At some point the token costs will increase so much, it will put all of these AI companies out of business.
Don't get me wrong. AI is definitely useful. The high costs are currently being subsidized by VC and other investors and unless there is a major breakthrough, it's just not going to work in the long-term.
I truly believe the cost per token for most people will eventually go approximately to zero even in the absence of subsidy unless there is a dramatic state intervention to curtail free computing and hardware R&D and manufacturing.
Your premise of full token commoditization has a major wrinkle though. GPU-centric data centers with never allow tokens to go to zero since you have a 1.5 year technical deprecation and 5 year life on the GPUs. Maybe Cerebras or similar chip builders, after writing model weights to silicon, will improve those returns on capital and enabling the scaling you are envisioning.
From my perch, AI at the edge is severely underserved, and the GTM for the hyperscalers mostly ignore it. So if someone figures out how to scale AI at the edge (nvidia + HF perhaps) then I think your premise is on point.
1) "AI companies will go bust because tokens will cost too much" >> 2) "Don't let them build data centers that will sit empty" >> 3) Demand outstrips supply by a lot >> 4) Token costs too much.
This is no different from the NIMBY's that block all the housing then sit in a huff saying "see! everyone's leaving. good thing we didn't build anything!" when the rents go up.
I think that's always been an issue in tech, but it feels like AI exacerbates the issue. I see people starting companies and services where their primary domain knowledge seems to be whatever their AI tells them. It's sometimes enough to get off the ground, but I wonder how long it takes for customers/clients who actually need the domain knowledge to bail.
Its just that in a gold rush, there are few winners and many losers. There used to be hundreds of car companies in the beginning, only a dozen or so survived long term.
I'm not sure whether the salaries of the new jobs are on par with those of the lost jobs.
Except that is hogwash, "attributed to AI" is basically the CEO screwing up something in the business and AI-washing their mistake because it's convenient.
For AI startups there eventually may be no tomorrow
How many of these startups are profitable
Perhaps the greatest "AI"-related job losses will be failed "AI" startups
More BS "jobs" created from throwaway investment capital instead of business success
I would not take extra hiring now as evidence that AI is preserving job numbers.
Not a super insightful article. The AI jobs apocalypse may or may not be real, but this article doesn’t provide any compelling evidence for its thesis.
It might be benefiting other areas, but I think as the technology improves a lot of those are going to start looking like what's happening in software development (which is arguably what the technology is best at right now).
So yeah, idk.
"The housing construction jobs are temporary, we shouldn't count them."
There will always be CapEx to spend.
Many of the current DCs being built out are also being powered or tied with renewable energy projects that were funded by the IRA, as many DC builders with deep pockets do know the political dimension. If this is what keeps the CHIPS, IRA, and the IIJA alive I'm happy.
Every industry and career type goes through ebbs and flows. The 2010s was a white collar boom, and the 2020s is now a white collar bust. Meanwhile the 2010s was a skilled trades bust and the 2020s is now a skilled trades boom.
I'd be rich if I knew what the next big thing was but the economy has been running on the assumption that there will be a next big thing for many, many decades. Each cycle has the same rhymes of "what happens when it ends" questions but there has always something.
For better or worse, it looks more like that we're gonna tile the whole damn planet in data centers, at a faster and faster clip.
High productivity is generally good for the employee, unless you're productive enough you run out of work to do. So how much code is needed?
I mean I use it every day of course, but the velocity of work doesn’t really seem to be budging lol. And frankly a lot of time wasted on AI made gadgets, and meetings about those gadgets (look at this dashboard Claude made! etc)
The joke is at the level of developers, the constant sharpening of AI merge-requests reviewing (sometimes too much) has made the new job of AI speedriding & reviewsfighting a serious brain squeeze.
The value of conscienciousness has skyrocketed, and the value of openness is still the same as it ever was.
I'm not disagreeing that some manual work require forms of intelligence that are undervalued in our society, and don't get me started about nurses, but still: What the Jeff Bezos!
It depends on how you define “spoils.” People define it to include notional paper wealth, and I think that doesn’t reflect the actual mechanics of economic imbalance. On paper, NVIDIA’s shareholders have increased their wealth by $4.8 trillion in five years, enough to buy all the residential real estate in New York City and LA combined. But $5 trillion in notional AI wealth isn’t actually the same as owning NYC and LA. The paper wealth doesn’t affect the average person meaningfully until someone actually buys something with it.
Given that, it’s more useful to define “spoils” in terms of finite goods and services and look at spending on those goods and services. That is what directly affects the average person—it’s what makes housing or education or Disney tickets or whatever harder to access.
And on that front, the income growth of the top 10% has had a much more significant impact on the average person. The top 10% accounts for half of consumer spending today, up from 35% in the 1990s. And that spending is coming from the upper middle class, not billionaires. Bezos might be worth a million times more than the average person, but he’s not buying a million times more housing or Disney tickets or hotel rooms. Instead, the average person is competing directly with the increasingly affluent top 10% for access to those finite resources.
That's not going to happen.
Transparent pixels harm the environment because they use energy and don't even do anything
As a programmer, but working in two non-SWE roles, I have turned job A 70% over to AI, and job B 90%.
Where systems and scaffolds can be put in place to manage various aspects of a job, would this not be considered, partially, an "AI job"?
If you work in government grants and R&D, what if agents can now triage 100-1000x the opportunity count that you could ever handle alone, and you had a pipeline system in place to manage said opportunities and relationships? And then you, as the human, become the mouthpiece and reviewer of hoards of filtered output? Is that Federal Grants & R&D position now not mostly an AI job?
I'd say a large percentage of digital-based jobs can be considered varying percentages of AI jobs.
AI is also creating a new class of white-collar jobs. Engineers build the models, data annotators label their inputs and judge their answers, “forward-deployed” engineers adapt them for customers, and newly minted “heads of AI” decide what companies should do with the technology.
> On September 4th the Bureau of Labour Statistics reported that the American economy added 162,000 jobs in August, far above expectations.
Good news that ought to be taken with a pinch of salt given how the current administration responds to unfavorable BLS data -
> On Friday, August 1, [2025] President Trump fired Dr. Erika McEntarfer from her role as the Commissioner of the Bureau of Labor Statistics (BLS). In social media posts announcing and confirming her firing, he said that he believed the jobs report released by the agency earlier that day included numbers that were “rigged in order to make the Republicans, and ME, look bad.”
https://www.epi.org/policywatch/firing-bls-commissioner-erik...
Last month got revised to being positive, at the same time that this month was being released.
Messenger 3: My Lord, news…
King: What?
Messenger 3: Lord Wessex is dead.
King: This news is not so good.
Messenger 3: Pardon, My Lord?
King: I like it not. Bring me some other news.
Messenger 3: Pardon, My Lord?
King: I LIKE NOT THIS NEWS! BRING ME SOME OTHER NEWS!!!
Messenger 3: Yes, My Lord!
(Messenger 3 leaves; King tosses things around angrily; Messenger 4 enters — actually just Messenger 3 pretending to be a new messenger delivering new news.)
Messenger 4: My Lord, news…
King: What?
Messenger 4: Lord Wessex is not dead.
King: Ah, good news! Let there be joy and celebration; let jubilation reign!
Trust me I do the same.
The current head of the BLS seems well qualified. Here are some select quotes from a NYT article [1]:
“Erica Groshen, who was commissioner during the Obama administration, also praised the selection. “He’s a B.L.S.-er, he’s a data nerd,” she said.”
“The firing, and Mr. Antoni’s subsequent nomination, raised concerns that Mr. Trump would seek to interfere in the agency’s operations and perhaps try to skew its estimates to his political advantage. “
“There is no evidence that has happened so far. The agency has been run since August by William Wiatrowski, its longtime deputy commissioner, who also led the bureau on an interim basis under President Joseph R. Biden Jr. Economists continue to view the numbers the agency produces as reliable.”
Can you cite any specific examples of wrongdoing? Or inaccurate numbers or figures outside of normal revisions?
When I see bad numbers from the BLS I say yes that’s probably true. When I see good numbers from the BLS I say yes that’s probably true.
[1] https://www.nytimes.com/2026/01/30/business/trump-bureau-of-...
also AI boosters: "we are going to build a million robots, and the robots will build factories producing robots, there will be infinite abundance, nobody will need to work anymore"
Maybe.
> nobody will need to work anymore
If we define "nobody" as very wealthy people who partake in that abundance while everyone else ... goes away.
Humans were simply nowhere near answering all the useful questions they had which software could answer. The problem was that software was expensive.
Now, software is cheaper and so people who couldn't afford to answer questions they've long wondered about can afford to answer them.
If the software isn't downright free – today, you can slap a $349 RTX 5060 [†] into any POS surplus computer and have an offline assistant, running Llama or Ornith, via Ollama in linux (i.e. the LLM and OS are FREE open-source software) [∑]
I have a working demo of this that has been shown/leant to several friends, and they're all surprised that it "all works so well, without being online, for only a few hundred dollars."
When I've demonstrated Mistral-small on my 5070Ti... that has been a real jaw dropper. I haven't shown anybody qwen3.8:27b, yet... but dammmmm, what a month of learning it's been.
----
A friend that was going through wifecancer confided in me that "you can ask it anything, without feeling embarassed" – and that has stuck with me (that so many smart people are afraid to ask simple questions [*], out of perceptional worries).
----
[†] (8GB DDR7) brand new from Wal-Mart
[∑] My first linux/LLM machine was built with setup help from Perplexity.ai (with a dozen "assists" - I am bluecollar, non-coder). This used an obsolete i5 (and $200 used VEGA64 GPU) to create a decent Llama3.1 LLM machine (~100wpm typeback, perhaps 70 tokens/s).
[*] even to their own detriment, of shame, when simple solutions often do exist
Temporarily. Token price will rise sharply when the bills finally come due.
That provides continuous downward pressure on token prices even if it isn’t Astra level
???
From a few months ago: https://www.economist.com/content-assets/images/20260516_DE_...
Note well: This does not mean that I agree with any part of what he said. Just explaining the term.
Now protests are getting louder and the midterms are close.
