Logos Stick said:

---

Employees will start doing shadow AI, paying for their own subscriptions. It's like crack. Once you use it to minimize the work you have to do, you're hooked.

---

What if you combined crack AND some AI! The productivity gains would be immeasurable!

And companies don't even know what or where any "finish line" is.

Best thing to do is make some money, save it, and be ready when the bubble go pop.

Over_ed said:

---

Logos Stick said:

---

Employees will start doing shadow AI, paying for their own subscriptions. It's like crack. Once you use it to minimize the work you have to do, you're hooked.

---

Correct, and they will willfully violate company security policies to do it.

---

I know people who will absolutely disregard a company policy if they think it unjust.

Wildmen03 said:

---

So many companies are racing to be first, but they don't know why they are racing or what they are going to do once they get to the finish line.

---

When we get to the finish line, we are going to fight the Terminators.

"Indica News"?

Companies will pay for more AI if it means that they can sell more for cheaper. If you're paying your workforce to then have them prompt ai, then you're duplicating costs to some degree. I can see a use case where AI solves the problem too fast and the company doesn't need it solved that fast. Sorta like overnight shipping - sometimes it's great to pay 60 bucks and my item is there tomorrow. Otherwise I'm good with paying 2.99 and it being there in 2 weeks. Ya know?

At least it's not just Sativa like the Babylon Bee.

Wildmen03 said:

---

So many companies are racing to be first, but they don't know why they are racing or what they are going to do once they get to the finish line.

---

Ian Malcolm has entered the thread.

It is the most addicting app ever created

Mr.Milkshake said:

---

It is the most addicting app ever created

---

This goes beyond AI porn brother.

Username respect though.

Not surprised at all. GitHub Copilot is moving to a usage billing model. That's going to drive up costs, and another problem with AI is unpredictable costs, as well. My company is considering blocking the more expensive (newer) LLMs.

Amazon was measuring people based on token usage (think of it as about how many words go into and out of the model you're using). A bunch of their employees spun up agents (think of something that automatically uses the model) to do worthless things over and over. They literally got people to waste their money on LLMs because they said they were tracking it. Not much useful going on, but plenty of costs.

So when does the dot com crash part II happen?

lobopride said:

---

So when does the dot com crash part II happen?

---

As soon as banks start to realize all of the debt companies spent on NVIDIA GPUs can't be paid.

Companies bought and bought and bought without knowing WHAT they bought... NOR what to do with it.

We ask companies about DPUs... they don't realize they were included in the BOM and don't know what to do with them.

The Gov is going to have to bail out the banks (again) because... like Home Mortgage loans... the bank can only reposses so much of something they can't use.

DallasAg 94 said:

---

lobopride said:

---

So when does the dot com crash part II happen?

---

As soon as banks start to realize all of the debt companies spent on NVIDIA GPUs can't be paid.

Companies bought and bought and bought without knowing WHAT they bought... NOR what to do with it.

We ask companies about DPUs... they don't realize they were included in the BOM and don't know what to do with them.

The Gov is going ping to have to bail out the banks (again) because... like Home Mortgage loans... the bank can only reposses so much of something they can't use.

---

Can you imagine the FPS all those GPUs could get you on a game of CoD or whatever?!?!

lobopride said:

---

So when does the dot com crash part II happen?

---

It's not. Costs are increasing because of a lack of deployable computation, not because people aren't using it. With more computation the costs will fall and the consumption will return. The current tokens are going to individuals that either have the most optimism or the highest and best provable use.

hph6203 said:

---

lobopride said:

---

So when does the dot com crash part II happen?

---

It's not. Costs are increasing because of a lack of deployable computation, not because people aren't using it. With more computation the costs will fall and the consumption will return. The current tokens are going to individuals that either have the most optimism or the highest and best provable use.

---

Dude, can you please translate that word salad into something a normal human can understand?

It's no surprise that companies are suddenly freaking over AI costs. The technology DOES NOT SCALE. Inference is ridiculously expensive, the providers are too busy focusing on training the next model and ignoring the part that is actually going to generate revenue.

Logos Stick said:

---

Employees will start doing shadow AI, paying for their own subscriptions. It's like crack. Once you use it to minimize the work you have to do, you're hooked.

---

You are a fool if you are in a position that uses Claude Code for work not to have your own account.

Things that I would keep on my personal computer and in a notebook for ideas for my own side business now live on my own AI Agent setup. I don't want any unique thought of mine or product idea outside of my work responsibilities being jeopardized because I used their token in development.

BigRobSA said:

---

I believe it.

Also, if corporate America would get smart, they'd cut the massive costs of numerous levels of "management" that have no intrinsic value: like managers, answering to managers, answering to managers, answering to a senior manager , answering to a director(s) , answering to Sr. Director(s), and so on....with several levels of VPs.

That's a **** ton of labor cost that can, and should, be cut before they cut the actual laborers making the product and/or "AI".

---

It's not the labor that's costly. It's the dishonesty.

Every layer adds another dimensions of bull**** to what gets fed to the execs. I honestly have no idea if the people at the top have any idea of what's really happening in the company. In my experience, by the D1 level, everything that got routed upwards was pure optics management.

AI might be useful just for logging every company communication at every level, and giving the CEO a way to ask "who ****ed this up", without having to route around his own management team to really find out. The opacity of middle management being undone would be a wonderful dividend from all this AI disruption.

If something doesn't make sense to you, then maybe don't be a ***** and consider it's coming from a lack of knowledge on your part.

Compute is the restriction. There's a lot of factors in getting a data center built and operational. Currently there is a shortage of compute relative to demand for compute. That caused Anthropic to put a governor on their subscriptions, defaulting their subscriptions to a lesser model and halving the number of tokens delivered per subscription, making those subscriptions less valuable. They have temporarily resolved that issue by contracting with SpaceX to use their unused capacity on Collosus I and Collosus II.

Companies see their expenses per productive output rise and decide to restrict access to individuals that actually know how to use the models productively, rather than some junior individual dicking around and wasting money. Not all token consumption is equal in utility, meaning not all token spend is equal in utility. They are consolidating on people that actually get the most productivity out of it while compute is restricted.

Reality is that what was a bleeding edge model 12 months ago now runs locally on a $10k computer. What was a bleeding edge model 24 months ago runs locally on a $600 computer. So if you're looking at the current model and thinking "that's magic!" In 12 months it is likely to be running on a high end computer, and in 24 months on a computer fit for grandma. For the first time in your lifetime computers are broadly and consistently gaining value, because the models are making them more productive. An H100 does more today than when it was brand new.

Future will see token use optimization that reduce computation per query. Future will see discernment built in to allocate questions to the appropriate model size further reducing token usage/cost of AI. Future will see a circumvention of the permitting process and public commentary of data center builds by launching compute into space.

AI is scalable. It is saturating deliverable supply. It is not like the internet build out where it is speculative value and dark fiber everywhere.

Holy: Anthropic’s ARR has reportedly now surged past $44B, up from $9B at the end of 2025, a nearly 5x jump, or roughly 389% growth, in just a few months.

The growth is driven mainly by enterprise Claude adoption and Claude Code, while inference gross margins allegedly improved… pic.twitter.com/amyXWJokaJ

— Chubby♨️ (@kimmonismus) May 2, 2026

Benioff was on All-In two weeks ago talking about Slack implementing tools that allow a CEO to ask what's really going on at their company. Not a coincidence that most of the layoffs recently are in those middle management positions. Cloudflares CEO was pretty blunt saying that they're laying off the measurers and keeping the doers.

There's some nuance there. That is self reported, non-GAP accounting. Want to look at the real data when they file for their IPO. Real revenues are likely much lower.

Oh and we know that graph is a complete lie because they filed a sworn affidavit with a federal court in February claiming their total revenues in the entire history of their company was only $5 billion.

Good. I hope AI goes the way of the Dodo bird.

ARR is a forward projection based upon past earnings. When you have $9 billion ARR it means you earned ~$2.25 billion in a quarter. Trace those claimed rates backwards and you get roughly $5 billion earnings lifetime by year end 2025 (give/take a couple $100 million).

Their filing stated they have lifetime revenue exceeding $5 billion in March. In March they almost certainly did not have finalized books to make a more specific statement other than what they had validated for 2025 and prior, hence the exceeding qualifier. When you're dealing with courts you don't try to get specific unless asked, because getting specific and being wrong causes legal troubles.

Based on that chart:
Lifetime year end 2024:~$250 million
Q1'25: ????
Q2'25: $1.125 billion
Q3'25: $1.75 billion
Q4'25: $2.25 billion
Total: $5.375 billion

Based on reporting:
Q1'26: $4.8 billion (113% increase QoQ)
Q2'26: $10.9 billion (127% increase QoQ)


It is claimed that Anthropic will show an operating profit in Q2'26 of >$500 million. Will there be bankruptcies/money lost in the AI boom? Absolutely, but if people are crossing their fingers hoping that the frontier model makers collapse under a lack of utility/viability of AI as a business then it's from a lack of understanding of what's happening.

ErnestEndeavor said:

---

There's some nuance there. That is self reported, non-GAP accounting. Want to look at the real data when they file for their IPO. Real revenues are likely much lower.

Oh and we know that graph is a complete lie because they filed a sworn affidavit with a federal court in February claiming their total revenues in the entire history of their company was only $5 billion.

---

GAAP revenue in Q1 2026 was $4.8B. The graph is not a lie; it's a projection that seems very reasonable.

AI is like any other tool. If you use it for what it is good for, it is game changing. If you use 1,000,000 tokens a month to save yourself 5 minutes a month, it is inefficient. Its not free. It has tremendous value when used well. It makes stupid people look lazy and stupider. It is a force multiplier for people who use it intelligently.

This. I'm building out my small teams workflows and reduced our compute 78% over the weekend. Just taking what we've built and optimizing. You have to spend tokens to figure it out. And then make it efficient. Compute costs will go down immensely and fast as well.

BigRobSA said:

---

DallasAg 94 said:

---

lobopride said:

---

So when does the dot com crash part II happen?

---

As soon as banks start to realize all of the debt companies spent on NVIDIA GPUs can't be paid.

Companies bought and bought and bought without knowing WHAT they bought... NOR what to do with it.

We ask companies about DPUs... they don't realize they were included in the BOM and don't know what to do with them.

The Gov is going ping to have to bail out the banks (again) because... like Home Mortgage loans... the bank can only reposses so much of something they can't use.

---

Can you imagine the FPS all those GPUs could get you on a game of CoD or whatever?!?!

---

hahahahhah.

Corporate America has to choose: spend on AI tokens OR spend on humans.

Since I've been in this industry for decades and know how the human mind works, here is how it will play out.

Human labor cost is expensive, we all know that. AI has just come in and the 6-10 AI companies (no one gives a damn about the smaller pretenders) have the monopoly. They need adoption, so they are taking losses to keep prices low. Stupid CEOs compare this artificially low price and fire their staff and make big long-ass tone-deaf announcements on Twitter about how humans are useless and AI is better. Eventually the market is going to catch up and this bait-and-switch won't go on for too long. AI prices will go up as AI companies (Meta, Anthropic, Google etc) need to recoup the zillions they spend on it. Where that does that come from? From their customers. The ones who fired humans and are now slaves of AI. Now don't give me the usual "free market conservative" line that other companies will come in and prices will go down. No, that won't happen as it takes years and years and lots of money to compete in this space. Only someone with the power of another Google can do it.

Bottomline: These companies will end up paying MORE than what they did to humans and with inconsistent results which fall below expectations. Who will they throw under the bus? AI? CEOs are cooked. But these slimy creatures will just shamelessly take their golden parachute and decamp, only to show up somewhere else to ruin another company

Hmm pay $100k to an experienced Steve, or pay $150k to an inconsistent Claude that needs supervision?

Ha ha ha ha ha.

Microsoft Flight Simulator would like to have a word with you.