February 2026: The "SaaSpocalypse" — iShares Software ETF plunges 27%, RSI hits a 15-year low

I'm Howard Uncle, CQF charterholder.

Over the past 30 days, the US software sector experienced a bloodbath — what traders are calling the "SaaSpocalypse."

The numbers are brutal:

• Adobe: $699 → $263 (−62%), wiping out its entire pandemic-era gains
• ServiceNow: Peak $239 → $98 (−59%)
• Salesforce: −48% drawdown, 10-year average P/E crushed from 46x to 14x
• Figma: Post-IPO $143 → ~$20 (−86%)

The software sector's Relative Strength Index (RSI) hit 15 — a 15-year low. Hedge fund net exposure to software dropped to a historic low of 3%.

What triggered this crash? Not a recession. Not earnings misses. Something more fundamental: Is AI enhancing software — or replacing it?

On January 30, 2026, Anthropic released 11 open-source AI plugins that automate legal drafting, sales pipeline management, and financial analysis. The market suddenly asked: If AI agents can do the work directly, do we still need traditional SaaS charging per seat?

But is this panic justified — or another buying opportunity in disguise?

Today I'll use Figma, ServiceNow, and Salesforce as case studies to teach you how to evaluate software company moats in the AI era. I'll break down:

• Why Generative UI (GenUI) is a bigger paradigm shift than AI coding tools
• How three companies are responding to the AI threat — with very different strategies
• A four-dimension framework to assess whether a software company's moat still holds
• Key dates and risks over the next six months

The method is reproducible — you can apply it yourself after reading. But remember: I teach methods, not buy/sell recommendations. Your money, your responsibility.

Ⅰ. The Real Paradigm Shift: Not AI Coding — It's Generative UI

1.1 What the Market Gets Wrong

Let's start with what spooked the market.

In January 2026, a wave of AI tools launched in rapid succession:

• Anthropic Claude Cowork: AI work assistant
• Lovable: $100M ARR in just eight months (Vibe Coding tool)
• Bolt/Cursor: Generate apps directly from text prompts

The market's knee-jerk reaction: "These tools will eliminate designers and developers. Software companies are losing their customers!"

Hold on. That's only the surface.

The real paradigm shift isn't "AI helps you write code." It's "AI generates the interface directly — no pre-design needed."

This is Generative UI (GenUI): interfaces generated in real time based on user intent, dynamically adapting instead of being pre-designed by humans and hard-coded in advance.

1.2 Why GenUI Is the Bigger Threat

Traditional software workflow:

Designer creates UI in Figma → Developer codes it → User gets a fixed interface

AI coding tools (Cursor/Lovable):

Designer creates UI in Figma → AI generates the code → User gets a fixed interface

Impact: Developer efficiency improves, but designers are still needed. Figma retains value.

GenUI disruption:

AI generates interface based on user intent → No pre-design needed → Interface adapts dynamically

Impact: The design step could be skipped entirely. Figma's value takes a massive hit.

Data point: According to Gartner's 2025 AI Predictions Report, by 2030, 90% of interfaces will be AI-customized rather than pre-designed.

This isn't fear-mongering. Picture this:

• You open an e-commerce site. AI generates a layout tailored to your shopping history and current mood — in real time
• You use enterprise software. AI rearranges modules and workflows based on how you actually work
• Every user sees a different interface because AI is generating it on the fly

If interfaces don't need pre-design, what's Figma for?

To be frank, this threat runs deeper than AI coding tools because it undermines the entire "design → develop" division of labor.

1.3 But Wait — Will GenUI Actually Work This Way?

Here's the critical engineering reality: the vision is compelling, but deployment is brutal.

The "real-time UI generation" scenario sounds incredible. But in production environments, it's nearly impossible. Three barriers:

Barrier 1: Cost economics don't work

• Every page load requires an AI model call to generate the interface
• For an e-commerce site with 1 million daily active users at $0.01 per API call: $10,000/day, $3.65M/year
• Traditional pre-designed interfaces? Near-zero marginal cost

Barrier 2: Latency is unacceptable

• Generating a complex interface takes 2–5 seconds
• Users waiting five seconds to see a page? They leave
• Traditional static pages load in under 500ms

Barrier 3: Output is unpredictable

• AI-generated UIs may have bugs, accessibility (a11y) issues, or XSS security vulnerabilities
• Each generation produces different results — impossible to systematically test
• Can enterprise applications tolerate this uncertainty? No.

So is GenUI a fake concept? No — but the real deployment path looks different from what most people expect.

1.4 How GenUI Actually Deploys: Offline Generation, Online Matching

The viable approach isn't "real-time generation." It's "heavy offline generation + lightweight online matching."

Let me explain with a concrete scenario:

Traditional approach (A/B testing):

1. Designers create two versions (A and B)
2. Split traffic 50/50 between them
3. After a week, pick the winner

GenUI-optimized approach (AI-powered micro-segmentation):

1. Offline phase (heavy AI usage):

   • AI analyzes user data and segments users into 50–100 micro-groups (not two — dozens)
   • For each segment, AI rapidly generates 5–10 candidate interfaces
   • Simulated A/B tests select the optimal version
   • Winning designs are compiled into code and deployed to CDN

2. Online phase (lightweight matching):

   • When a user visits, a lightweight classifier (<10ms) identifies their segment
   • The pre-generated interface for that segment loads instantly
   • Zero additional latency. Zero real-time AI cost

What's the essence of this approach?

The classic "precomputation vs. real-time computation" tradeoff — AI era edition.

Analogies you'll recognize:

• Search engines don't build indexes when you search — they pre-index everything
• Recommendation systems don't train models in real time — they train offline and infer online
• CDNs don't fetch from origin servers every time — they pre-cache at global nodes

GenUI follows the same logic:

• The AI "heavy lifting" happens offline (generation, testing, optimization)
• Online, it's just lightweight matching (user → pre-built variant)
• Result: AI-powered personalization with the stability and low cost of traditional approaches

What does this mean for Figma?

Under this model, the designer's role doesn't vanish — it transforms:

• No longer: Manually designing 50 interface versions
• Instead: Defining design systems, component libraries, and constraint rules
• Then: Letting AI generate, test, and iterate within those constraints

Analogy: Architect vs. construction crew

• Architect (designer): Defines style, structure, and safety standards
• Construction crew (AI): Builds quickly within those constraints, creating variants and optimizing

So Figma's value isn't "drawing interfaces" (AI will handle that). It's "building design systems and collaboration platforms" (AI can't replace that).

If Figma successfully pivots to become the "design system platform for the AI era," its moat holds. If it stays a "drawing tool," it's in real danger.

1.5 Jensen Huang's Latest Take: Software Won't Be Replaced by AI

On February 4, 2026, Nvidia CEO Jensen Huang said this at the Cisco AI Summit:

"The idea that software will be replaced by AI is the most illogical thought in the world."

His core argument:

• AI is a user of tools, not a replacement for tools
• The act of coding has been devalued (AI writes code), but the capability and tooling for building software have appreciated
• The act of drawing interfaces has been devalued (AI generates them), but design decisions and design systems have appreciated

In plain English:

• Wrong: "AI will replace developers and designers. Software companies will die."
• Right: "AI will eliminate low-value repetitive work, but high-value creation and decision-making become more important."

Historical parallel: When Excel appeared, accountants' manual calculation skills were devalued. But their financial analysis capabilities became more valuable. The accounting profession didn't die — it evolved.

The same pattern applies:

• Figma's "interface drawing" function will be devalued
• But "design systems" and "collaborative decision-making" will appreciate
• The question is whether Figma can successfully pivot

That's why Huang called the replacement idea "illogical."

Software companies that fail to evolve will be eliminated — but the cause will be "failing to integrate AI," not "being replaced by AI."

1.6 So Why Is the Market Still Panicking?

If GenUI's real path is "offline generation + online matching," and Jensen Huang himself says software won't be replaced — why did software stocks crash this hard?

Market panic operates on three levels:

Level 1 (Surface): Misunderstanding the technology

• Most investors hear "AI generates interfaces" and assume "real-time generation"
• They conclude Figma will be completely replaced
• Panic selling follows

Level 2 (Deeper): Doubts about transition speed

• Even knowing GenUI means "offline generation," investors worry:
  • Can Figma pivot to a "design system platform" fast enough?
  • Can ServiceNow and Salesforce integrate AI into their product cores?
• History is littered with companies that had the right strategy but moved too slowly

Level 3 (Deepest): Business model repricing

• Even if companies successfully transition, the business model may change:
  • From "per-seat pricing" to "usage-based pricing"
  • ARPU (Average Revenue Per User) could drop 20–40%
  • Valuations need repricing

This explains why even rational analysis saying "software won't be replaced" can't stop the bleeding.

Investors aren't pricing in "whether it gets replaced." They're pricing:

1. What's the probability of successful transition?
2. How much profit gets lost during the transition?
3. What's the post-transition business model worth?

Right now, nobody has clear answers to these three questions.

So the market chose "shoot first, ask questions later" — and will revisit once data arrives.

What does this mean for us?

For long-term investors, current valuations (historic lows) may represent opportunity. But for short-term traders, this decline may not be over — the "answer verification period" needs at least 6–12 months.

Let me be direct: nobody can guarantee software stocks have bottomed.

Ⅱ. Three Case Studies, Three Different Fates