铝时刻：为何AI将软件变为钢梁

The Washington Monument was completed in 1884, and at its very peak sits a 100-ounce pyramid of solid aluminum. At the time, that pyramid was a flex of unimaginable wealth. Aluminum was more expensive than silver, more difficult to refine than gold. It was a “Precious Metal,” reserved for jewelry and the monuments of kings.

Two years later, Charles Martin Hall and Paul Héroult independently discovered the electrolysis process. They crashed the price of aluminum by 90% overnight.

On February 5, 2026, Wall Street is waking up to the realization that the B2B SaaS industry (Salesforce, Workday, ServiceNow) is the Washington Monument’s capstone. For two decades, the industry treated code as a precious metal. Companies paid recurring royalties for it and built trillion-dollar valuations on its scarcity.

Then Claude Code arrived. Just like the Hall-Héroult process, it did not just make the metal cheaper. It turned a scarce asset into a structural commodity.

This week’s “Great Rotation” is where capital is violently fleeing the BVP Nasdaq Emerging Cloud Index for shares of Duke Energy and Caterpillar. This is not a correction. It is a repricing of reality. Software is no longer Gold (it is Girders). And builders do not pay a monthly subscription for girders.

The News: The Rotation to “Real” Assets

Investors looking at portfolios this week see violent internal churning despite a flat S&P 500. The BVP Nasdaq Emerging Cloud Index is down 14% week-over-week, hitting lows unseen since the 2022 rate hike panic. Meanwhile, the Industrial Select Sector SPDR (XLI) and Utilities Select Sector SPDR (XLU) are hitting all-time highs.

The mainstream financial press, including Bloomberg and CNBC, calls this “Sector Rotation.” They attribute it to “rate jitters” or “valuation concerns.”

They are wrong. This is not a macro story about interest rates. It is a technology story about Net Dollar Retention (NDR).

Investors are realizing that the core metric of the SaaS business model (NDR) is mathematically broken in a world where a customer can replicate a product for $50 of compute.

The Theory: The Aluminum Moment

To understand why this crash is structural, historical context from 1886 is necessary.

Before Hall-Héroult, aluminum was rare not because it did not exist (it is the most common metal in the Earth’s crust), but because the bond between aluminum and oxygen in bauxite ore was incredibly strong. Breaking that bond required immense energy and complex chemical reduction.

The Hall-Héroult process used electricity to brute-force that bond, turning electricity into aluminum.

Code is the New Bauxite

In 2024, “Code” was rare. It was locked inside the minds of Senior Engineers (the Bauxite). Extracting it required salaries of $300,000 and months of time. Because extraction was expensive, the resulting software was precious. Companies rented it and cherished it.

In 2026, Agentic AI (specifically the new class of coding agents led by Claude Code and DeepSeek-V3) is the Hall-Héroult process. It turns Electricity (Compute) directly into Code.

The bond is broken.

$$\text{Cost of Software} = \frac{\text{Complexity}}{\text{Human Hours}} \rightarrow \text{Cost of Software} = \text{KwH} \times \text{Model Efficiency}$$

When the denominator shifts from “Human Hours” (scarce, expensive) to “KwH” (commoditized, scalable), the price of the output must collapse.

The “Ice Trade” Fallacy: Why SaaS Dies

The most chilling parallel is not aluminum. It is ice.

In the 19th century, the “Ice Trade” was a massive industry. Ships hauled blocks of ice from frozen New England lakes to India and the Caribbean. It was a logistics empire. They thought they were in the business of selling ice.

They were not. They were in the business of selling Cold.

When refrigeration was invented, it allowed “Cold” to be generated at the Edge (right in the customer’s kitchen). The user no longer needed the logistics network (the ships). They just needed the machine and the electricity.

SaaS is the Ice Ship. Companies like Salesforce built massive, complex logistics networks (Sales teams, Customer Success, Cloud Infrastructure) to deliver “Features” (Cold) to users. Customers paid a premium because they could not generate features themselves.

AI Agents are the Refrigerator. Now, a CTO can look at a $50,000/year contract for a niche internal tool and realize they can spin up a dedicated agent to build this exact workflow in Python. They can host it on a $5 droplet and own it forever.

The “Logistics Network” of SaaS (the sales team, the marketing, the flashy UI) is now dead weight. The customer can generate the “Cold” (the utility) directly at the point of consumption.

The Numbers: 90% Deflation

The math of “Disposable Software” reveals the deflationary impact.

In 2024, building a custom CRM for a 50-person cleaning business required:

• Agency Build: $40,000 upfront + $5,000 maintenance.
• SaaS (HubSpot): $12,000/year forever.

In February 2026, verified benchmarks show Claude Code can scaffold, test, and deploy a functional, CRUD-based CRM in approximately 45 minutes of autonomous loop time.

• Token Cost: $12.50.
• Hosting: $20/month.
• Total Year 1 Cost: $252.

The Math of Collapse:

$$\text{Value Destruction} = \$12,000 (\text{HubSpot Revenue}) - \$252 (\text{Compute Cost}) = \$11,748$$

That $11,748 gap is the deflationary pressure hitting the tech sector. It is money that was Tech Revenue, now becoming “Consumer Surplus” or simply vanishing.

This drives Salesforce (CRM) down while Duke Energy (DUK) rallies. The value moves from the “Code” layer to the “Energy” layer. The $252 goes to the utility company and the data center. The $11,748 stays in the pocket of the cleaning business owner.

The Gray Area: The “Girders” Bull Case

Is this the end of software? No. It is the beginning of Ubiquitous Software.

When aluminum crashed to $0.20/lb, usage did not stop. It shifted from jewelry to airplanes. Builders constructed skyscrapers. Families wrapped leftovers in it.

The volume of software code in existence is about to explode by 1,000x. Software will exist for everything.

• A bespoke app just for a family reunion? Done.
• A logistical routing script for a single delivery truck’s Tuesday route? Done.
• A disposable financial model for one specific meeting? Done.

But users do not pay subscription fees for foil wrap.

The winners of the “Aluminum Moment” will not be the people trying to sell code at jewelry prices. It will be:

1. The Bauxite Miners: The energy grid and data center REITs (Equinix, Dominion Energy).
2. The Smelters: The Foundation Models providing the “process” (Google, Anthropic (though their margins will also compress)).
3. The Airplane Builders: The companies that use this practically free software to build entire new industries in the physical world (Tesla, Anduril).

What Comes Next: The De-Financialization of Ops

The market is entering a period of “CapEx-heavy, OpEx-light” IT.

For the last 15 years, the trend was OpEx. “Don’t buy servers, rent AWS. Don’t build tools, rent SaaS.”

The trend is reversing. It is now cheaper to Own than to Rent, because the cost of creation (CapEx) has fallen below the cost of renting (OpEx).

Companies will stop buying specialized software seats. They will buy Compute Blocks. They will treat software generation like electricity: a utility metered by the second, used to power the real work of the business.

The SaaS bubble is not popping because the tech failed. It is popping because the tech succeeded too well. It turned gold into aluminum. And while aluminum is useful, keeping it in a vault is foolish.