I'll be at Reindustrialize 3.0 in Detroit on June 16 and 17, and there's one thing I want to put on the table before I get there.
The American reindustrialization conversation is louder and more serious than it's been in my career. Capital is moving. Policy is shifting. The workforce question is finally getting the attention it deserves. The strategic case for a real industrial base, the kind that can build at scale and survive a contested decade, is broadly accepted in rooms where it used to get rolled eyes.
That's all real progress. I'm not here to talk anyone out of it.
But I've spent the last few years on the implementation side of industrial AI, working with operators on the actual floors where these systems either work or don't. And the part of the conversation I keep waiting for, the part that decides whether any of the rest of it pays off, is mostly missing. Not from people doing the work. From the public conversation. From the panels and the press releases and the policy memos.
The missing part is the operating layer.
Where the work actually breaks
Most of what's being written about reindustrialization assumes that once the capital lands and the policy stabilizes and the workforce trains up, the operating layer underneath will keep up. It won't. The systems that run American industry today, the ERPs and procurement workflows and production decisions made at six in the morning when the data is wrong and somebody has to act anyway, weren't built for what reindustrialization is asking of them. And almost nobody is staffing or funding the work to fix that.
I want to be careful here, because the easy version of this argument is to wave at “AI implementation challenges” and move on. That's not what I mean.
What I mean is specific. Most enterprise AI initiatives in industrial environments don't reach production. The numbers vary depending on how you define production, but credible analyses land somewhere between two-thirds and four-fifths of all projects stalling out before they create operating value. That's not a model problem. The technical state of the art has been more than capable for industrial use cases for a while. It's not a budget problem either. The companies running these initiatives are spending real money.
The work breaks at the handoff. Between the proof of concept that demoed beautifully in a steering committee meeting and the actual operator who has to use the system on a shift floor with imperfect data, partial training, and a job that doesn't pause for a software rollout. That gap is the last mile, and it's where the entire industrial AI thesis succeeds or fails.
This matters for reindustrialization because the same gap exists at every layer of the new industrial economy. The factory goes up. The supply chain reshores. The procurement contract gets signed. And if the operating layer underneath those investments can't carry the weight, the company ends up with a beautiful new facility running on the same disconnected systems that were breaking its operations a decade ago.
That's the version of reindustrialization that fails quietly. From where I sit, it's the version most companies are quietly building.
What I'm actually seeing
Across the industrial clients we work with, I've been watching a progression. Companies move through it in a specific order, and skipping a step is the most common cause of stall. I'll walk you through the four layers, with one example each from work I've been close to.
Layer one is visibility. Most industrial companies I've worked with can't answer the simplest question about their own operations, which is what they paid for what and when. They run on ten or more systems that don't talk to each other. The gaps get stitched together in spreadsheets that go stale the moment they're saved. Decisions about what to order and from whom get made on partial information, and everyone in the building knows it's partial.
I sat in a conference room with one industrial client's procurement team while they pulled up their ten different ERPs side by side. A senior buyer realized in real time that her own organization had paid three different prices for the same part across three vendors in the same quarter. She wasn't surprised by the price difference. She was surprised that no one in the company had been able to see it before. That moment is why this work matters.
We built an ontology across all ten systems for that client. Now they can see what they're paying for, what they're paying, and when it'll arrive. Those are the two real levers in any supply chain, cost transparency and delivery timing. Without seeing both, no investment, no model, and no procurement reform changes the outcome. The Palantir Foundry word for what we built is ontology. I'm using it on purpose. It's not a database. It's a structured representation of how the business actually works, and it's the foundation everything else gets built on.
Layer two is cost reduction. Once a company can see, it can act. The first thing it usually acts on is cost, because operating costs in industrial environments are full of unnecessary spend that's accumulated over decades and that nobody owned the responsibility to question.
We worked with an advanced manufacturing client to cut the daily cost of getting operational data off their equipment from roughly $10,000 a day to $700. About a 93 percent reduction on a line item the company had treated as fixed for years. We did it while also increasing the uptime of the equipment itself, which matters because it shifts value from cost-out, which is defensive, to throughput, which is offensive. American industry needs both, and most companies aren't yet doing either at scale.
Layer three is margin engineering. Cost savings inside operations matter, but the strategic question is whether they translate into price competitiveness in the market. That's the actual question reindustrialization has to answer. Whether American manufacturing can compete on price isn't only a question of labor cost or trade policy. It's also a question of whether companies can surface the savings already sitting inside their own operations and pass them through to the products consumers actually buy.
We worked with one global manufacturer on exactly this. Found efficiencies inside their production and procurement that translate to roughly $15,000 of margin on a family SUV. The implication of that number is the difference between an American-built vehicle being affordable for a working family and being out of reach. Reindustrialization that doesn't produce affordable American goods isn't reindustrialization in any meaningful sense. It's a more expensive version of what we already have.
Layer four is autonomous execution. This is the layer where I think the next decade gets decided.
We're working with a company that's rebuilding how American homes get built. The goal is homes that are meaningfully cooler in literal temperature terms, and meaningfully cheaper to construct. The system we built reads the bill of materials for any given build, identifies what's needed, pulls every supplier the company has ever ordered from along with the prices paid, proactively introduces additional qualified suppliers who can fulfill the same parts, and submits the bid request through an automated workflow.
The human role on that team has shifted. They're not executing procurement anymore. They're setting strategy, watching for the cases that genuinely need judgment, and exception-managing the rest. This is what agentic AI actually looks like when it works in industrial settings. It's not a chatbot. It's not a copilot. It's the operating layer making decisions and acting on them, with humans steering rather than driving.
Almost nobody in the public reindustrialization conversation is talking about this layer. They should be. It's where compounding advantage gets built.
The progression matters
Visibility, then cost reduction, then margin engineering, then autonomous execution.
Most American industrial companies are stuck somewhere between the first and second layers. A few are reaching for the third. The companies that will compound advantage through the 2030s are the ones building toward the fourth, and they're going to do it whether the public conversation catches up or not.
Here's what bothers me. The reindustrialization debate isn't going to become serious until it starts measuring the country's progress on this gradient, because this gradient is what decides whether the capital, policy, workforce, and geopolitical investments produce a competitive industrial economy or just a more expensive one. Right now we're measuring the wrong things. We're counting groundbreakings and tariff schedules and reshoring announcements, which are easy to count. We're not yet counting whether the companies executing on those announcements can actually run the operating layer underneath them, which is harder to count and also the thing that matters.
What I'm watching for from here
There are three audiences I keep thinking about when I look at the gap between where most industrial companies are and where the leaders are heading.
The first is policy. The infrastructure of reindustrialization is necessary and not sufficient. Tariffs do real work. So do procurement reform and the capital incentives in CHIPS and IRA. None of them changes the operating layer underneath, and from where I sit there aren't yet serious proposals that take that gap seriously. The next round of industrial policy needs to fund the systems inside the factory, not just the factory itself, and the people writing that policy will need to listen carefully to operators about what that actually requires.
The second is the operators running industrial businesses today. The AI initiative the CEO has been briefed on, the procurement modernization the CFO is reviewing, and the supply chain visibility the COO is fighting for aren't three projects. They're one project. The companies treating them as separate are the ones whose investments are failing to compound. The leaders I'm seeing make real progress have stopped running these initiatives in parallel and started running them as a single program with one operating thesis underneath. That shift is harder than it sounds and it usually requires someone senior with cross-functional authority to force the integration.
The third is capital. The durable moats in industrial AI aren't in the models. The models are commoditizing and that trend is accelerating. The moats are in the implementation capability. Companies and partners that can ship governed, operator-trusted systems into production at scale, repeatedly, across messy enterprise environments, are the ones that will compound through the cycle. From where I sit, that's where the multi-decade alpha lives, and it's underpriced relative to what's getting funded right now.
Why we're joining NAIA
Foxtrot just joined the New American Industrial Alliance.
I want to be honest about why we did this, because I'm hearing a lot of “we joined a trade group” energy in this category lately and most of it lands as theater. We joined because reindustrialization is a coalition project, and the coalition needs a voice for the operating layer.
The companies inside NAIA are doing the work that matters. They're building the factories, the aircraft, the energy infrastructure, the defense systems, the materials base of the next American industrial era. Foxtrot's job inside that coalition is the implementation layer. The part that turns the AI investments those companies are making into outcomes in production rather than dashboards in slide decks.
I'll be at Reindustrialize 3.0 in Detroit on June 16 and 17. I want to meet the founders and operators and investors and policymakers who are doing the work, especially the ones wrestling with any of the four layers I just walked through. There's more than enough work to share, and the country needs us to do it well.
If you're going to be there and any of this resonates, find me. I'd rather compare notes than give a speech.
Pete Vigneaux is a Commercial Portfolio Lead at Foxtrot Services, a Foundry-native Palantir partner founded by Palantir alumni. Foxtrot helps industrial companies close the last mile gap of enterprise AI by delivering production-grade, governed workflows on Palantir Foundry. Foxtrot recently joined the New American Industrial Alliance, and Pete will be at Reindustrialize 3.0 in Detroit on June 16-17, 2026.