The Capability Gap Reshoring Won’t Close 

$165 billion. That's what TSMC has committed to US semiconductor manufacturing, making it the largest foreign direct investment in American history. The headlines call it a turning point, but I'm not convinced. 

During the 2020-21 semiconductor shortage, I was in Microsoft Azure's Cloud Sourcing and Supply Chain org, leading data center infrastructure planning. What I saw then shapes how I think about reshoring now. 

When the shortage hit, traditional sourcing practices collapsed across the industry. Long-term forecasts meant little while suppliers were triaging allocations, and companies without pre-existing relationships were scrambling. We weren't scrambling as Microsoft had invested in those relationships before the crisis hit. 

I flew out to meet a supplier's executive team in person and, sitting across the table from their operations leadership, saw what years of relationship-building had earned us: a real conversation about shared priorities, not a negotiation over scraps. That trip helped secure the supply we needed, and that trust was converted into allocations when it mattered most. 

Capability vs Capacity

Here’s the distinction I keep coming back to: Capacity is not capability. Capacity is what you own. Capability is what you can execute. 

In semiconductors, capacity means fabs, equipment, and allocation commitments on paper. Capability means process engineers, chemical suppliers, yield optimization routines, and the trust relationships that determine who gets supply when everyone is short. 

The CHIPS Act put $39 billion toward domestic manufacturing. As of mid-2025, $30.9 billion has been awarded, and only $6 billion actually disbursed. Fabs are under construction (capacity), but Taiwan didn't become the center of global semiconductor manufacturing because of capital. TSMC controls nearly 70% of the global foundry market. They built that dominance over four decades: optimizing yields, establishing a supplier ecosystem, and refining maintenance routines over millions of production hours. You cannot replicate that ecosystem by building a building.   

The Capability Gap Is Already Showing

TSMC's first Arizona fab was scheduled for 2024, but workforce shortages pushed it to 2025. The company flew in over 1,000 technicians from Taiwan to train local workers, and the fab is now operational, but the second fab has slipped to 2027-2028. 

Additionally, Intel's 18A process node has persistent yield problems. In October 2025, its CFO said yields are "adequate to address supply, but not where we need them to be in order to drive the appropriate level of margins." Industry-standard yields aren't expected until 2027.  

These are symptoms of industrial atrophy. The US semiconductor workforce is down 43% from its 2000 peak, and the Semiconductor Industry Association projects a 67,000-worker shortfall by 2030. Capital is flowing, but capability isn't keeping up. 

What Actually Secured Supply

The companies that survived the 2020-21 shortage weren't the ones with the biggest purchase orders, but the ones that treated supply as a risk modeling problem. At Microsoft, we extended forecast windows and stress-tested demand scenarios that we could bring to suppliers as a foundation for shared commitment. Securing supply meant giving something first: extended horizons, sharing risk, and being transparent about our demand signal. 

In our latest report on Continuous Planning, we call this signal-to-action compression: the ability to sense a shift, interpret the implications, and adapt at market tempo. Organizations with mature planning capabilities recover faster and absorb shocks better, while companies running transactional supplier relationships end up having to explain to customers why they are going to miss their commitments. 

What I'd Do Differently Now

If I were still in the seat: 

• Stress-test your supplier relationships now, not during the next crisis. The companies that secured supply in 2021 built trust before they needed it.
• Model capability decay, not just capacity growth. Most supply plans assume symmetric recovery, but real industrial systems don't work that way. Capability erodes faster than it rebuilds.
• Extend your commitment horizons. The quarterly cadence wasn't built for structural supply constraints.

The semiconductor industry is entering another period of volatility. AI demand is straining advanced node capacity, geopolitical tension is accelerating diversification, and billions in reshoring capital are about to collide with reality. Fabs without capability are just expensive buildings. Capital follows capability, not the other way around.