Monday MAMLMs: Where Will the Money to Pay for All the “AI” DataCenters Come From? Modern Advanced Machine-Learning Models (MAMLMs) will have value as very big-data, very high-demension, very...
While your analysis of the unsustainable aspects of Uber’s business model ten years ago was thoughtful and carefully documented, today’s comments on Uber were not. You never explained how a company that lost $33 billion in its first 14 years years—because of all the issues you had originally pointed out—magically became profitable. You didn’t point out any errors in your original analysis that recent events had exposed or point out anything Uber had done to become more efficient.
Uber’s turnaround can be reduced to four major points: (1) Uber abandoned every aspect of its original business model that had driven its meteoric growth and popularity (prices well below everyone’s actual costs and expansion into areas that had been poorly served because they were extremely unprofitable) (2) the post-pandemic realization of massive anti-competitive market power because the scorched-earth retaliation against any potential market entrant, regulator or critical journalist that Kalanick had established meant they could raise prices and cut service without fear of market discipline (3) their successful post-pandemic war against any democratically elected government that attempted to provide rudimentary labor law protections for drivers (e.g. spending hundreds of millions to nullify California’s Proposition 22) (4) their post-pandemic implementation of surveillance pricing (first order price discrimination) for riders and (more importantly) drivers, things that would never have been possible in competitive markets. These issue have been understood for some time; for a fuller explanation see https://www.nakedcapitalism.com/2025/02/hubert-horan-can-uber-ever-deliver-part-thirty-five-what-drove-ubers-recent-8-billion-pl-improvement.html
You said “I was wrong” about failing to see that Uber could become profitable. Unless you want to defend a company finally figuring out how to maximally exploit artificial market power, the problem here is your failure to analyze or understand Uber’s post-2023 profitability. That failure totally undercuts the faux-humility of your “I was wrong about Uber, maybe I was wrong about LLMs.
My recent American Affairs article documents the huge parallels between Uber and OpenAI—two companies whose established business model had absolutely no possibility becoming sustainably profitable, but had to create bubble-like enthusiasm that could subvert cognition of huge, well documented business/financial problems. *Understanding the LLM Bubble*
I was in NYC recently and I used both Uber and taxis. The price was about the same. Basically taxis were forced to reduce fares due to Uber, then Uber was forced to increase their prices to make a profit. So neither option is a bargain, and both are equally expensive.
Uber's claim--endlessly repeated during its first decade--was that its growth had nothing to do with regulatory evasion or the $13 billion from investors that funded its predatory pricing--it was due to its superior technology that made it much more efficient than traditional taxis. Uber never substantiated any of these claims and they were all lies and could be refuted with objective industry data.
Uber's below cost pricing pre-pandemic (which drove the $33 billion in losses) were egregiously anti-competitive, and jacking up prices post-pandemic was egregiously anti-competitive. It never had any legitimate efficiency advantages. Anyone defending Uber is defending the right of Silicon Valley investors to subvert competitive markets.
I don't understand Brad's desperate reach for a crystal ball. Investors are legitimately looking for a new utopia; regulators are legitimately fearing a new dystopia. The rest of us are in it for the ride. "Are we there yet?" isn't a very useful question, since we don't know where we are going. LLMs are not like a next-day weather forecast.
That being said, I find that Brad's very intelligent speculations help me better understand the present, if not the future.
One thing to consider is that a lot of corporate processes that look like perfect fits for AI have already been highly automated. Unlike AI, they are auditable.
Look at flight dispatching at a major airline. This involves filing the flight plans, checking the weight and balance, calculating and ensuring the fuel load and making sure the plane can take off and land. By the 1980s when I did a code inventory for a major airline, the code was extremely sophisticated and embedded all sorts of sometimes hard earned knowledge. In theory, the airline could run with the dispatchers doing little more than hitting OK now and then, at least if no problems came up. It is hard to imagine how AI could improve this process.
It's a similar case for a shipping or inventory management company. Unless it's a momma-poppa operation, odds are they have sophisticated algorithms already at work, and those algorithms are auditable and can account for the odd size of SKU7812556 and the problems making a U turn on Forest Road 6257.
Ditto for the code validating credit card purchases. Ditto for the code running internal combustion engines. Ditto for the code running production machinery. There's a lot of highly tuned code is already paying for itself. If the code is going to run a million times, odds are there isn't much room left for AI. That means AI is going to have to get good at one offs and small production runs. Given the high cost, those are going to have to be high value applications.
With the birth rates falling below replacement world wide, it seems those jobs will become more automated.
My age cohort is the fastest growing.
Robots to lift us patients from beds and wash our bodies clean and put on a clean diaper. Robots to clean the room. A robot to keep us Alzheimer's patients engaged. Robots to assist us in getting dressed and in and out of the bathroom. So many tasks to do.
From what I read the care for the aged is a very tiring task that wears out caregivers. If you want to sell billions of devices, mobility assistance robots is the tech to develop. I am on the ragged edge of needing that assistance device now.
I think "copilots that help programmers refactor and remember syntax" is a pretty big understatement with respect to the current capabilities of, say, Claude Code.
What if OpenAI had not given ChatGPT to everyone at first? Because I think that led to the insane defensive investment, which required too many investors to promise the sun, the moon, and the stars. Maybe the hardware investment can be used for other models which are less cool but have a moat and profit. Which also probably requires getting off the NVDA software.
While your analysis of the unsustainable aspects of Uber’s business model ten years ago was thoughtful and carefully documented, today’s comments on Uber were not. You never explained how a company that lost $33 billion in its first 14 years years—because of all the issues you had originally pointed out—magically became profitable. You didn’t point out any errors in your original analysis that recent events had exposed or point out anything Uber had done to become more efficient.
Uber’s turnaround can be reduced to four major points: (1) Uber abandoned every aspect of its original business model that had driven its meteoric growth and popularity (prices well below everyone’s actual costs and expansion into areas that had been poorly served because they were extremely unprofitable) (2) the post-pandemic realization of massive anti-competitive market power because the scorched-earth retaliation against any potential market entrant, regulator or critical journalist that Kalanick had established meant they could raise prices and cut service without fear of market discipline (3) their successful post-pandemic war against any democratically elected government that attempted to provide rudimentary labor law protections for drivers (e.g. spending hundreds of millions to nullify California’s Proposition 22) (4) their post-pandemic implementation of surveillance pricing (first order price discrimination) for riders and (more importantly) drivers, things that would never have been possible in competitive markets. These issue have been understood for some time; for a fuller explanation see https://www.nakedcapitalism.com/2025/02/hubert-horan-can-uber-ever-deliver-part-thirty-five-what-drove-ubers-recent-8-billion-pl-improvement.html
You said “I was wrong” about failing to see that Uber could become profitable. Unless you want to defend a company finally figuring out how to maximally exploit artificial market power, the problem here is your failure to analyze or understand Uber’s post-2023 profitability. That failure totally undercuts the faux-humility of your “I was wrong about Uber, maybe I was wrong about LLMs.
My recent American Affairs article documents the huge parallels between Uber and OpenAI—two companies whose established business model had absolutely no possibility becoming sustainably profitable, but had to create bubble-like enthusiasm that could subvert cognition of huge, well documented business/financial problems. *Understanding the LLM Bubble*
(https://americanaffairsjournal.org/2026/02/understanding-the-llm-bubble/)
I was in NYC recently and I used both Uber and taxis. The price was about the same. Basically taxis were forced to reduce fares due to Uber, then Uber was forced to increase their prices to make a profit. So neither option is a bargain, and both are equally expensive.
Uber's claim--endlessly repeated during its first decade--was that its growth had nothing to do with regulatory evasion or the $13 billion from investors that funded its predatory pricing--it was due to its superior technology that made it much more efficient than traditional taxis. Uber never substantiated any of these claims and they were all lies and could be refuted with objective industry data.
Uber's below cost pricing pre-pandemic (which drove the $33 billion in losses) were egregiously anti-competitive, and jacking up prices post-pandemic was egregiously anti-competitive. It never had any legitimate efficiency advantages. Anyone defending Uber is defending the right of Silicon Valley investors to subvert competitive markets.
I don't understand Brad's desperate reach for a crystal ball. Investors are legitimately looking for a new utopia; regulators are legitimately fearing a new dystopia. The rest of us are in it for the ride. "Are we there yet?" isn't a very useful question, since we don't know where we are going. LLMs are not like a next-day weather forecast.
That being said, I find that Brad's very intelligent speculations help me better understand the present, if not the future.
One thing to consider is that a lot of corporate processes that look like perfect fits for AI have already been highly automated. Unlike AI, they are auditable.
Look at flight dispatching at a major airline. This involves filing the flight plans, checking the weight and balance, calculating and ensuring the fuel load and making sure the plane can take off and land. By the 1980s when I did a code inventory for a major airline, the code was extremely sophisticated and embedded all sorts of sometimes hard earned knowledge. In theory, the airline could run with the dispatchers doing little more than hitting OK now and then, at least if no problems came up. It is hard to imagine how AI could improve this process.
It's a similar case for a shipping or inventory management company. Unless it's a momma-poppa operation, odds are they have sophisticated algorithms already at work, and those algorithms are auditable and can account for the odd size of SKU7812556 and the problems making a U turn on Forest Road 6257.
Ditto for the code validating credit card purchases. Ditto for the code running internal combustion engines. Ditto for the code running production machinery. There's a lot of highly tuned code is already paying for itself. If the code is going to run a million times, odds are there isn't much room left for AI. That means AI is going to have to get good at one offs and small production runs. Given the high cost, those are going to have to be high value applications.
Re:scramble for janitor and home-health jobs.
With the birth rates falling below replacement world wide, it seems those jobs will become more automated.
My age cohort is the fastest growing.
Robots to lift us patients from beds and wash our bodies clean and put on a clean diaper. Robots to clean the room. A robot to keep us Alzheimer's patients engaged. Robots to assist us in getting dressed and in and out of the bathroom. So many tasks to do.
From what I read the care for the aged is a very tiring task that wears out caregivers. If you want to sell billions of devices, mobility assistance robots is the tech to develop. I am on the ragged edge of needing that assistance device now.
I think "copilots that help programmers refactor and remember syntax" is a pretty big understatement with respect to the current capabilities of, say, Claude Code.
What if OpenAI had not given ChatGPT to everyone at first? Because I think that led to the insane defensive investment, which required too many investors to promise the sun, the moon, and the stars. Maybe the hardware investment can be used for other models which are less cool but have a moat and profit. Which also probably requires getting off the NVDA software.