Detroit Warehouse Seeks to Out‑Produce Century‑Old Steel Mills With Three Employees

The Money and the Machine

Nox Metals, a three-person team in a Detroit WWII-era warehouse, just closed an $11.5 million seed round led by Hyperion with Palmer Luckey, Y Combinator, Jared Friedman, RoboStrategy, Operator Collective, DTX Ventures, and Alumni Ventures joining — betting that AI-run metal factories can reshore defense supply. Founder and CEO Zane Hengsperger told Crain's Detroit Business he expects the round to reach $14 million. The company, seven months old at announcement, had already hit $330,000 in annual revenue with three employees while bootstrapping through Y Combinator's summer 2025 batch. GetLatka's data confirms the revenue figure.

The pitch centers on a 30,000-square-foot facility in Southwest Detroit that once fed the Arsenal of Democracy. Hengsperger said the company will revitalize the plant this summer, staffing it with what he calls "techno-industrialists" — a hybrid of robotics engineers, AI specialists, and metallurgists. The goal: vertically integrated factories run by AI, designed to eliminate human bottlenecks in raw-material delivery. Nox Metals currently supplies custom-cut aluminum plate and bar to roughly 100 customers nationwide, from small machine shops to defense contractors including Anduril and SpaceX. By utilizing AI automation, the company claims it can cut lead times from days to seconds.

Hengsperger founded Nox Metals in 2025 with a stated mission to reindustrialize the US supply chain. The seed funding will scale the team and expand the Detroit headquarters. Whether a startup that began as a SaaS business can translate software velocity into heavy-industry throughput remains the open question the round was designed to answer.

How an AI Factory Would Work

Nox Metals has not publicly disclosed its process stack, sensor architecture, or model training pipeline. The company's website and seed-round announcements describe the vision — AI‑controlled metal production in Detroit, but omit the technical specifics that would let engineers evaluate feasibility. That silence is standard for a pre‑revenue startup still buying its first induction furnaces in a sector where incumbents guard process know‑how as trade secrets.

The clearest public parallel is the U.S. Air Force's AI‑FORGE program. A robot prototype developed at The Ohio State University with the Air Force Research Laboratory, the Advanced Robotics for Manufacturing Institute, CapSen Robotics, and Yaskawa Motoman arrived at Warner Robins Air Logistics Complex in late January 2023 and completed a six‑hour field test forging simple parts under depot‑like conditions. The system's objective, as described by Dr. Michael Groeber of Ohio State, is to "make shapes but also control how we get there, in order to produce repeatable results" — a formulation that maps directly to the three subprocesses Nox must automate: smelting, casting, and quality control.

In smelting, the control problem is thermal and chemical. Electric arc and induction furnaces already run on PID loops; the AI layer would replace fixed setpoints with a model that ingests real‑time spectrometer feeds, off‑gas analysis, and power‑draw waveforms to adjust charge mix, power profile, and slag chemistry on the fly. The Air Force research highlights the decisions a human blacksmith makes continuously — "how to orient the component, where to apply force, how much heat to use, and when to put it back into the furnace", and those same variables exist at furnace scale: scrap grading, oxygen lance timing, carbon injection, and tap‑to‑tap cycle optimization. No public paper shows a closed‑loop AI controller doing this end‑to‑end in a commercial mini‑mill, though researchers at MIT's Materials Systems Laboratory and at Tata Steel's European pilots have published reinforcement‑learning approaches for individual setpoints.

Casting adds geometry and solidification physics. The AI‑FORGE system uses force‑controlled manipulation and in‑process thermography to close the loop between deformation and microstructure. Translating that to a continuous‑cast or near‑net‑shape casting cell means fusing mold‑level thermal imaging, ladle‑nozzle flow models, and strand‑guide actuation into a single policy that minimizes segregation, centerline shrinkage, and surface cracks. Yaskawa Motoman's involvement suggests industrial robot arms — likely high‑payload, hollow‑wrist models, handle the heavy manipulation; Nox would need equivalent or custom motion platforms rated for foundry environments.

Quality control is where the data density pays off. The Air Force's far‑term goal — "make almost anything" with repeatable properties, requires inline non‑destructive evaluation tied back to the process model. Eddy‑current arrays, laser ultrasound, and high‑speed X‑ray computed tomography can generate terabytes per shift. The AI task is not just defect detection but root‑cause attribution: linking a subsurface inclusion at meter 12 of a cast bar to a specific ladle‑temperature excursion three minutes earlier. That traceability is what defense primes — Lockheed Martin, RTX, Northrop Grumman, demand for flight‑critical forgings, and what Nox must deliver to credibly claim "defense‑grade" output.

The labor context sharpens the technical bar. The Air Force explicitly cites a labor shortage in trade skills such as metal fabrication as a driver for robot blacksmiths. Nox's Detroit location puts it in the same hiring pool that Ford, GM, and Stellantis draw from — a pool that has shrunk as experienced tool‑and‑die makers retire. An AI‑run factory that still requires a PhD metallurgist per shift fails the economic test. The system must encode enough process knowledge that a technician with an associate degree can supervise multiple cells, intervening only on exceptions the model flags with calibrated uncertainty.

What would differentiate Nox — proprietary sensor fusion, a foundation model pre‑trained on decades of melt‑shop data, a novel furnace geometry that simplifies control, remains undisclosed. Until Nox publishes a technical white paper, demonstrates a pilot heat, or files patents that reveal the architecture, the technical approach is a hypothesis, not a specification. The seed round buys the furnace time to turn hypothesis into data.

Why the Pentagon Needs New Melt Shops

The Pentagon's current procurement surge makes the supply-chain case for domestic, AI‑optimized metal production concrete. In August 2026 alone the U.S. Army put out a request for more than 130,000 GMLRS rockets to be delivered by 2034, the Department of Defense awarded Raytheon a $22.9 billion contract to accelerate Tomahawk output, and seven‑year framework agreements were signed with Boeing and RTX to expand Standard Missile‑3 component production. Those programs share a common bottleneck: each missile, rocket, and jet engine requires high‑integrity alloys — nickel‑based superalloys for turbine blades, maraging steels for motor casings, titanium and aluminum forgings for airframes, that today flow through a fragile, foreign‑dependent supply chain.

Manufacturing data underscores the strain. Defense News reported in July 2026 that 155‑mm artillery ammunition production remains hamstrung by industrial‑base shortfalls, and in early August the same outlet documented "significant challenges" in U.S. Air Force jet‑engine manufacturing. The root cause traces back decades: the number of prime defense contractors has collapsed from roughly 50 in the 1980s to five today, and those primes outsourced critical sub‑components — often to overseas foundries, while cost‑plus contracting gave little incentive to invest in domestic capacity. A senior Pentagon acquisition official said in January 2026, "We've outsourced those things to other countries for a long time. So now we're redomesticating them, increasing our supply."

The rare‑earth dependency is the starkest example. China controls three-fifths of global mining and more than nine-tenths of refining, per the International Energy Agency, and the U.S. Geological Survey estimates the United States relies on China for roughly seven in ten rare‑earth imports. Those elements — neodymium, dysprosium, samarium, are essential for the permanent magnets in F‑35 actuators, Virginia‑ and Columbia‑class submarine motors, Predator drones, Tomahawk guidance kits, radar arrays, and JDAM kits. In October 2025 Beijing announced export restrictions on rare‑earth materials for foreign militaries, handing Xi Jinping direct leverage over U.S. weapons production. The Trump administration responded with a 100 percent tariff on Chinese goods and an unprecedented equity‑stake, price‑floor, and offtake agreement with MP Materials, the largest U.S. rare‑earth miner, while the Defense Department's Office of Strategic Capital issued a conditional loan to Sunrise Energy Metals to build a full scandium value chain starting from its Australian Syerston Project.

AI‑run mini‑mills like those Nox Metals plans to deploy in Detroit attack a different but adjacent layer: the bulk and specialty alloys that feed the same weapon systems. The Exascale Foundry partnership between Autonomous Resource Corporation and Oak Ridge National Laboratory, announced in April 2026, demonstrates the model. Its initial focus is high‑temperature nickel superalloy turbine components for autonomous air‑vehicle engines produced via metal binder jetting, with ORNL's Peregrine AI software — trained on nearly 2 million additive‑manufacturing layers, providing real‑time adaptive control and quality assurance. That closed‑loop, data‑driven qualification process is exactly what defense programs need to certify new domestic melt sources without the years‑long test campaigns traditional foundries require.

Pentagon procurement policy is shifting to enable it. The department is moving from cost‑plus to fixed‑bid contracts, paying on delivery rather than reimbursing overruns, and using the Defense Innovation Unit and Office of Strategic Capital to fund new entrants directly. The same senior official described the new approach: "We're going downstream from the prime contractors and looking at what are the core components that are common across systems… solid rocket motors go in a lot of different systems from drones to hypersonic missiles to ballistic missiles and they're a scarce resource. We're creating a wartime supply chain that can be flexible and can deliver against the demand that we have."

For Nox Metals, the implication is clear. If its AI‑controlled Detroit factories can produce defense‑grade steel, aluminum, and nickel alloys with traceable, repeatable microstructure — and do so at volumes that matter for missile casings, rocket motor housings, and airframe forgings, they plug a gap the primes can no longer fill alone. The tariff environment reinforces the economics: effective rates on construction‑grade metals hit a 40‑year high of 25‑30 percent in 2025, and Deloitte found nearly half of engineering and construction executives classify their supply chains as "fragile due to geopolitical tensions." Domestic melt capacity that is programmable, auditable, and scalable becomes a strategic asset, not just a cost center.

Incumbents Strike Back

The incumbents are not waiting for AI‑driven mini‑mills to prove themselves. ArcelorMittal, the world's largest steel producer with 34 integrated and mini‑mill facilities across 14 countries and 55.6 million tons of output in 2025, has moved aggressively to embed AI across its own operations. In June 2026 the company announced a strategic collaboration with AWS to converge operational technology and IT on a single cloud platform, extending AI to the edge of production environments at steelmaking sites worldwide. The partnership includes a comprehensive workforce education program designed by AWS to address what Deloitte identifies as the top obstacle to industrial AI adoption: lack of internal expertise. ArcelorMittal's new greenfield plant in Andhra Pradesh, India — slated for first production in December 2028, will be the company's first facility to deploy AI, automation, and robotics at scale from day one, targeting 8.2 million tonnes annually in phase one.

Cleveland‑Cliffs, which acquired ArcelorMittal USA for $1.4 billion in 2020, framed that deal explicitly around the economics of scale. Chairman and CEO Lourenço Goncalves said at the time that "steelmaking is a business where production volume, operational diversification, dilution of fixed costs, and technical expertise matter above all else." The combined entity controls a dominant share of the discerning automotive steel marketplace, a position reinforced by ArcelorMittal's ongoing supply relationship with Gestamp through its 35% stake in Gonvarri. Nucor and Steel Dynamics, the two largest U.S.‑based minimill operators, have seen analyst sentiment strengthen. Both companies have long relied on electric‑arc furnace flexibility; the new variable is whether they can match the data‑driven process control that startups like Nox Metals promise without the baggage of legacy blast‑furnace footprints.

Automotive OEMs are applying pressure from the demand side. General Motors has locked in long‑term supply agreements with domestic flat‑rolled producers that specify exacting metallurgical tolerances, tolerances that AI‑controlled melting and casting could theoretically hit more consistently than human‑supervised heats. Tesla's looming Optimus robot production plans add another layer: the company's "unboxed" manufacturing philosophy requires steel that arrives at the stamping line with near‑zero dimensional variance. ArcelorMittal's XCarb lower‑carbon steel, now supplied to Amazon and AWS data centers under a multi‑year framework agreement, shows how decarbonization mandates are becoming a parallel procurement filter. Europe's revised carbon‑border adjustment mechanism and tariff‑rate quota system, which reduce duty‑free import volumes and raise out‑of‑quota duties, further insulate large European producers from low‑cost imports, but they also raise the bar for any new domestic entrant seeking to qualify as "green" steel under defense or automotive specs.

The pattern is clear: incumbents are absorbing the AI threat by industrializing it themselves, leveraging balance sheets that dwarf Nox's $11.5 million seed. Their counter‑move is not to block the mini‑mill model but to out‑scale it, pairing cloud‑native AI with existing melt shops, rolling lines, and global logistics networks. For automakers, the calculation is simpler: they will buy from whoever delivers defense‑grade and automotive‑grade consistency at competitive cost, whether that steel comes from a 150‑year‑old blast furnace retrofitted with edge inference or a Detroit startup's clean‑sheet cell. The next two years will test whether Nox's architecture can survive first contact with the qualification regimes that protect the incumbents' moats.

The Federal Toolkit

The federal toolkit for domestic metals manufacturing has expanded sharply since 2021, and Nox Metals' Detroit build-out arrives as several programs move from pilot to scale. The Defense Production Act remains the most direct lever: its Title III authorities have provided U.S., Canadian, and Australian companies with grants, loans, loan guarantees, and purchase commitments to reopen mines, build refining and recycling facilities, and produce battery-grade materials, a Council on Foreign Relations report published February 2026 noted. The same report notes that the Department of Defense and Department of Commerce are now "taking equity-like stakes and financial warrants in companies, in addition to providing offtake agreements, price floors, debt, and guarantees" to shore up supply chains, a shift from pure grant-making toward investment structures that resemble venture capital.

The clearest template for a metals startup is the Vulcan Elements agreement announced November 2025. That $1.4 billion vertically integrated rare-earth magnet supply chain, already operating, finances its expansion to 10,000 tonnes annual capacity with a Direct Loan from the Department of Defense's Office of Strategic Capital, federal incentives from the Department of Commerce under the CHIPS and Science Act, and $550 million in private capital. ReElement Technologies, Vulcan's recycling partner, secured an OSC Direct Loan matched by private capital. The OSC, established in 2022 and now operating under the Department of War branding formalized by executive order in September 2025, explicitly targets the "valley of death" between pilot and commercial scale for defense-critical technologies. Nox Metals' AI-run mini-mill concept, domestic, automated, and positioned for defense-grade alloys, fits the OSC's mandate.

Parallel Energy Department programs add another layer. The DOE's Battery Materials Processing and Recycling Grant Program and ARPA-E's RECOVER initiative for wastewater recovery have begun funding projects that extract critical minerals from legacy waste streams, the CFR report found. The Interior Department, under a secretary's order, is streamlining regulations to make mine-waste recovery projects eligible for federal funding and mapping federal mine-waste inventories. For a Detroit operation that could recycle automotive and industrial scrap into prime alloys, these waste-recovery incentives are directly relevant.

The intelligence community runs its own channel. In-Q-Tel launched its Compass Fund in September 2023 to fill the early-to-growth stage (seed through Series B) equity gap in critical-mineral technology. While Nox Metals' $11.5 million seed round came from venture investors, the Compass Fund's existence signals that downstream defense customers track, and sometimes pre-fund, the same supply-chain layer Nox targets.

The CHIPS and Science Act, though semiconductor-focused, created a precedent for commerce-department manufacturing incentives that now extend to adjacent hardware supply chains. A Seeking Alpha analysis from 2026 notes that "the CHIPS Act, Defense Production Act authorities, and targeted federal incentives are anchoring capital flows into semiconductors, grid equipment, defense, and quantum computing, supporting long-term domestic capacity expansion." Nox Metals' investors will likely map the company's roadmap to those capital flows.

At the state level, the research record is thinner. Crain's Detroit framed the Nox raise as part of a "reindustrialization push," but no specific Michigan program, tax credits, grant pools, or workforce partnerships, appears in the provided digests. The CFR report recommends a national critical minerals innovation strategy, a multiyear magnet independence initiative, a waste-recovery program, and a critical-minerals venture fund, all structured around the National Energy Dominance Council established early in Trump's second term. It also urges performance-based procurement standards that would let substitute alloys compete on function rather than composition, a change that would benefit an AI-optimized mini-mill able to tune chemistry per order. None of these are law yet; they indicate where the next authorization bills may direct money.

For Nox Metals, the immediate playbook is visible: pursue an OSC Direct Loan for the first production line, layer CHIPS Act incentives for any semiconductor-adjacent alloy work, and position scrap-recycling throughput to qualify for DOE and Interior waste-recovery grants. The Vulcan Elements deal proves the stack works. Whether Michigan adds its own capital, and whether the next defense authorization codifies the CFR wish list, will determine how fast the second and third factories follow.

Can Detroit Build the Workforce?

Nox Metals enters its growth phase with a team of three, founder Zane Hengsperger and two others, per board data current as of August 2026, and a mandate to scale fast. The $11.5 million seed round, which he told the outlet he expects to reach $14 million, is explicitly earmarked for "product development, conduct market research, and make key hires," Daily Detroit reported. That phrasing mirrors the standard seed-stage playbook, but the operational context is unusual: the company is simultaneously standing up a vertically integrated, AI-controlled metal factory in the same 30,000-square-foot World War II-era building in Southwest Detroit, targeting a summer 2026 move-in.

The hiring profile implied by that build-out is narrow and deep. Nox's stated architecture, AI orchestrating smelting, casting, cutting, and quality control in a single facility, demands a hybrid workforce that does not exist in any single talent pool today. Robotics engineers who understand high-temperature process control, ML specialists who can train models on sparse metallurgical data, and metallurgists willing to write code or at least embed with software teams. Hengsperger calls them "techno-industrialists," a term that signals the blend rather than any established job title. No public job postings from Nox appear on major boards as of this writing, but the company's Y Combinator summer 2025 pedigree and its customer list, Anduril, SpaceX, roughly 100 machine shops and defense contractors, give it recruiting signal in circles where "defense tech" and "hard tech" overlap.

Detroit's labor market supplies half the equation. The region carries generations of tool-and-die, CNC, and automotive production expertise. The broader auto OEM shift toward EVs has displaced skilled tradespeople who know metal. But those workers rarely have robotics or AI fluency. Conversely, the Midwest's robotics talent has been concentrating in logistics automation, not primary metals. Nox sits in the gap. Its ability to recruit will depend on whether it can offer equity packages competitive with better-funded robotics plays while selling a mission, reshoring defense-grade metal, that resonates more than warehouse picking.

The ripple effect is speculative but structurally plausible. The 30,000-square-foot facility running multiple production cells could absorb 50 to 100 technical hires over two to three years if the AI-driven throughput claims hold. That would make Nox a meaningful anchor tenant in Southwest Detroit's industrial corridor, pulling in vendors for sensors, refractory materials, and maintenance robotics. Michigan's reindustrialization incentives could subsidize training partnerships with local universities or community colleges. No such partnerships have been announced.

The risk is timeline. Hengsperger's "this summer" factory launch target leaves months, not years, to hire the core team that commissions the first cell. If the seed round stretches to $14 million as projected, the next tranche likely hinges on demonstrated headcount and factory milestones. For now, the talent impact is a hypothesis backed by capital and a building, not yet a headcount.

The Gaps That Could Kill It

Nox Metals enters a capital-intensive industry with a seed round that barely covers the down payment on a single production line. The $11.5 million, potentially growing to $14 million per Hengsperger's comments to Crain's Detroit Business, must fund the revitalization of the 30,000-square-foot WWII-era factory in Southwest Detroit, the development of AI control systems for smelting and casting, and the hiring of a team that currently stands at three people. Even with a vertically integrated model that skips certain upstream steps, the capital gap between seed funding and operational scale is wide.

Technical scalability presents the first hurdle. The company's current revenue comes from supplying those custom-cut aluminum products to about 100 customers, including defense contractors Anduril and SpaceX. That business relies on CNC machining of sourced aluminum, not the end-to-end AI-run smelting, casting, and quality-control loop Nox Metals envisions. Moving from custom aluminum blocks for CNC machining to autonomous melt-shop operations introduces metallurgical variables that software alone cannot resolve: alloy consistency, inclusion control, thermal gradients during solidification, and real-time defect detection at production speeds. The research describes the goal as "eliminating human bottlenecks," but the bottleneck in primary metals is often physics, not labor.

Regulatory barriers compound the technical challenge. Defense-grade aluminum must meet specifications requiring documented process control, traceability, and third-party certification. The company has not publicly disclosed its quality-management certifications, and the research contains no mention of Defense Logistics Agency qualification or Defense Production Act Title III engagement. Without those, the "defense supply chain" narrative remains aspirational. The company's stated mission to "participate in the reindustrialisation of the US supply chain" will eventually collide with ITAR, EAR, and CMMC compliance requirements that favor incumbents with established security clearances and audit histories.

Competitive response is already visible. Nucor, Steel Dynamics, and ArcelorMittal, all referenced in the broader research as active in capital allocation, operate melt shops with integrated metallurgical labs and decades of process data. These companies are not standing still; they are adding automation, sensor networks, and predictive-maintenance AI to existing assets. The "AI-run factory" differentiator shrinks when incumbents retrofit rather than rebuild.

Talent acquisition in Detroit adds another variable. The research notes Hengsperger plans to "scale the team" using seed proceeds, but the Midwest robotics and AI labor market is tightening. Nox Metals competes for the same techno-industrialists, often against better-funded employers with clearer paths to liquidity.

Open questions the research cannot answer: What is the unit economics of an AI-run melt shop at 10,000 tons per year versus 100,000? How does the company handle scrap recovery and alloy adjustment without a staffed metallurgical lab? What happens when the WWII-era building's structural limits constrain equipment layout? And critically, can a seven-month-old company with three employees and $330,000 in revenue execute a multi-year factory build-out while maintaining the customer relationships that generated that revenue?

The seed round buys optionality, not certainty. The next 18 months will test whether the "techno-industrialist" model can translate software velocity into metals throughput.

Working in frontier tech? Zero G Talent tracks the openings: see every open ASML role, browse frontier tech jobs, openings at Stripe, and the people building the field.