That we are in an AI bubble is a given. The thing about bubbles is that there is no telling when it will pop.
In the late 90's there was an internet/telecommunication bubble. Cedar Rapids was at the heart of this. McleadUSA became one of the largest employers as it became a massive CLEC (Competitive Local Exchange Carrier). By the time of it IPO in 199x it was valued at $20 billion. Many of my friends had stock options were hundreds of thousands of dollars. Although some money was taken off the table, most held on and as the industry imploded my friends jumped ship but none had cashed out the options at the peak.
The same was true for friends at Yellowbook (eventually acquired by Mcleod USA).
Why bring this up now?
This summer SpaceX, OpenAI, and Anthropic are about to IPO. I'm leaving SpaceX out of this because the story that interests me most is the impact on San Francisco - where most Open AI and Anthropic employees live and work there.
The eye-watering valuation OpenAI is seeking is $1 trillion. Anthropic is seeking a $965 billion valuation. (Personally, I think Anthropic will have a higher valuation by the end of 2026 but that's a story for a different podcast.) 1800 OpenAI employees and 1300 Anthropic employees will become multimillionaires over night. The average single family home in the bay area is going for $2.15 million. A starter house of less than 1000 sq. feet is going for over a million.
Now 3100 people will suddenly have the money to buy. Currently San Jose and San Francisco (both bay area/silicon valley) are the two most cities for housing in America. Econ 101 says they are about to take that to a new level.
Yay for them! But... Does a rising tide lift all boats? How does a janitor afford to live there? A policeman? A barrista? Or my fraternity brother Danny who is a swimming/diving coach in Palo Alto?
So econ issue #1 for me: What happens when there is suddenly a whole lot of wealthy(wealthier) people in a city?
********
In the late 90's I was working weather living mostly in western Iowa/Nebraska. When I returned home I'd join a poker group that mostly guys working for McLeod. I definitely heard how about their stock options and was mildly jealous. I'm living in a place I didn't like making $50k. They are making about the same but worth hundreds of thousands in stock options. Yay for them but I had to ask myself wtf I was doing. One of my former weather employees settled down, got married and went to work at Rockwell Collins (Now Collins Aviation). Their stock also jumped after he started so he was making what he made working wx but now had options valued over $100k. Having once owned/operated the largest bar in Cedar Rapids with his brothers, Neil chose to devote his free time to fixing up old pinball machines/video games and developing an amazing home brew beer system. Again, a great story for a different podcast.
********
Business/Econ issue #2:
For a rapidly growing company there are three concerns:
- Customer Acquisition Costs (CAC): How much does it cost for a company to gain new customers?
- How much does it cost to service them once you've acquired them?
- How high is the churn/turnover (people choosing to leave you)?
Groupon IPO'd at its peak with a valuation of $17.8 Billion. People loved Groupon! Half priced meals? Yes, please! The problem? Customers were addicted to the discounts and didn't become loyal to the restaurants. Many restaurants pulled their offers and as of today it is valued at $614 million. That's 3.5% of the IPO valuation. Even though Groupon acquired customers quickly and cheaply they have not been able to keep them.
When ChatGTP 3 (OpenAI) launched in November 2022 I was amazed. I started paying $20/month for their premium service. By the time 4 launched I saw the free service was good enough for me and stopped paying. The problem for the OpenAI/Anthropic/Google Gemini/Microsoft Co=Pilot, etc. is that none of them are remotely close to having revenue cover the costs of doing business.
Google with all of its phenomenal cash assets announced last week it is raising $80 billion to pay for faster AI expansion.
What... what if these AI's will never be worth the cost of making them? AI will continue to exist but change dramatically. While AI can put several categories of employees out of work: Technicians reading mammograms, paralegals, stenographers, etc. the product they produce may become a commodity. At that point nobody/no company will care whether they use Anthropic or Gemini to get the results they need. (There is an argument that some AI's will become specialized in a particular field... I love https://www.harvey.ai/ Harvey is named after the lead in the legal drama Suits.)
Back to the guys playing poker. My first clue they were screwed is when they were describing McLeod's Customer Acquisition Costs. Given the tight margins in the industry it would take years to pay back the cost of each new customer acquired. And if they jump ship? Eeek.
Fortunately for Cedar Rapids the bubble didn't pop so much as the air seeped out. Even as McLeod USA telecom and it's acquisition of Yellowbook went belly up, the avionics jobs in CR grew in the post 9-11 military expansion.
Nobody will shed a tear for Silicon Valley when the AI bubble pops. That it will happen is a given. The when? Economist Paul Krugman, Robert Shiller and many others started pointing out problems in the banking/real estate industry in 2005. (I remember because I was teaching Political Economics and I made my students read about it.) The bubble grew exponentially from there before finally popping in 2008. I don't think we will have to wait 3 years to see the AI bubble pop, but I'd argue its impact will be ginormous. If it weren't for capex spending on AI, we'd already be in a recession.
That's also why of all the candidates on my side running for President I short the chances of Governor Newsom... but again, a story for a different podcast.
My goal in the next few weeks is to zoom with some friends who worked at McLeodUSA. I'm curious if they agree with the parallels I am seeing.
********************
The search engine/AI I used to get data was Google's Gemini. I didn't bother providing links because pretty much everything I wrote above is likely to change in the coming days/weeks.
