Tariffs & Network Redesign:
the difference between reacting and restructuring

Over 2 million U.S. manufacturing jobs have been announced via reshoring and foreign direct investment since 2010, according to the Reshoring Initiative's 2024 Annual Report. The reshoring narrative is real, but the execution is lagging badly. Construction takes years. Supplier ecosystems take longer. And without deep domestic networks, new plants risk becoming costly assembly hubs still dependent on the imported components tariffs were meant to replace.

The structural challenges are not hidden: they're just inconvenient to say out loud. U.S. labour costs ($25–30/hr) still far exceed Asian competitors ($6–7/hr) without advanced automation. Nearly 500,000 manufacturing jobs remain unfilled because modern factories require skills the current workforce doesn't yet have. Companies announce reshoring under policy pressure, then delay or cancel when subsidies expire. Who actually does the landed cost math before committing $200M to a new facility? In most cases I've observed, that analysis comes after the announcement, not before.

Genuine restructuring looks different from the press release version. The starting point is a full landed cost model: tariffs, transit time, safety stock implications, service level impact, not just unit price. China+1 diversification is a deliberate intermediate step, not a permanent answer. Vietnam's tariff exposure has shifted again: temporarily more competitive under the current flat-rate structure, but with USTR running accelerated Section 301 investigations on multiple Southeast Asian markets. The teams that built diversification plans in 2023 assuming Vietnam was a clean answer got caught once. Building the same assumption now carries the same risk. Rebuilding supplier ecosystems requires qualification timelines and dual-sourcing logic that take 12–18 months minimum. And the part most companies skip entirely: connecting the network decision to S&OP. A new sourcing region changes lead times. Changed lead times change safety stock. Changed safety stock changes working capital. That cascade has to live somewhere beyond the procurement team's spreadsheet. What I don't have a clean answer to is how fast a mid-size company can realistically build this optionality: the playbook exists, but the organisational bandwidth to execute it while also running operations is a different question.

The bottleneck, in most cases, isn't the analysis. It's the level at which the decision gets made. A supply chain team can identify the exposure, model the scenarios, and present the options. But committing to a new supplier ecosystem, absorbing higher unit costs during a transition, or accepting elevated safety stock while a new lane stabilises: those are not procurement decisions. They require a mandate that comes from above. The teams navigating this environment well have one thing in common: they got this conversation onto the right agenda, early enough to still have options.

Reshoring is not a supply chain strategy. It's a trigger for one. The work is in the network design, not the announcement.

The landed cost calculation is where most reshoring decisions break down. A component that costs $5 in China and $8 in Mexico looks 60% more expensive, until you add the tariff (now running at ~30–35% effective rate on most Chinese goods, layering Section 301 duties on top of baseline MFN rates), the transit time differential (35 days vs 5 days), the safety stock required for ocean variability, and the markdown risk from delayed availability. When you model the full picture, the gap narrows dramatically, and in some cases reverses entirely.

For a company with €30M in sourced goods, a 34% effective tariff on a major China lane represents over €10M in additional annual cost. That number is now large enough to justify nearshoring decisions that were previously marginal. Here is the mechanism most models miss: a shorter, more predictable lane requires less safety stock. Less safety stock means less working capital tied up in buffer inventory. That freed capital can offset a significant portion of the higher unit cost from a nearer-shore supplier, sometimes completely. The companies that see this are the ones running the full model. The ones that don't are comparing unit prices and wondering why the CFO keeps pushing back.

And there is a deeper problem behind the CFO conversation. Every team I have seen fight hardest for a low unit cost was optimising the wrong number. An IMU is a figure on a spreadsheet. What actually wins is net margin at the end of the chain. The two are not always friends. The logic runs like this: you source far away chasing the unit price. Quality is not quite right. The shipment arrives late. You miss the selling window. You start discounting. You end up at the outlet. You sell the rest the following year. You lost while thinking you were winning. The tariff was never the problem. The cascade was.

When the tariff waltz started last year, I'll admit I smiled. Not because it wasn't serious, but because I expected the numbers to change every week, which they did. But the question I kept coming back to wasn't "what does this cost us." It was: does it still make sense to produce where we're producing? Not every category has a real answer to that question. Some products can only be made in certain regions, at the quality and price point the market expects. Others have more flexibility than most sourcing teams are willing to admit. The problem is that most companies never separated those two categories. They applied a blanket response to a question that needed product-by-product thinking. That's where the real conversation should have started.

The deeper issue is that the new world is not giving us a new normal. It is giving us instability as the constant. Reacting is useful in the short term, but if reacting is your strategy, you lose over time. The companies that built optionality before the pressure arrived are not smarter. They just accepted instability earlier than everyone else did, and started building before they had to. What this tells me isn't that the strategy was wrong: it's that anyone who used the first round of tariffs to build supplier optionality across multiple regions is now navigating this from a position of strength. Anyone who waited for clarity before moving is still waiting.

The tariff structure today has a sunset clause in ten weeks. Section 122 expires in late July 2026, and USTR is running accelerated Section 301 investigations on 16 countries with new rates expected before that date. Vietnam looks temporarily competitive again. The word "temporarily" matters. Anyone building sourcing scenarios right now needs to model what happens after July, not just today. What I still have not worked out is whether the teams not doing this work are missing the analysis, or whether they have it and cannot get anyone above them to act on it. I have seen both. From the outside, they look identical. But they are completely different problems, and only one of them gets solved with a better model.

The more complex the chain, the more this holds. The easy objection is: too complex to move quickly. That is true. But the correct conclusion is not to wait. It is that the conversation has to start earlier, the scenarios have to be stress-tested before the pressure arrives, and the what-if questions have to be on somebody's desk before the disruption makes them urgent. The companies that started those conversations two years ago do not have perfect answers. They have options. That is the difference.

Most network redesign decisions are optimized for today's tariff snapshot. That is the wrong target. Tariff policy in the current environment is not a stable input; it is a variable. A network move that looks compelling at 34% may look unnecessary at 15% and insufficient at 54%. The teams that redesign for current rates and stop there are making a capital allocation decision based on one scenario out of four or five plausible ones. When the rate changes, they redesign again. That cycle is expensive and slow.

Three things AI makes tractable before you commit. First, run an HTS (Harmonized Tariff Schedule) classification audit before any scenario modeling: feed your import records into an AI tool and ask it to cross-reference your declared HTS codes against the product descriptions and flag discrepancies. Misclassification is more common than most teams expect, and optimizing against a wrong tariff baseline produces confident arithmetic on a false foundation. Fix the classification first. Second, build a scenario robustness matrix: take your top 10 sourcing lanes by spend, define four tariff scenarios (current rates, full escalation to negotiated ceiling, partial rollback to pre-2025 baseline, sector-specific carve-out), and ask the AI to calculate total landed cost for each lane under each scenario. The output is not a recommendation; it is a map of which lanes are exposed under all scenarios, which are only exposed under escalation, and which are genuinely resilient across the range. That is the analysis that makes a network redesign durable rather than reactive. Third, model the transition cost explicitly: for each move the matrix flags as robust, ask the AI to estimate the working capital requirement, including qualification lead time, dual-sourcing safety stock buffer, and service level degradation window. This number is almost always absent from the initial business case and almost always changes the sequencing of what you do first.

The specific failure mode: scenario modeling is only as good as the scenario definitions. If your four tariff scenarios are not grounded in actual policy mechanisms (negotiating leverage, WTO constraints, sector lobbying dynamics), you are running math on arbitrary assumptions. Use public trade policy analysis to calibrate the scenarios before you run the numbers. Garbage scenarios produce confident garbage outputs.

AI with your numbers is leverage. AI without them is theater.