Anthropic and General Catalyst: Markets Misprice Resilience

Amazon's stock fell roughly 95% from its peak in 2001 while many of its dotcom peers went into liquidation — yet the company emerged stronger because it had already spent 1999 expanding fulfillment capacity from 300,000 square feet to more than five million. That excess capacity gave Amazon optionality. It opened its network to other retailers and independent sellers, turning idle warehouse space into a platform business.

That's the historical anchor of a new argument from Anthropic and General Catalyst: resilience is a growth asset that markets still price as an afterthought. For most of the last century, that was defensible. Resilience meant surviving disruption, a defensive property rewarded only in hindsight. But as disruption becomes more frequent and varied, the companies that absorb shocks fastest are the ones that grow while competitors retrench.

AI accelerates this divide. It threatens incumbent business models while giving companies a powerful instrument for absorbing shocks. Organizations running on AI-enabled processes are cheaper to operate and have more room to act when conditions change.

One of the most expensive mistakes in markets, the authors argue, is confusing "still growing" with "structurally sound." Recurrence is not the same thing as durability. Markets price the risks of disruption more accurately than the businesses built to withstand it. The qualities that make a company resilient are visible in advance — the challenge is measuring durability before disruption makes its value obvious.

Where does the mispricing show up?

The authors identify four places where the gap between resilience's cost and its value plays out:

• Demand resilience. AI is exposing how conditional some recurring SaaS revenue really is. Software embedded in core processes and compounding on proprietary data has held up better than seat-based software for work AI can now do. Investors, though, have historically screened by sector label rather than depth of entrenchment.
• Supply-chain resilience. Markets punish disruption yet give little credit to companies that spend to avoid it.
• Capital-stack resilience. Private-equity sponsors bought hundreds of software companies in 2021 on heavy debt, assuming recurring revenue would service it. Not all recurring revenue is equal.
• System resilience. A company can appear resilient while consuming resilience from the system around it.

What does AI do to SaaS revenue?

Healthcare, defense and energy sit at the opposite end of the spectrum from vulnerable SaaS. Regulatory barriers and institutional trust make them harder to displace digitally, while demand for essential services can rise when conditions deteriorate. A hospital becomes more vital to its community during a crisis — the same conditions under which a SaaS business may see customers pull back and growth falter.

Defense and energy can behave the same way. Aramco spent years building flexibility into its system, including the East-West Pipeline and multiple pipeline and storage options. Those investments paid off within days of the Iran war and disruption in the Strait of Hormuz, helping Aramco remain one of the world's most reliable oil suppliers.

What does resilience cost before it pays?

TSMC is the clearest live case. It pays a significant near-term cost to diversify manufacturing beyond Taiwan. Its overseas factories are more expensive to build and operate, weighing on margins as they come online. Markets see that cost immediately and the protection only later. TSMC's Arizona factories have run at yields matching Taiwan since late 2024. The option paid off; it simply took longer than a quarterly model tolerates.

The private-equity software trade tells a parallel story. The duration of buyout debt is fixed. The duration of the revenue underneath it is variable, and AI is shortening it unevenly. Software that genuinely sits inside a customer's operations has proved more durable. The difference was visible from the start, but investors focused on recurring revenue rather than what made it recur.

Who consumes resilience versus who builds it?

The system-level point may be the sharpest in the piece. Electrical grids are built with spare capacity to absorb shocks. AI data centers increasingly draw on that capacity, transferring risk from individual companies to the broader system. The North American Electric Reliability Corporation has sharply raised its forecast for peak electricity demand and in May issued its highest-severity alert after a sudden, massive swing in computing demand.

The authors' prescription: the most resilient companies will invest in the systems they depend on rather than treating spare capacity as theirs to consume.

The principle extends beyond infrastructure. JPMorgan has built resilience through a balance sheet designed to absorb financial shocks and a business diversified well beyond investment banking. That structure has helped it withstand periods of stress that left less diversified peers exposed.

What does this change for investors?

Building resilience carries a cost. Redundant capacity, stronger balance sheets and operational flexibility all require investment before they are needed. But those investments buy optionality and the ability to keep moving when conditions change and competitors cannot.

That is the market's blind spot. Companies can spend years building resilience that weighs on near-term returns while its value remains invisible until conditions deteriorate. The result is a market that prices the cost of resilience before it prices its value.

Amazon in 2001 was that kind of company. There are many more of them now — and the authors' closing implication is that practitioners who learn to measure durability in advance, rather than in hindsight, hold the edge.