AI Changes the Discount Rate, Not Just the Growth Rate

Why traditional valuation frameworks break when the cost of intelligence goes to zero.

Chamath Palihapitiya said something on the All-In podcast recently that most investors heard but few internalized: if AI is real, and it works the way the market is pricing it, then the natural consequence is that most software companies become less valuable, not more.

That sounds backwards. The market is paying 40x, 60x, 80x earnings for anything that touches AI. But Chamath's point was precise: if AI commoditizes the ability to build software, then the moat that justified a 15x revenue multiple for SaaS companies evaporates. The hard thing becomes easy. And when the hard thing becomes easy, margins compress.

He's right. But I think the argument goes further than most people realize. The problem is not just about earnings estimates. It's about the discount rate itself.

The PE Compression Thesis

Here's the standard bull case for AI-adjacent tech companies: AI increases productivity, which increases margins, which increases earnings, which justifies today's multiples. Simple.

But run the math the other direction. If AI makes it 10x cheaper to build a competing product, what happens to the incumbent's pricing power? If an AI agent can replicate 80% of what a $50K/year SaaS tool does, what happens to that company's net retention rate?

The answer is not "earnings go up." The answer is "the entire earnings trajectory that justified a 60x PE multiple was wrong."

Consider the current state of affairs. Microsoft trades at roughly 33x forward earnings. Salesforce at 27x. ServiceNow at 58x. Palantir at 150x+. These multiples assume durable competitive advantages that compound over decades.

AI doesn't destroy these companies. It does something worse for their shareholders: it makes their advantages temporary. When any reasonably funded startup can use Claude or GPT to build a credible competitor in months instead of years, the premium you pay for "we have a 10-year head start" shrinks dramatically.

The Part Nobody Models: The Discount Rate

Every DCF model has two inputs that matter: the cash flow projections and the discount rate. Analysts spend 95% of their time on the cash flows and almost none on the discount rate. They plug in WACC, which is itself derived from a beta that measures historical volatility against the S&P 500, and they move on.

But the discount rate is supposed to capture risk. And AI is changing the risk profile of every tech company in ways that historical beta cannot measure.

Think about it this way. If you're building a DCF for ServiceNow today, you're probably using a discount rate somewhere between 9% and 12%. That rate implicitly assumes ServiceNow's competitive position is roughly as risky as it was over the last 5 years. Same moat, same switching costs, same enterprise sales cycle protecting the business.

What if that's wrong? What if AI-powered workflow automation tools reduce switching costs because they can migrate data and replicate processes trivially? What if the enterprise sales cycle compresses because a VP can prototype a replacement using an AI agent before the renewal call?

If the risk is higher, the discount rate should be higher. And a higher discount rate does more damage to a DCF than lower growth estimates. A company growing cash flows at 20% per year but discounted at 15% instead of 10% loses roughly 35% of its intrinsic value. That's not a rounding error. That's the difference between "buy" and "overvalued."

Where This Gets Concrete

Let's run some actual numbers. Take a hypothetical SaaS company doing $500M in free cash flow, growing at 20% annually, with a current WACC of 10%.

Standard DCF, 10-year model with 3% terminal growth: intrinsic value is roughly $22 billion.

Now bump the discount rate to 13% because AI has made the competitive landscape meaningfully riskier. Same cash flows, same growth rate. The intrinsic value drops to roughly $14 billion. A 36% haircut from a 3-point increase in the discount rate.

The market is not pricing this. Every major tech stock's implied discount rate (which you can back-calculate using a reverse DCF model) assumes business-as-usual competitive dynamics. The discount rates baked into NVDA, MSFT, and PLTR valuations are historically low, implying historically low competitive risk.

That assumption deserves scrutiny.

The Counter-Argument (And Why It's Partially Right)

The obvious pushback: incumbents will use AI too. Microsoft is embedding Copilot everywhere. Salesforce has Einstein. ServiceNow has its own AI layer. They're not standing still while startups eat their lunch.

This is true. The largest companies will benefit from AI more than most because they have the data, the distribution, and the customer relationships to deploy it. The moat doesn't disappear overnight.

But the moat gets narrower. And when moats get narrower, multiples compress. You don't go from 60x PE to 10x. You go from 60x to 35x. And for a stock priced at 60x earnings, a re-rating to 35x is a 42% decline even if earnings grow exactly as the street expects.

This is the scenario that kills returns: the company executes perfectly, hits every estimate, and the stock still goes down because the multiple it deserves is lower than the multiple the market assigned it.

What Smart Investors Should Do

I'm not saying sell all tech. I'm saying your valuation framework needs to account for a world where competitive advantages decay faster than they used to.

Practically, that means three things:

First, stress-test your discount rate. If you're using 10% WACC for a software company, ask yourself: "Would I still use 10% if I believed AI makes competitive moats 30% less durable?" Run the DCF at 10%, 12%, and 14% and see how much the valuation changes. If a 2-point increase in discount rate kills your thesis, your thesis depends on assumptions about competitive durability that AI is actively undermining.

Second, focus on companies with moats that AI can't replicate. Network effects (Visa, Mastercard). Regulatory capture (insurance, defense). Physical infrastructure (pipelines, cell towers, data centers). These are the businesses where a 10% discount rate still makes sense because an AI agent cannot build a competing railroad.

Third, look at PE ratios in historical context. The current median PE for the technology sector is roughly 28x. The 20-year average is closer to 22x. If AI compression brings multiples back to historical norms, that's a 21% decline embedded in the sector that has nothing to do with earnings.

The Bottom Line

The market is treating AI as pure upside for tech companies. Higher productivity, higher margins, higher earnings. But the same force that increases productivity also increases competition. And increased competition means the premium you pay for being a tech company should shrink.

The question every investor needs to answer is simple: does AI make this company's advantages more durable or less durable? If the answer is "less," then the stock is overvalued at current multiples regardless of what earnings do next quarter.

Most investors are asking "will this company use AI to grow faster?" That's the wrong question. The right question is: "does AI make it easier for someone else to compete with this company?" For most of the software sector, the honest answer is yes. And that honest answer is worth 20-40% of the current stock price.

The DCF doesn't lie. But it only tells the truth if you're honest about the discount rate.