The $500 Billion Illusion

Why Industrial Policy Can't Buy Knowledge

There’s a bet being placed right now using federal incentives to catalyze private capital. Since 2020, companies have announced more than $540B in private U.S. semiconductor projects on the assumption that capacity can be turned into capability on political timelines. That the US can "friend-shore" critical minerals processing away from China. That capital, deployed at scale, can substitute for decades of accumulated knowledge.

Here's what the data actually shows: it's not working. And the reason why reveals something profound about how capabilities are really built.

The Scoreboard (January 2026)

Let's start with the numbers, because they tell a story most people aren't hearing yet.

TSMC share (two lenses; not directly comparable)

• Top-10 foundries by revenue (TrendForce): 2Q25 = 70.2% (record)
• Pure-play foundry share (IDC forecast as reported): 2025 = 66% (up from 2024)

The point isn’t the exact percent, it’s that the dominance hasn’t meaningfully budged yet.

Wait. Read that again. TSMC's dominance is increasing while the US deploys hundreds of billions to build domestic capacity. Despite dozens of announced U.S. projects, including ~17 new fabs, and tens of billions in federal subsidies and loans. Despite $32 billion in direct subsidies and $29 billion in loans.

Critical minerals (China's position as of 2023-2024):

• Lithium refining: 60%
• Battery cell capacity: 77%
• Cathode production: 78%
• Battery components (illustrative 2022 shares): China produced roughly ~70% of cathodes, ~85% of anodes, ~82% of electrolytes, and ~74% of separators.

This is after the US Defense Production Act, the EU Critical Raw Materials Act, and billions allocated to "friend-shoring" to Australia, Canada, and allied nations since 2022.

The pattern is consistent: massive capital deployment, minimal capability transfer.

What Everyone's Getting Wrong

There's a simple assumption embedded in these policies. It's so obvious that nobody questions it:

If you build the physical capacity (fabs, refineries, equipment), competitive capabilities will follow within 5-10 years.

I'm calling this the Capacity = Capability Fallacy. And this is costing hundreds of billions.

The assumption makes intuitive sense to Western policymakers and executives. After all, in financial capitalism, capital is the universal solvent. Need distribution? Buy a network. Need technology? Acquire a startup. Need manufacturing? Build a facility.

But something strange happens with advanced manufacturing capabilities. They don't transfer. Let me show you why.

The Knowledge You Can't Buy

There's a concept from economic complexity theory called "personbytes". Think of it as knowledge measured in person-years: the accumulated experience sitting in human brains and organizational routines.

Here's the key insight: personbytes cluster geographically because much of the knowledge is tacit. You can't write it down. You can't codify it in a manual. A process engineer learns by watching another engineer, iterating through problems, accumulating intuition about when a tool is about to fail or why a yield dropped 2%.

The TSMC Example

TSMC didn't become dominant by spending more than competitors. Intel has outspent them. Samsung has comparable capital. What TSMC has is something else entirely:

• 40 years of accumulated process knowledge (since 1987)
• Deep customer co-development relationships with Nvidia, Apple, Qualcomm, AMD
• Taiwanese cluster effects: nearby suppliers, educated workforce, government support
• Causal ambiguity: even TSMC can't fully explain why their yields are better

This is what strategy scholars call a VRIN resource (Valuable, Rare, Inimitable, Non-substitutable). The "inimitable" part is crucial. Intel can't copy it even when they know what TSMC is doing, because the knowledge is embedded in 40 years of trial and error, organizational learning, and supplier ecosystems.

In 2Q25, overall foundry revenue (top-10) rose 14.6% QoQ, while TSMC’s revenue rose 18.5% QoQ and its share reached 70.2%. The gap is widening, not closing, despite competitors having access to the same equipment suppliers (ASML, Applied Materials, Tokyo Electron).

The Historical Timeline Nobody Wants to Hear

Let's look at countries that actually built competitive manufacturing capabilities:

South Korea (1962-2000):

(World Bank, current US$)

• 1960: $159
• 1980: $1,746
• 2000: $12,710
• ~ three decades to reach global top-5 in economic complexity (ECI ranking #5 by 1995)

The path: Textiles → Wigs → Light manufacturing → Steel → Shipbuilding → Electronics → Semiconductors → Automotive

Each stage built capabilities for the next. You couldn't skip steps. Korea tried (remember their failed aircraft industry in the 1970s) and learned this lesson the hard way.

Shenzhen (1980-2024):

• Started with simple assembly (toys, textiles)
• Progressed to consumer electronics
• Now dominates complex products requiring 1000+ components
• Timeline: 44 years to become the world's electronics manufacturing hub

Notice the pattern? Human learning rates constrain capability building. You can't compress 25 years into 5 by spending 5x more capital.

Why China Dominates Critical Minerals (And Will Keep Dominating)

The rare earth and lithium stories are almost identical.

1990s-2000s: The West outsourced mining and refining to China for cost and environmental reasons. "Let them deal with the toxic waste while we focus on high-value design and assembly."

2010: China restricted rare earth exports. Prices spiked 10x. Western governments panicked.

2010-2015: Multiple attempts to restart Western refining. Molycorp in California. Lynas in Australia. Most failed economically when prices normalized because they lacked:

• Scale economies (China had built volume)
• Process expertise (decades of iteration on separation chemistry)
• Integrated supply chains (China controlled mining and refining and magnet production)

2022-2024: History repeats with lithium. Massive government subsidies announced. But here's the timeline analysts are actually predicting:

"Structural bottleneck persists with resolution timeline extending to 2030-2035+"

That's the baseline best case. And it assumes everything goes right: permitting, environmental approvals, technical success, market prices that justify continued operation.

The Theory of Constraints Meets VRIN Resources

There's an elegant synthesis here from two different strategy frameworks.

Theory of Constraints (Goldratt): Every system has a bottleneck. The way to improve system performance is to identify and exploit that bottleneck.

Resource-Based View (Barney): Sustainable competitive advantage requires VRIN resources. When a resource is inimitable, competitors can't replicate it regardless of capital.

The synthesis: When the bottleneck is a VRIN-protected capability, you can't "elevate the constraint" by adding capacity. The properties that make it valuable (tacit knowledge, path dependence, causal ambiguity) are the same properties that make it unreplicable.

What This Means for Policy

Current industrial policy assumes:

Capital Investment → Physical Capacity → Competitive Capability

Reality is:

Capital Investment → Physical Capacity
Competitive Capability → Requires 15-25 years of learning, adjacent moves, ecosystem development

The gap between these two models is costing hundreds of billions.

What Works Instead: Following the Adjacent Possible

Countries and regions that successfully built capabilities followed what complexity economists call "product space navigation". Each capability you have determines which new capabilities are feasible to develop next.

Think of it as a board game where:

• You can only move to adjacent squares
• Each square you occupy makes new adjacent squares accessible
• Trying to jump across the board fails

Korea's successful moves:

• Textiles → Electronics assembly (adjacent: basic manufacturing discipline)
• Electronics assembly → Consumer electronics (adjacent: understanding supply chains)
• Consumer electronics → Semiconductors (adjacent: precision manufacturing)

Failed leaps that don't work:

• Oil exports → Aerospace (too distant: Saudi Arabia tried this)
• Agriculture → Semiconductors (too distant: skip too many capability steps)

This is why Shenzhen will likely retain >40% of global electronics manufacturing through 2035 despite:

• Rising Chinese wages
• Geopolitical pressure for diversification
• Massive subsidies to alternative hubs (Vietnam, India, Mexico)

The prediction from complexity theory: Alternative hubs will handle simple assembly but fail to replicate Shenzhen's full-stack capabilities for complex products. Timeline to reach comparable capability density: 15+ years minimum.

Why? Because you can't airlift the personbytes. The knowledge networks are embedded in the place.

The Uncomfortable Implications

If I'm right, here's what follows:

For Semiconductor Policy

• US will add capacity but remain dependent on Taiwan for leading-edge chips through 2035+
• Intel's catch-up is structurally blocked by VRIN-protected TSMC capabilities, regardless of subsidies
• Geographic risk (Taiwan) remains the dominant issue, not fixable with domestic investment alone

For Critical Minerals

• China maintains 60-90% refining dominance through 2030-2035 minimum
• "Friend-shoring" enables partial diversification but not independence
• Substitution innovation (alternative battery chemistries, reduced rare earth content) is more effective than capacity building

For Supply Chain Strategy

• Stop investing in elevation (adding capacity) of VRIN-protected constraints
• Start investing in optimization (working around constraints, demand shaping, process improvement)
• Accept that some dependencies can't be eliminated, only managed

The One Scenario Where I'm Wrong

This analysis assumes current approaches continue. There is a path to success, but it requires accepting the real timeline:

Realistic 2040-2045 goals:

• Develop capabilities through adjacent moves (don't try to leapfrog)
• Invest in education and workforce development (the actual bottleneck)
• Build regional ecosystems, not isolated facilities
• Accept 20-25 year timelines for competitive capabilities

What doesn't work:

• 2030 political promises
• Capital-only strategies
• Assuming capacity = capability

The hardest part? Telling the public that the $540 billion investment will take 20+ years to produce results, not 5-7. But that's what the historical evidence shows.

Final Thought: The Humility of Human Learning

There's something almost beautiful about this constraint. In an age where we can simulate protein folding and train models on trillions of parameters, we still can't compress human learning.

A process engineer needs 5-10 years to develop real expertise. An organization needs 15-20 years to build the routines and relationships that constitute competitive manufacturing. A cluster needs 20-30 years to accumulate the supplier density and knowledge networks that make it sticky.

Capital can build buildings. It can't buy time.

The policymakers who understand this will make better bets. The ones who don't will keep learning expensive lessons.

The data in this analysis draws from economic complexity theory (Hidalgo), resource-based view (Barney), and empirical case studies of TSMC, critical minerals constraints, South Korea's development trajectory, and Shenzhen's manufacturing ecosystem. Market data current as of January 2026.