Understanding AI Bubble and Revenue Projections

I keep seeing the same pattern destroy SaaS companies: AI makes their customers insanely productive. Those customers need 80% fewer seats. Revenue falls off a cliff. The pricing model is literally eating itself. I've executed pricing transformations across 4 SaaS turnarounds. What worked 18 months ago now destroys value - every SaaS company is racing to embed AI, and it's breaking their revenue models. The automation paradox: AI makes customers wildly productive, so they need fewer seats. You just automated away your own revenue model. 85% of SaaS companies have abandoned pure per-seat pricing. The holdouts are learning why the hard way. Here's what actually works now: Track different data. Old way: Seats, tiers, revenue per account. New way: Token consumption, API calls, automated workflows. Found one enterprise using AI to replace 10 seats while consuming 100x the resources. Seat pricing misses this completely. Price outcomes, not access. Old way: ROI = human hours saved. New way: Automated resolutions, workflows completed. Saw $500/month AI running entire departments. Customer saves $2M annually. Your pricing is broken. Build hybrid models. Old way: Per-seat with usage tiers. New way: Base subscription + AI consumption. Example: $X base platform fee + $Y per 1,000 AI resolutions. Revenue jumps 3x. Churn drops. Value finally makes sense. Model the seat apocalypse. Old way: 20% churn assumptions. New way: Accounts dropping from 50 to 10 seats but 10x-ing AI usage. Price it right = 2x revenue. Miss it = -60%. Prove value first. Old way: Show features, hope they get it. New way: "Our AI resolves 1,000 tickets = 40 human hours." Now $2/resolution pricing clicks. Without proof, you're just taxing AI. CS becomes AI coaches. Script: "You're paying for 50 seats but AI handles 30 of those workflows. Let's optimize." Fewer seats, higher revenue. Trust wins. Real-time transparency. Token usage dashboards. Cost predictions. 80% alerts. Show exactly what AI costs vs human alternative. Black box pricing = dead company. Most SaaS companies still add 50% "AI premiums" to seat licenses. Meanwhile, Salesforce charges per conversation. Zendesk per ticket resolved. The leaders already moved. But the window's closing. Companies with consumption-based AI models report 38% higher growth. Foundation models commoditize by 2030. We have maybe 24 months. After that, it's a race to the bottom. The fundamentals from my 4 turnarounds still apply - but the game has changed. We used to price software that helped humans work. Now we're pricing software that replaces them. Get this transition wrong and you'll watch competitors eat your market share. Get it right and you own the next decade.

A few highlights from our friend Anastasios A. at Bienville's recent "Bubble Warning" below: In our history of following markets, there have been few and far between five-year profit growth estimates that have sustainably topped the 30% annum pace. S&P Semis (an A.I. proxy) long-term EPS growth forecasts are running at 70% - which compounded results in 14.2x; over four times the tech sector’s 1999/2000 and current 27% annum rate that compounds to 3.4x. Tech valuations in general and semiconductors in particular have gone parabolic. Semis are trading at a whopping 5 sigma above trend, while Tech has vaulted to all-time high valuations (trumping the dotcom era peak), hovering around 2.5 sigma above trend. Google search trends for AI are at all-time highs. But worry not! The new "Price-to-AI" metric fashioned by Morgan Stanley’s Sherry Paul makes the NASDAQ look dirt cheap! The big risk (perhaps postponed after yesterday's report) is that NVDA will eventually miss expectations as revenue gains do not appear to be volume-based but dependent upon price. A comparison of COGS to revenues proves the point. Q/Q growth has already crested. While the AI bubble may have more air to intake, the mania is in the late innings. Similar to QCOM peaking in Dec-99, there are high odds that NVDA may also top out before the major indexes. Piling into these cult stocks is fraught with danger at the current juncture. There is nothing wrong with moving to the sidelines, as the aftermath of the tech bubble was brutal!

"Investors have poured $330 billion into about 26,000 A.I. and machine-learning start-ups over the past three years, according to PitchBook, which tracks the industry. That’s two-thirds more than the amount they spent funding 20,350 A.I. companies from 2018 through 2020." "A.I. start-ups have been challenged by that gap between spending and sales. Anthropic, which has raised more than $7 billion with backing from Amazon and Google, is spending about $2 billion a year but pulling in only about $150 million to $200 million in revenue, said two people familiar with the company’s financials, who requested anonymity because the figures are private." Turns out "losing money on every sale, but making it up on the volume" is not a good business strategy. My main lesson from the first tech bubble is that when venture money is plugging the hole between what a service costs and what the end user will pay, bad things will happen. That does not mean that there will not be winners and even big winners in the AI startup-land. It means that diligence is even more important. In particular, understand what value is being offered to end users and whether it supports the revenue model of the venture. #venturecapital #ai