstop the rhetoric or puncture it. Hiring logs should show ~1,000 new roles, with a majority of the permanent headcount in skilled factory work. Output metrics should move toward the promised 4x. Power draw per unit of optical bandwidth should trend down in line with the 50% reduction thesis. If those conditions are met, expect competitors to mirror the move, and expect the political support to harden, regardless of who occupies the White House. If they’re not, the industry will have to own that the clearest, numbers-backed attempt to tie AI’s growth to domestic manufacturing jobs fell short.
For a publication like this one, the drama is not in the press release but in the payroll. Sherman is where AI’s employment thesis gets audited. If Nvidia’s supplier strategy can mint durable, higher-skill roles alongside cheaper, denser compute, then “AI replaced me” starts to sound less like a eulogy for labor and more like a relocation notice—from the back office to the plant floor. The difference will be visible in a parking lot at shift change. And that, finally, is a metric you don’t need a benchmark to understand.