• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

How AI will fail

Ididar said:
SpaceX is going the same route. They're going hardcore into data centres and chips because Grok is falling behind everyone else. xAI is quickly going to become a hardware company, not an AI company.
Click to expand...
they just spent 60 billion on cursor and both benefited from each other quick.

Cursor gave xAi near best in class data, xAI gave cursor best in class compute will have to see a couple of cycle if they can keep an interesting quality per dollar (which arguably they fully achieved with grok 4.5), a bit more expensive than Muse spark but significantly cheaper than the biggest open weight model like GLM 5.2-K3.

It is not keeping up with openAI-anthropic (no one does) but it is with everyone else, ahead of google until they update gemini arguably and they were the last to start in the race, so more pacing people than falling behind.

And because xAI is in part involved (intent layer) of self driving and robotics, the spaceX physic-aeorospace AI simulation, and other first hand use case, they will stay an AI company no matter what externally pretty much and if tesla and spaceX merge (highly probable) they could get more and more involved in tesla product AI.
 
Claude Fable 5 failed me.

Adding "tribute" and "personal usage" doesn't really change what's being asked, unfortunately — the spec is still a point-for-point recreation of the Doom Slayer: the Praetor Suit, the Praetor helmet emblem, the exact color scheme, the signature loadout. Whether it's called a replica or a tribute, the output would be the same protected character design, so I have to give you the same answer as before.
 
Not failing today!

Patch verified and applied correctly at lines 21998-22015 of gateway/run.py. When an approval request fires:

_approval_notify_sync runs after redacting credentials
Calls text_to_speech_tool("Thank you sir, may I have another?") via Piper TTS (using your configured voice/model)
Generates audio file in temp directory
Then sends the approval prompt to chat
 
SOXX is now 200 pt. drop from the top, as of today's nose dive close, but it's still very high, mutual fund closes their book on Halloween day, and hedge fund closes their book in Nov. 30, so I expect much more profit taking as they have to show their book for the yr. on huge profit
 
Last edited:
AI will not fail and it's unfortunate that 99% of people think all AI is chat bots and customer service. But the reality is companies have AI doing so much more and it needs a lot more processing power than chat bots. IMHO, this isn't close to over.

Edit:
Minor tweaks for clarity.

But also, to add that companies that don't rely on AI for LLM have absolutely zero use for DeepSeek or Kimi. You can't use either one to look at a pathology slide and identify cancer cells. You can't use either one to find you a possible super conducting material.
 
Last edited:
I agree the technology is not close to over, but people like Big Short legend Michael Burry is saying the market is ahead of itself. AI does make $, but not the kind of P/E that these chip stock is showing.
 
Happy Hopping said:
I agree the technology is not close to over, but people like Big Short legend Michael Burry is saying the market is ahead of itself. AI does make $, but not the kind of P/E that these chip stock is showing.
Click to expand...
So, historically, nearly every major technology has had a bubble. Bubbles are the norm, not the exception. So, a crash isn't unexpected. But how long until the inevitable crash is the question. After all, markets can remain irrational longer than you can remain solvent.

But the one thing I really fear is the whole notion of having the government invest in these companies. Because the returns might not pay off in your lifetime. And then the question comes about, do you as a citizen actually directly receive money from it, or is it the government using it for their own spending? Basically a backhanded way for AI companies to steal your money, but this is more of a soapbox thing.
 
When CBDCs and carbon footprints become part of our personal data, the government will need AI to manage and monitor these matters in detail and in a centralized manner; this is one of the reasons the government is investing in and implementing AI in its administration.
So, yes, some projects or companies in the field of AI will fail, but not all of them.
 
Nytegard said:
But the one thing I really fear is the whole notion of having the government invest in these companies. Because the returns might not pay off in your lifetime. And then the question comes about, do you as a citizen actually directly receive money from it, or is it the government using it for their own spending? Basically a backhanded way for AI companies to steal your money, but this is more of a soapbox thing
Click to expand...
You might not but your progeny will. Americans need to stop thinking about the 'me' and think about 'us'. At least thats the way marketing machine in China makes you think. The boomers over there are all about future generations.
 
The Lurker said:
You might not but your progeny will. Americans need to stop thinking about the 'me' and think about 'us'. At least thats the way marketing machine in China makes you think. The boomers over there are all about future generations.
Click to expand...

The USA has pushed individualism so strongly that causes real problems when trying to think about communities as a whole. Not impossible, but more difficult in the US than other countries.
 
The Lurker said:
AI will not fail and it's unfortunate that 99% of people think all AI is chat bots and customer service. But the reality is companies have AI doing so much more and it needs a lot more processing power than chat bots. IMHO, this isn't close to over.
Click to expand...

The whole anti-AI sentiment is setting up people for disaster. They will be left behind. This is not a maybe situation.

After using AI quite a bit in the last few months and really learning how to delegate agents, run workflows, etc - I now see this technology is as revolutionary as the internet. Yes, many people are trying to use LLMs to one-shot slop hoping to strike it rich quick, but I've got 20 years of coding experience along with a PhD in engineering, and I safely say: my days of programming are over. They are just flat out done. It's no longer something I can present a strength.

Because it's now incredibly inefficient and counterproductive for me to write any more code, when LLMs are doing that job well enough in a fraction of the time. And I'm actually more than fine with this, because I find it much more stimulating/fun to architect along with LLMs as a co-brain + solutions implementer + verifier. I'm basically there for the beginning of the task to form plans and guidelines, and then also there at very end to make sure the final output works.

SAAS is also completely dead, now that I can spin up my own "personalized apps" with no logins or internet connections needed.

FWIW this is my personal AI project (I'm trying to develop deterministic bots that still feel human-like for my game - I'm working on an offline version - which is eventually going to simulate the online version entirely).

View: https://youtu.be/V17V3NUbM1w?si=yS2OjfS4HGsUh_8X

The above took about a month of heavy prompting/use on a Max 20x account. It's something that would have easily taken me a year or more if I was doing everything myself. I no longer care for the anti-AI sentiment. Edit: I should clarify that; I no longer care for that sentiment for the sake of counterculture. There are real problems I am against - for example, I don't like that datacenters are being built next to residential areas causing issues for the residents. But that's an issue that doesn't really have to do with the tech itself.
 
Last edited:
blade52x said:
SAAS is also completely dead, now that I can spin up my own "personalized apps" with no logins or internet connections needed.
Click to expand...

That's an incredibly naive approach. SaaS won't be the same, but it will exist where it provides value. Take cloud based email. THe service goes beyond providing an MTA. It also provides all the services you need for email to work well. Server redundancy, storage redundancy, a devops upgrade and fault handling system to patch it all, reputation management, large provider peer negotiations and back channels, etc. If you don't know how to do all that, you don't know enough to prompt your AI and check that what it spits out is viable.

Software dev has a skewed perspective of where AI is on the overall adoption and profitability timeline because they work in one of the areas AI is best at and don't realize what impacts success overall and where the industry is at with that. You are also able to exist well in the loss leader buffet model. So the cost looks very reasonable. Once you start graduating into the pay for everything you eat at market rate side of things, the price goes up a LOT. If they don't get that price down, it's going to hit a wall. We are using it a lot for dev work at my job. We are rapidly approaching costs where it's going to be costing more than hiring people, so it has to be better than hiring people. Even if it is, it's going to hit the same constraints on spending that staffing does when money is tight. Even if they can get the down to maintain that cost advantage at scale, they risk spending on build out incorrectly enough to kill their company.

Will AI go away? No. Does that mean the bubble won't pop and mess up a lot of people's plans? Also no.

One of my overall predictions for this is that Apple will wind up a big player at some point. Not because they are going to do it better, but because they are not going to lose money investing, will take short term advantage by simply renting from the market, and have a very good chance of having lots of cash on hand when some of the existing players are selling off assets and IP while closer competitors are carrying too much debt to outbid them.
 
raz-0 said:
That's an incredibly naive approach. SaaS won't be the same, but it will exist where it provides value. Take cloud based email. THe service goes beyond providing an MTA. It also provides all the services you need for email to work well. Server redundancy, storage redundancy, a devops upgrade and fault handling system to patch it all, reputation management, large provider peer negotiations and back channels, etc. If you don't know how to do all that, you don't know enough to prompt your AI and check that what it spits out is viable.

Software dev has a skewed perspective of where AI is on the overall adoption and profitability timeline because they work in one of the areas AI is best at and don't realize what impacts success overall and where the industry is at with that.
Click to expand...

Agree. I'm in engineering on the special missions/military side of the aviation industry. Nobody has an AI model that is going to go in and measure up an existing aircraft for me, model components from physical parts, come up with design to mount everything, write the stress reports, and then write cert plans, test reports, and so on. Can it help with limited stages of that process? Yes. Is it wiping it out? No. Is it even, right now, providing any serious benefits? Not really. The only place where AI is making headway in my side of industry is in drawing reviews against standards. There are some decent efforts underway to provide people tools to take the review process out of the schedule but not much else (yet). It will get better over time but I have no AI in my workflow right now and nobody is screaming at me to introduce it. We are starting to look at using AI to do some interesting imagery analysis of the video we record but that is in its early stages right now. All custom stuff having to be developed. That won't happen overnight.

But, yes, coding is at the front of the edge right now both because it is good at it AND the people who understand code also understand best how to work with AI. They have a deeper understanding of the processes involved and how to manipulate it. Toss AI at an accountant and they're going to spend all their time scratching their head. You can't expect people who's job is not in the tech side of computers to adopt AI as quickly as those who live and breath computer hardware and software.
 
blade52x said:
The whole anti-AI sentiment is setting up people for disaster. They will be left behind. This is not a maybe situation.

After using AI quite a bit in the last few months and really learning how to delegate agents, run workflows, etc - I now see this technology is as revolutionary as the internet. Yes, many people are trying to use LLMs to one-shot slop hoping to strike it rich quick, but I've got 20 years of coding experience along with a PhD in engineering, and I safely say: my days of programming are over. They are just flat out done. It's no longer something I can present a strength.

Because it's now incredibly inefficient and counterproductive for me to write any more code, when LLMs are doing that job well enough in a fraction of the time. And I'm actually more than fine with this, because I find it much more stimulating/fun to architect along with LLMs as a co-brain + solutions implementer + verifier. I'm basically there for the beginning of the task to form plans and guidelines, and then also there at very end to make sure the final output works.

SAAS is also completely dead, now that I can spin up my own "personalized apps" with no logins or internet connections needed.

FWIW this is my personal AI project (I'm trying to develop deterministic bots that still feel human-like for my game - I'm working on an offline version - which is eventually going to simulate the online version entirely).

View: https://youtu.be/V17V3NUbM1w?si=yS2OjfS4HGsUh_8X

The above took about a month of heavy prompting/use on a Max 20x account. It's something that would have easily taken me a year or more if I was doing everything myself. I no longer care for the anti-AI sentiment. Edit: I should clarify that; I no longer care for that sentiment for the sake of counterculture. There are real problems I am against - for example, I don't like that datacenters are being built next to residential areas causing issues for the residents. But that's an issue that doesn't really have to do with the tech itself.
Click to expand...

Loving this, gotta get a better mic though to make your voice over easier to follow. Keep posting updates for sure!
 
Ididar said:
Agree. I'm in engineering on the special missions/military side of the aviation industry. Nobody has an AI model that is going to go in and measure up an existing aircraft for me, model components from physical parts, come up with design to mount everything, write the stress reports, and then write cert plans, test reports, and so on. Can it help with limited stages of that process? Yes. Is it wiping it out? No. Is it even, right now, providing any serious benefits? Not really. The only place where AI is making headway in my side of industry is in drawing reviews against standards. There are some decent efforts underway to provide people tools to take the review process out of the schedule but not much else (yet). It will get better over time but I have no AI in my workflow right now and nobody is screaming at me to introduce it. We are starting to look at using AI to do some interesting imagery analysis of the video we record but that is in its early stages right now. All custom stuff having to be developed. That won't happen overnight.
Click to expand...

I heard similar spiel about machine learning 10 years ago when that was still in its public infancy. Now every company doing anything scientific has their own ML team - if not producing anything than at least for show, to claim they are on par and up to date with modern talent. You have no AI in your workflow for a similar reason: what you do works. It has always worked. But so did the world pre-ML. And eventually, just like with ML - it's going to start to become about cost and efficiency. Who can produce a valid solution faster? Or cheaper? Or both? It's going to be those with AI in their workflows, because unlike ML which has problems its clearly is not suitable for, LLMs/AI are extremely broad in what and where they can be utilized.

P.S. Use the cloud-based LLMs to help develop your custom stuff (if allowed). They will help make that happen overnight. Because yes, they will create and test your data processing pipelines far faster than any person would manually develop on their own. You still need the person-architect overseeing it all, but yeah... :).
 
Last edited:
Ididar said:
Agree. I'm in engineering on the special missions/military side of the aviation industry. Nobody has an AI model that is going to go in and measure up an existing aircraft for me, model components from physical parts, come up with design to mount everything, write the stress reports, and then write cert plans, test reports, and so on. Can it help with limited stages of that process? Yes. Is it wiping it out? No. Is it even, right now, providing any serious benefits? Not really. The only place where AI is making headway in my side of industry is in drawing reviews against standards. There are some decent efforts underway to provide people tools to take the review process out of the schedule but not much else (yet). It will get better over time but I have no AI in my workflow right now and nobody is screaming at me to introduce it. We are starting to look at using AI to do some interesting imagery analysis of the video we record but that is in its early stages right now. All custom stuff having to be developed. That won't happen overnight.

But, yes, coding is at the front of the edge right now both because it is good at it AND the people who understand code also understand best how to work with AI. They have a deeper understanding of the processes involved and how to manipulate it. Toss AI at an accountant and they're going to spend all their time scratching their head. You can't expect people who's job is not in the tech side of computers to adopt AI as quickly as those who live and breath computer hardware and software.
Click to expand...

Coding is also at the leading edge not just because it started there more or less, but also because it has a lot of structured verifiable input and there was an absolute ton of the stuff that was already vetted to function out there to train on.

Lots of other things, even if they have potential applications, are having to work at getting good training data sets in place. Your examples from your industry are tractable, but you likely have to start doing things with an eye to creating a training data set, or go back and try to massage things into a training data set. Then you get to see if you have enough to be successful at it, and balance that with how to not leak anything that's a trade (or other kind) of secret to not undermine your own business.

But many things also don't have verifiable answers. It's why one of the things people are focusing on where AI can't help right now is matters of taste. I suspect to be any good in that realm, if it can, the stuff will have to be affordable enough for everyone to be walking around with personally trained models of sufficient oomf to do the job.
 
The software example is shortsighted. People are showing demos that cost thousands of dollars in tokens and look like generic demos freely available anywhere on the internet posted by hobbyists and students. Even before AI clever engineers could put together full apps quicky while writing very little code using open source libraries that were (presumably) battle tested and reliable vs totally unknown code spit out by an LLM. As a bonus these are teams of humans working together and all profiting vs paying crazy amounts to giant corporations that have a nifty side gig of surveilling everything you do and think.

Now one place I do believe the hype is security, finding and chaining exploits is something machines can do that very few people can.
 
myzz said:
The software example is shortsighted. People are showing demos that cost thousands of dollars in tokens and look like generic demos freely available anywhere on the internet posted by hobbyists and students. Even before AI clever engineers could put together full apps quicky while writing very little code using open source libraries that were (presumably) battle tested and reliable vs totally unknown code spit out by an LLM. As a bonus these are teams of humans working together and all profiting vs paying crazy amounts to giant corporations that have a nifty side gig of surveilling everything you do and think.

Now one place I do believe the hype is security, finding and chaining exploits is something machines can do that very few people can.
Click to expand...
AI has its use, and all but the most ardent haters realize it's here to stay. But it's also not wrong to state that it's being oversold. And that's just basic logic.

For example, let's say you're using AI to build nuclear reactors. And one's being proposed to be placed in your neighborhood. Now, a year ago, there was a 20% chance of an AI built nuclear reactor causing a Chernobyl level accident. Would you want that reactor built? Of course not, and I think that would be common sense. Now, today, Sam Altman, Mark Zuckerburg, or Elon Musk comes out and states their new AI is 100% perfect in building a nuclear reactor. Would you still want that reactor being built based on their word alone?

This is the same type of logic of why some people didn't want the covid vaccine, even though we were told it was 100% safe.

First, people lie.

But second, even if they're telling the truth, and even if you give the benefit of the doubt and assume that the AI is perfect, you'd be insane not to want it verified for at least a couple years. And I've seen too many people say, why not just have the AI document the code? Sorry, but documentation can be wrong also.

Until we go through a vetting period, the shift in jobs will probably have to be somewhat slower than the AI evangelists are predicting.
 
Happy Hopping said:
I agree the technology is not close to over, but people like Big Short legend Michael Burry is saying the market is ahead of itself. AI does make $, but not the kind of P/E that these chip stock is showing.
Click to expand...
Agreed. This is the same thing we have seen over and over and over....why is this time different? The dot com wasn't wrong. They just went too fast in the short term.
 
Speaking of music. We have fake A.I. bands like "The Velvet Sundown" and "The Fossils". Complete with A.i. generated images of the band, album artwork, and social media accounts claiming to have seen them live. The whole works. My GF and I listened to both both "bands" and they just sound off. Sure there are countless others out there now. These things will get better no doubt. Knowing several musicians both amateur and professional across several genre, I cannot say I will be happy about that.

"The Velvet Sundown: The AI Band Controversy Explained"

https://www.berklee.edu/berklee-now/news/velvet-sundown-ai-band-controversy
 
OutOfPhase said:
Velvet Sundown sounds like a line of undergarments for those who have just given up.
Click to expand...
Stained undergarments at that. The instruments on that album reminded us of people who play technically perfect lacking soul and feeling. No surprise given how it was generated.
 
RE: stock valuations you have to remember that stocks are priced in US Dollars which are currently being inflated away. Sure, so are earnings etc. but the flood of dollars looking for a place to not burn away to nothing will push prices up anyway.
 
It isn't AI failing that I'm most worried about. It is the students use of AI in assignments without checking the results prior to submitting said assignments. Taking away or limiting students ability to think for themselves is a travesty.

A Mississippi professor ran a test to see if anyone was using AI to for their assignments. The results are unsurprising.

https://www.foxnews.com/media/missi...als-hidden-way-expose-ai-cheating-viral-video
 
I bought my 3080 during the crypto craze and paid too much for it. I'm glad I already have my 7900XTX and won't need a new card for a while. Hopefully AI will die down. Then maybe we can finally admit as a species to stop buying into fads. But that'll probably never happen.
 
LukeTbk said:
Chinesse innovation can look way more efficiant than it actually is from the outside, as we do not see the 50-100 failures for each success story, they are simply 100% unknown of the outside worlds.

There was 300-400 Chinese ev car manufacturer at the starting line, lot of capital destruction to find out the top 5 winners, those winner are so good because they had to go throught the most competitive open market bloodbath in the world (and learned a giant amount of things to not do that does not work from the hundreds of failures around them) and individually they do look lean from the start as they were, but as a whole endavour not necessarily that much more efficiant that silicon valley.

Chinese Ai probably look like this has well inside in, as for deepseek model efficacy for what we know had less efficiant inference than what google gemini was about to launch (which was probably already using MoE and others advancement who are all based on google papers, they were openly by 1.5 pro launch) at the time and certainly nothing special now (same for kimi or glm 5.2):
View attachment 814039
Click to expand...
Alibaba’s Qwen3.8-Max Rivals Fable 5, Goes Open
 
https://www.linkedin.com/posts/dion...-opensource-ugcPost-7490299225197420545-DhZG/
1785864409776.png


https://www.linkedin.com/pulse/american-ai-domination-so-fast-scott-ortkiese-enlif/

1785864590432.png

https://www.linkedin.com/in/scott-ortkiese-012486104/​

August 3, 2026

The Tech Oligarchs, Bankers, and Lawyers Fed Their Greed with a Closed-Source Shakedown.

PART II Sequel to “Not a Cold War. A Market Rotation” 1 August 2026

Note to Readers: I wrote this as 20,000-word long-form article. Readers interested in a deeper dive, should contact me at so@throughlinesynthesis.com

China Built the Alternative. Open-Source Now Wins on Security, Pricing, and Performance, and the Trap Is Closing on American Greed.​


Yesterday I argued that the U.S.-China artificial-intelligence (AI) race is a market rotation, and that Washington is losing it.

Today I want to talk about the financial layer. The American AI stack is not just losing developer share. It is a solvency story, and the American greed coalition is hanging itself with rope it manufactured, sold to itself, expensed as capital investment, and financed with paper the private-credit funds are still marking at par.

The coalition is a stack: chip vendors (Nvidia, Advanced Micro Devices, Broadcom), hyperscalers (Microsoft, Amazon, Google, Meta, Oracle), a two-name customer credit portfolio (OpenAI, Anthropic), banks (Goldman Sachs, Morgan Stanley, JPMorgan), private-credit funds (Apollo, Blackstone, KKR, Ares, Blue Owl, Sixth Street), venture capital houses (Sequoia, Andreessen Horowitz, Founders Fund), law firms (Wachtell, Cravath, Skadden), accounting firms (Ernst & Young, KPMG, PricewaterhouseCoopers, Deloitte), comp consultants (Semler Brossy, FW Cook, Pearl Meyer), a captured press, captured think tanks, and a captured political apparatus. Each takes a fee. None is accountable for the ending.

On the other side of the trade is China, deploying the same technology as public infrastructure, on open weights, at cost, to the entire Global South.

The $250 billion backstop​


On 31 July 2026, tech reporter Ed Zitron reported that Nvidia would guarantee up to $250 billion of the SoftBank/SB Energy data center, with an additional $350 billion of Nvidia graphics-processing-unit (GPU) financing to fill it.

The chip vendor is guaranteeing the customer of its own chips, using its own money, to justify the data center that will house those chips.

This is a closed loop.

The cascade​


This is not one deal. It is a pattern.

1.Nvidia backs CoreWeave and Lambda. OpenAI’s primary neocloud footprints.

2.Advanced Micro Devices (AMD) backs Crusoe and Anthropic. AMD/OpenAI, originally 6 gigawatts, was quietly reframed as a warrant deal.

3.Broadcom guarantees an Anthropic-Apollo tensor-processing-unit (TPU) financing of about $35 billion. Same debt product. Different customer wrapper.

4. Google backs Cipher, TeraWulf, and a $15 billion data-center project. Google’s Anthropic exposure is now estimated at $76 billion for 2027 alone.

Seven-plus backstops. Four chip vendors. The same handful of customers. No transparency, only deceit.

Backstops that already failed​


The market has already run this test in 2025. It failed. Three headline vendor-customer commitments were announced with fanfare. None has produced a delivered gigawatt or a wafer at scale.

Repeat: The market has already run this test in 2025. It failed. Three headline vendor-customer commitments were announced with fanfare. None has produced a delivered gigawatt or a wafer at scale.​


1.Broadcom to OpenAI, 10 gigawatts. No delivery timeline. Silence since.

2.AMD to OpenAI, 6 gigawatts. Reframed as warrants. No chip delivery visible.

3.SK Hynix plus Samsung to OpenAI, 900,000 high-bandwidth-memory (HBM) wafers per month. Exceeds global HBM capacity. Mathematically impossible.

Three commitments. Zero closures. And the market rewarded every announcement anyway.

The post-revenue economy​


beta&t=zBlzAJTvnfIYW-x7Wy-xJfM2MNYURiU5isVqIDjrEXs.png

Catch-22, Making Up Losses on Volume?



1.OpenAI plus Anthropic have booked roughly $1.1 trillion of forward compute commitments through 2030.

2.Combined 2025 revenue: about $17 billion. Combined 2025 loss: about $26 billion. OpenAI’s audited 2025 loss alone: $20.9 billion.

3.Anthropic’s projected 2027 spend at Google Cloud alone: $76 billion. Plus $25 billion at Amazon Web Services (AWS). Before Anthropic’s own revenue crosses $10 billion.

That is a 65 to 1 ratio of commitments to revenue. That is not a growth curve. That is a solvency problem, dressed as a timing problem.

The credit tape is already repricing​


1.CoreWeave option-adjusted spread (OAS) is now above 900 basis points (bps). That is death-zone territory.

2.CoreWeave’s latest debt priced at roughly 9 percent yield. Worse than the average junk-rated borrower in the U.S. high-yield index at 7.2 percent.

3.Amazon’s July 2026 bond auction was oversubscribed only 1.6 times against a historical norm of 4 to 5 times.

4.Meta’s free cash flow collapsed from $8 billion to $800 million in one quarter. That is 90 percent of the cash cushion at the most cash-rich hyperscaler, consumed by capital expenditure (capex).

The credit market is ahead of the equity market. Again.

The 2008 language returns​


In the interview, Ed Zitron used the exact phrase “financial innovation” to describe what is happening in the neocloud debt stack.

That is the vocabulary of the 2007-08 subprime cycle. Collateralized debt obligations (CDOs), synthetic exposures, and monoline guarantees were sold as innovations right up to the point at which they detonated.

Financial innovation is the sound a system makes when it is running out of real customers and starting to invent them.

Prices are collapsing at the top of the cycle​



1.OpenAI cut its cheapest model, Luna 5.6, by roughly 80 percent. Cut its mid-tier model, Terror, by roughly 20 percent.

2.DeepSeek V4 Flash undercut the new pricing immediately. Blue Origin imposed a token cap to protect margin.

3.When prices collapse and commitments do not, the gap becomes the loss.

What breaks first, if this unwinds​


1.Neoclouds first. CoreWeave and Lambda. Thinnest capital cushions, most concentrated customers.

2.Private-credit funds second. Apollo, Blue Owl, Ares. Marking to model, not to market.

3.Chip vendors third. Nvidia, AMD, Broadcom. Tested on the size of their backstops when customers cannot service the obligations they took on.

4.Hyperscalers fourth. Meta already blinked. Google is booked at $76 billion for Anthropic 2027. That number will move.

5.Public equity last. It always is.

Why the West cannot win​


The West is not losing on engineering. It is losing on business model. The American AI stack is a rentier business. The Chinese AI stack is public infrastructure. A rentier business cannot outrun a free-download business once the free download reaches performance parity. Parity is here.

Behind the four wins on security, cost, performance, and control sit ten industrial capabilities the American coalition pretends do not exist: rare-earth refining at 85 to 90% global share, STEM graduates at three-to-one, Semiconductor Manufacturing International Corporation (SMIC) at 7 nanometers, Huawei’s Ascend accelerators within a factor of two of Nvidia’s H100, 400 gigawatts of new capacity a year at half the American marginal power cost, water-sited data centers, 1.1 billion domestic internet users, state coordination without state capture of the fee stack, an adjacent industrial commons in batteries and solar and electric vehicles and robotics, and a public-goods AI release cadence out of Alibaba’s DAMO Academy, Huawei’s 2012 Lab, and the Chinese state quantum program feeding DeepSeek, Qwen, Kimi, GLM, ERNIE, Doubao, MiMo, Hunyuan, and LongCat, aligned with the entire Global South.

Pax Silica unmasked​


The label the coalition uses for itself is Pax Silica. Palantir Technologies executives use it. Anthropic’s policy team uses it. Eric Schmidt’s Special Competitive Studies Project uses it. The Center for a New American Security and the Hudson Institute use it. Every declining empire names its extraction project after a peace. Pax Romana. Pax Britannica. Pax Americana. Each was a marketing document.

Pax Silica is a malignant, ego-driven, farcical moniker. It is self-congratulation dressed as strategy.

What the label conceals is four familiar drives:

1.Political capture. The federal government is a shareholder. Enforcement of existing disclosure and antitrust law would impair its own equity.

2.Self-enrichment. Sam Altman’s OpenAI equity, Dario Amodei’s Anthropic equity, Jensen Huang’s Nvidia stake, Masayoshi Son’s SoftBank portfolio, and carried interest at the VC houses. A wealth transfer unseen since the Gilded Age.

3.Entitlement. The presumption that the coalition deserves to own global AI infrastructure by virtue of leading first. The same claim the British East India Company made in 1780 and United States Steel made in 1890. Neither survived competition.

4.Contempt for the citizenry. The rate base, the retirement accounts, the water table, and eventually the tax base are treated as the coalition’s underwriting collateral.

The sanctimony inventory. The coalition vilifies China for behaviors the American AI stack demonstrably practices, at greater scale, with fewer disclosures. Data harvesting: trained on the whole internet without consent, accuses TikTok. Surveillance: every query logged in a United States-jurisdiction data center, accuses Chinese cloud services. State direction: executive orders, procurement, and export controls, accuses China of state capitalism. Intellectual-property theft: trained on every copyrighted book, article, and codebase, accuses China. Human-rights abuse: deployed inside a border-and-detention regime deporting people to third-country prisons, accuses China. Each accusation is projective. The coalition is describing itself.

How this ends for America: six moves​


1.Credit reprice completes. CoreWeave option-adjusted spread past 1,500 basis points. Amazon’s next AI-linked issuance 100 basis points wider. Meta free cash flow negative two quarters. Private-credit AI portfolios begin to mark down. 6 to 12 months.

2.Equity reprice follows. Nvidia, Microsoft, Amazon, Meta, Alphabet, and OpenAI reprice 30 to 50% over 6 to 9 months.

3.Retirement-wealth hit. With Magnificent Seven concentration at 30 to 40% of median household equity, $8 to $12 trillion nominal household wealth destruction. The 55-to-65 cohort takes the worst of it. This is the political detonator.

4.Political crisis. The government cannot enforce remedies without impairing its own equity. Hearings become theater. The Securities and Exchange Commission is captured. 2028 is fought over retirement destruction and bailout terms, and neither party is trusted to run either side of the trade.

5.Industrial hollowing. Utilities that overbuilt against AI demand are left with stranded transmission, stranded generation, and community water shortages, reallocated by public utility commissions jurisdiction by jurisdiction across 2027 to 2030.

6.Dollar reprice. Dollar share of global reserves falls below 50% by end-2028. Federal interest bill hits 30 to 40% of revenue. The fiscal state can no longer fund military, entitlements, and AI industrial policy simultaneously.

Moves 1 through 3 are already in motion. Once Move 3 completes, the political window closes, and Moves 4 through 6 run on autopilot.

The rope, the merchant, the buyer​


Part I said the empire is revocable to its users. Part II says it is revocable to itself, by a coalition that manufactured the rope, sold it to itself at full price, expensed it as capital investment, and financed it with paper marked at par.

The rope is the closed-weight, token-tax business model. The merchant is Nvidia and Advanced Micro Devices and Broadcom and the hyperscalers financing the customers who buy their chips. The buyer is a two-name credit portfolio (OpenAI and Anthropic) whose end-market revenue does not cover the compute bill and whose only path to solvency is a bailout characterized as national security, paid by the same American public whose retirement accounts, water tables, and electricity bills already paid for the buildout.

The credit market has begun to reprice this. The equity market has not. The rotation is already happening. The reckoning is what happens next.
Click to expand...
 
MrGuvernment said:
https://www.linkedin.com/posts/dion...-opensource-ugcPost-7490299225197420545-DhZG/
Click to expand...
many element of that piece seem a bit over sold.

- OpenAI and anthropic ARR are getting closer to 100 billions than 15 billion, they do not usually book that much compute, join entity like stargate LLC do. They do not tend to commit much, it is conglomerate with the softbank-microsoft-other and themselve that do, to take that Nvidia 250 billion example it is with Softbank not really with openAI.
Using 2025 numbers for the fastest growing big company in world history is highly misleading here.

- Renter with proprietary custom ASIC for its proprietary inference can beat free to download open weight and for that level of needed hardware renting will always be popular, it is not like people will not rent what they need for Deepseek giant models. OpenWeight has been quite good for a while now.... this is not something new, if anything Fable vs them do not show a closing gap, but a potential opposite, the moment they achieve to close distillation of them by third party.
 
AceGoober said:
It isn't AI failing that I'm most worried about. It is the students use of AI in assignments without checking the results prior to submitting said assignments. Taking away or limiting students ability to think for themselves is a travesty.

A Mississippi professor ran a test to see if anyone was using AI to for their assignments. The results are unsurprising.

https://www.foxnews.com/media/missi...als-hidden-way-expose-ai-cheating-viral-video
Click to expand...
It's even more fun when it's the professor using AI to write an exam without checking the results prior to administering it to his/her students. I've seen that firsthand now twice, and have heard of it secondhand a couple times, too. (I am a university professor in STEM.)

I'm as skeptical of AI as anyone, and generally consider it a fancy plagiarism machine, but it is a valuable tool for many workloads. I built my career as much to escape routinized scut work as to do interesting (and occasionally even useful!) work, but I really, really wish I had access to AI tools 20 years ago.
 
You must log in or register to reply here.
Back
Top