Behind the WSJ GenAI Numbers is History and this Collison
(updated 11.15.2025 to add in Telco and smart input from my GenZ consultant out of Singapore)
Yes, the Hiddenburg is hitting the Titanic as it hits an iceberg.
Today’s WSJ article Big Tech’s Soaring Profits Have an Ugly Underside: OpenAI’s Losses is at the end.
WSJ Solid Article - But Misses Two Big Points
Basically, the smarties at WSJ are highlighting just how much the big AI firms are spending to support growth versus their actually quite impressive revenue numbers. OpenAI could double its revenue each of the next two years to approx. $50B. Meanwhile, they would still have a loss measured in 12 digits. Yep, 12.
That, however, is not the problem. The problem is innovation and history conspiring to fly the Hiddenburg into the Titanic.
But First, Shocking Reality
To be clear as a jaded long term in the trenches tech / BU leader strategist start-up infused chap I do think what we are seeing in the AI space across GenAI, Agentic, Hyperscale/Hyperspeed, Supercomputing and Quantum is just tech magic, a new level and pace of change.
If I had suggested a bet five years ago that was has happened in AI would happen or that green men from Mars speaking French would land on earth, bets would have split pretty evenly. The insiders would have said that much tech change that fast was just impossible/unthinkable.
The impacts will be Industrial Revolution scale and bigger than most think. We will see quantum leaps across science and business while the time gap between scientific 'discovery/breakthrough' and real-world commercial impact shrinks by 80% or more. The changes and impact as startling and impressive as they have been over the past few years, are just the very tip of the iceberg (bad metaphor). I can explain in detail why we are about to hit 10 exponential, somewhat interconnected, curves that will change the world but that's another article and we have already been talking about a 4th Industrial Revolution since Davos 2017.
(The next 8 years will make the last 4 years of huge change look small).
The Pain
There were two streams of pain in an era of Industrial Revolution(s).
One was just competitive wreckage. In 1912 there were 400 vibrant companies making cars in the US. Twenty years later 360 were dead, 37 were struggling and 3 owned 80% of the market.
The second was that jobs were destroyed while new jobs emerged. Often that happened in geographically different locations and with mismatched skills. The textile industry that was the anchor of the Yorkshire economy or the same in Phili or the Glasgow Shipbuilding sector or canal transport left and never came back. Urbanization boomed because we mechanized farming and created industrial factory floor jobs. From 1870 to 1920 there was a lot of labor based strife because the rate at which agricultural jobs were lost was not matched by new industrial jobs, geographic friction/mismatch, skills mismatch and the more cyclical nature of industrial jobs being much more sensitive to boom/bust cycles.
In this current AI revolution it isn't as much as jobs vanish, as it is that AI augments humans, processes become intelligent and errors/rework are proactively eliminated, all of which makes things that much more efficient in terms of labor content. At a macro level that means less jobs, less work. New jobs are created but my best guess on a five to ten year horizon is the efficiency cuts will be 10 - 20x greater than the job adds. On top of that AI is being applied to improve quality and process beyond human augmentation. The flagship BYD plant has 97% process automation, 30% efficiency gains from AI and intentional work on what they call Quality Automation has reduced rework an manual inspections have reduced labor time per car by as much as 85%. Ironically EV's a more labor intensive to make than combustion engine vehicles but still the labor share per drops every year because of tech/AI.
I have mentioned before we are not prepared and we are not preparing.
Harvey , the brilliant AI platform for the legal sector, augments humans doing lawyer stuff. You then need less lawyers/paralegals. That is their entire model. They have a valuation of $5B. They were born three years ago. Good news is they have hired people to work at Harvey which are 'new era' jobs. Bad news is it's about 500 total, also mostly not lawyers.
The other huge wave impacting employment numbers is what is happening in Robotics, a field that is just starting its acceleration curve. Once that industry architects for unit cost impact and is less fixated on making robots like humans that curve will skyrocket. I was at FedEx for 17 years in the early , so fun, huge growth days. If I was designing robots to run logistics warehouses I am pretty sure none of them would be bipedal forward sensing meat sacks with just two arms. Not really sure I would need any humans onsite. In 1995 my team put in a fully robotics based pick and pack function in Penang for Intel chips. An order say from Austin would come in. It would get picked/packed in Penang and no human would touch anything until that container got opened for the morning sort to FedEx vans at the local station in Texas. 1995. We are now just at the beginning of a Robotics exponential curve.
I advise a wicked smart CEO at a cool cancer biostatistics two sided network startup Olira that reduced their human coding needs by 90% using AI tools (and improved quality.)
From 1873 to 1929 there were 14 depressions/recessions starting with the Long Depression of 1873 kicked off by 5% of the banks failing and lasting for 65 months. In the 46 years prior to the Great Depression and the crash there were 29 years of depression/significant recession.
The Point of the WSJ GenAI Story That Was Missed, Actually Two Points
Again this is pretty clear if you see the history of tech developments be it infrastructure, weapon systems, transport or tech itself. The big difference with tech is the speed at which is has evolved and the new accelerated pace.
The computational cost to achieve GPT-3.5-level performance has dropped more than 280-fold since late 2022. That's like you buying a new Bentley in 2022 and me getting one now for $1300. Or my flight to LAX from wonderful Newark dropping to 68 cents round trip. Okay that's a mid week, Spirit Airlines, probably ted up in the cargo hold fare but still a real fare quote from today, so you get the point.
So what happens to sectors that have marvelous innovation , huge capital inflows, massive revenue / growth and big losses? Innovation happens.
Genius Barksdale (real name Jim - Fedex Pres.) used to say 'revenue growth hides a lot of sins". I just heard Henry Curr say "valuations hide a lot of underlying weakness" , on the marvelous The Economist annual Future Cast chat.
Typically Wave One innovation companies get murdered by Wave Two who look at the gaps, architectural improvements, tech leverage and design for innovation plus cost structure. Wave One innovators don't spend a lot of time designing for cost structure even though Toyota invented concurrent engineering and the commensurate design side so long ago there was still a Berlin Wall (in Berlin).
Kimi2 from Moonshot.ai in China, founded 3 years ago, was released in July and has 85% better inference cost structure. than comparators The overall space is on an early trajectory of roughly 80x cost structure improvement.
So Wave One versus Wave Two lessons from history is the first thing not mentioned in the WSJ article. The second is that there is always a secondary migration after big platform changes/advancements to the profit pools that are the consequence of the new platforms.
History Again - Profit Pool Migration
There was huge wealth creation as the roll out of rail networks happened. A lot of it was tied to speculation and huge stock price increases. The most intense speculative booms occurred in the U.S. during the 1860s–1890s, and in the UK during “Railway Mania” in the 1840s, when railroad shares dominated investment markets and created fantastic fortunes for those with capital and insider connections. What does that sound like ?
A tip of the hat to my awesome daughter, Esther, who a few years ago as a New School history major showed me the the Politics and Poetics of Infrastructure, sort of Infrastructure, Tech, History and Anthropology blended together (see Brian Larkin ) which is about the ripple effects and societal changes primary and intentional infrastructure build out has. Her case in point was NYC pre Croton Aqueduct and post, ie, fresh water supply. The NYC economy, 1820 to 1860, grew about 11x the bulk of that coming after Croton which had 10 ripple effects in healthcare, sanitation, etc.
Cornelius Vanderbilt was, depending how you calculate it, top 5 or 10 richest people in the world. He literally had more money in the bank than the US Treasury. In 1860 he shifted his empire building/investment focus from steamships to railroads.
In 1873 the bubble burst, stocks crashed, 1/4 of railroad firms went out of business/bankrupt and that triggered the Panic of 1873 with years of depression and social unrest in an era with no real public safety net. As the richest country in ta fierce global market competition he world we should worry about our current proactive job supply focus/system (because we don't have one) or for that matter the lack of a decent safety net for those in need. Forget any moral, poltical or ethical rationale but not mamaging those better dilutes national competitiveness and the future talent supply. Imagine two countries 10% to 20% of the people in the future talent pool (kids) experience food insecurity. What does that do to the economic engine and competitiveness? The economic cost of not fixing this is less than the ongoing impact by a huge margin. US GDP per capita is $86k which is 57% more than Canada and 165% that of Japan.
In parallel though what happened was value migrated from the once new tech of railroads to the secondary areas where it spawned new models, profits and industry. Standard Oil and Sears are two prime examples that were innovations literally enabled by rail . As that happens the infrastructure folks don't get fully commoditized but they do lose their shine and astronomical valuations.
Today that is possibly a little bit worse because of the speed of Wave Two coming. Couple that with value migration for AI moving from infrastructure to impact (like Harvey) and you have a similar railroad story.
Telco Sidebar
In the Telco world with 4G the value migrated above telco networks especially in the US. Think of the windfalls of Facebook, Uber, the App Store etc. 5G which first went live in early 2017, is literally a software defined, AI intensive, fat bandwidth, super low latency Edge Compute platform run and built by companies that have oceans of valuable data. Reminds me of railroads and oil fields and Standard Oil/Rockerfeller. Still Telcos are stuck and leave the value migration to others. Two of the three big ones spend bzillions on ads saying they have the best network and how Bily Bob can call from some rural place without buildings in sight and the other just says 'Nu uh, our network is better'. The third has shaken up leadership and is moving beyond network to 'customer centricity' . At FedEx I was part of the team that helped land the big national prize for Service/Customer Centricity award ... in 1990. Minor detail almost 85%-90% of cell phone revenue (and almost all of the margin) comes from urban/suburban areas where maybe 1-2% of those customers time (aggregated) is used traveling in truly rural areas.
More on this later but the coming tsunami of change anchored in a 'Demand for Intelligent' and harvesting the profit pools in being proactive, preventative, pattern matched, predictive, precise, personalized, process performant, peer connected permissioned, and pragmatic (Toby's 10P's) fits the assets 5G telcos have but only as ecosystem solution partners not as just the plumbing supplier. TMo has a smart team after some innovative things in this area but like all three big players the percentage of revenue they get from non network Enterprise/Consumer solutions is tiny. There is a small window for a strategy refresh.
One other thing to think about if you are a telco is satellites. Launch costs (per satellite) are down by a factor of 650 and unit data transmission costs once in space are down 1500x, maybe more , over the past 15 years. What that looks like when we have three providers with 5k to 6k LEO satellites is hard to fathom but it is coming fast. The top three firms will have 16k to 20k satellites up within about two years. China's G60 constellation will have 12k.
China is Going After Wave Two
China, via its Action Plan for Global Artificial Intelligence (AI) Governance (published July 2025), is already officially focused on key domains above the infrastructure for AI:
Scientific Research and Technology Innovation
Industrial Development and Manufacturing - including robotics and intelligent supply networks
Consumer Services and Applications
Public Welfare, Healthcare, and Education (includes smart/efficient cities)
I can also pretty much guarantee they are applying Wave Two innovation thinking about impact and cost structure and 'bang for the buck' to manufacturing, supply chain, weapon systems and military logistics.
Ukraine / Weapons Systems Sider Bar
Ukraine and its clever drone approach has clearly shown a cost-optimized architectural and design effectiveness path with drones. An anti tank Javelin missile strike costs $200,000 . A Ukrainian one way drone with a TM-62 Russian mine (ironic) strapped to it might cost $900. Also if you shoot a Javelin missile at me I can often spot exactly where you are. The F35 if it were designed with a bang for the buck, an optimized lifetime cost structure approach would be hundreds of billions of dollars cheaper for the life of the fleet and better. Ditto the Littoral Combat Ship. Why does an FP5 long range missile cost 1/5th of a Tomahawk, carry 2.5X the explosive payload , travel twice as far, and flies faster ? It is largely because it was designed/made in Ukraine where the scarcest primary input for any weapon system is money. They thought through the lifetime unit cost structure per unit impact, which US weapon systems almost never do nor giant US AI platforms it seems. Yes, a Tomahawk is 'which window do you want it to fly through' accurate but with 2,000 lbs of high explosive on it and a 5 to 1 cost advantage does that count much?
Conclusion
Industrial Revolution scale change drives two types of destructive change: commercial wreckage and massive, high friction job displacement/elimination .
Wave Two innovations kill off a lot of the Wave One firms and profit, focus and speculation migrate above the 'infrastructure'.
So Wave Two is about a new plateau of smart innovation often as the speculative shine and valuations on wave one migrate away to the profit pools the new infrastructure has created. Wave Two does things wave one never did.
We used to have 50 years to absorb an Industrial Revolution scale change. This time we will be lucky if it's a decade.
Modern Wave Two Examples:
In Hyperscale/Hyperspeed data Ocient is a narrow example. It is 1000x more effective per unit spend in some areas. That doesn't mean they killed all the slower more costly folks although some of that is starting to happen. It means new profit pools emerged because with prior tech it just took too long and cost too much to actually attempt it. A great example is what they did with the National Crime Bureau in the UK. There are different consequences for criminals if I can get complex analysis done in 3 minutes versus 3 days. Workday is a Wave Two company. I fear Waymo, Xpeng, Baidu, BYD will be Wave Two to Tesla on driverless cars simply because they started later, leveraged more modern architecture, not just for the cars but for production. The unit cost trend for BYD over the last decade is mind-blowing. It is intentional. It is anchored in design for that journey not just the journey on the road.
History. It does not really often repeat itself unless you abstract to a level where the pattern fits. But what history does do is obey the gravitational and physical forces of change and like a boulder gaining momentum crashing down a hill follows a particular trajectory, one that is pretty predictable until it loses energy and stops. Of course, then there's another boulder with its own trajectory.
Again, we are not prepared and we are not preparing. More on that later.
The one fundamental positive element in all this is humans are bad at change and all businesses are process systems run by and full of humans so the pace of tech improvement will be throttled at the coal face/factory floor by the fact that we are humans. In large change management efforts the people issues are very often mismanaged/under-managed. Most companies can't tell you explicitly what change management model they use. But that's a different article for later.
Thank you both for reading this, switching to decaf :-)
