Crazy money spent, 100s of billions.

I dont think Ive seen a good discussion on how its monetized, just ever increasing stock prices over the spend.


Anyone got an answer? What am I missing?

Three possibilities from measuring AI roi in my opinion

1. AI isn't ready, the cost benefit curve with energy doesn't warrant the investment, etc. In each of these cases, across publicly traded companies you might see an uptick in opex but little else. This will lead to a very large market correction

2. AI roi converges around automation. Across public companies, we see a reduction in SG&A. Corresponding uptick in unemployment. I think this still leads to a market correction. I would remain worried about market fundamentals because I think pricing models for token consumption today are unlikely to remain consistent or even within an order of magnitude from a pricing perspective so headcount reduction today creates more problems in the future. I hate the message around AI replaces 90% of jobs. I think the people stating that I have a very dim view of humanity.

3. AI continues to evolve around domain specific architecture and models. This has the effect of increasing organizational maturity. Across any business function it's not just that people are moving faster. It's that they are more effective because AI is delivering insights from deep pattern recognition that low to mid-level performers would otherwise miss. This will reflect all across the income statement and balance sheet, but most notably on the top line. All else held constant, if we see top line growth over the next 12 months while sg&a/cogs remain controlled, that's your ROI for AI. same story on the balance sheet with things like your cash conversion cycle and inventory levels.

It will eat market share starting with software and expand into physical world. Where it doesn't replace existing products, those products will use it. Run that TAM

That's because the AI companies really don't know either. All they care about right now is getting funding.

Altman started with retail subs and realized that isn't going to cut it. Announced earlier this year they wouldn't be profitable until 2030 at the earliest.

They have recently cut bait on new ideas and are now switching focus more to corporate because that's where the money seems to be after watching Anthropic jump ahead of them on the profitability curve by focusing on corporate from the get-go.

But now corporate is putting the brakes on their internal users in the last 30 days due to outrageous costs to even participate in it.

So, at this point, I think we are back to not knowing.

Sounds like its possible the human cost vs AI cost could still be prohibitive.

Maybe it starts slower and more niche than the idea that we're all being replaced in 12 to 18 months.


It may be true that we could all be replaced, just not in a cost effective manner

The ROI for the business owners come when they sell the business.

That is, the venture capitalists who invested in AI companies will take them public, and sell their ownership stake in the business for a profit.

They really don't care about the underlying cash flow or profitability of the business. They just give them enough capital to cover payroll and operating costs as long as needed to get the company valuation to its peak, and then they go public and profit right before the house of cards collapses.

GeorgiAg said:

---

Discussion on Bloomberg today about this. When these go public, there will be more pressure for ROI. The analyst really liked Google and Microsoft rather than Anthropic, OpenAI, etc.. because the former companies have other cash cows and can outlast the others for ROI from AI.

---

Valid and interesting though because MS Copilot has been a disaster. They've only had a 3% uptake rate in their existing MS user base from article I read the other day.

That's almost sad.

I'm really heavy in Google.

GeorgiAg said:

---

I'm really heavy in Google.

---

Until proven otherwise, I continue to think Google will lead the consumer/retail space, while Anthropic will remain the best of breed in the corporate/industrial market. This is from casual conversations within the industry (customer viewpoints, primarily), and these vendors' relentless primary focus on their intended market.

Thoughts?

I bleed maroon said:

---

GeorgiAg said:

---

I'm really heavy in Google.

---

Until proven otherwise, I continue to think Google will lead the consumer/retail space, while Anthropic will remain the best of breed in the corporate/industrial market. This is from casual conversations within the industry (customer viewpoints, primarily), and these vendors' relentless primary focus on their intended market.

Thoughts?

---

I love Anthropic as well. I use it heavily at work.

YouBet said:

---

GeorgiAg said:

---

Discussion on Bloomberg today about this. When these go public, there will be more pressure for ROI. The analyst really liked Google and Microsoft rather than Anthropic, OpenAI, etc.. because the former companies have other cash cows and can outlast the others for ROI from AI.

---

Valid and interesting though because MS Copilot has been a disaster. They've only had a 3% uptake rate in their existing MS user base from article I read the other day.

That's almost sad.

---

They'll add it as an upgrade and raise prices. Can claim 100% adoption that way.

techno-ag said:

---

YouBet said:

---

GeorgiAg said:

---

Discussion on Bloomberg today about this. When these go public, there will be more pressure for ROI. The analyst really liked Google and Microsoft rather than Anthropic, OpenAI, etc.. because the former companies have other cash cows and can outlast the others for ROI from AI.

---

Valid and interesting though because MS Copilot has been a disaster. They've only had a 3% uptake rate in their existing MS user base from article I read the other day.

That's almost sad.

---

They'll add it as an upgrade and raise prices. Can claim 100% adoption that way.

---

Microsoft updates are like a game of Where's Waldo. We use to right click, then click....Now we right click and a less useful menu pops up, eventually you find Waldo.

The AI-native startups are the ones to keep an eye on... A lot of the major players are offering a platform with more than just AI capabilities to protect them from a reset on AI valuations...

At some point you just ask AI that question. Ask it to monetize itself.

These frontier models are going to have an almost impossible task monetizing. No one truly trust them and any business that devolops on them will get any true innovation stolen and rolled into the model. They are monitoring everything even Anthropic's ZDR (Zero Data Rentention) clearly states it. The Open Source models are getting just as good GLM 5.2 scores better than Opus 4.8 on many metrics. The cost is 1/10 the price and honestly most companies should be hosting these models themselves for true privacy. Once this shift happens it is over for the frontier models.

Most AI offerings are shifting to a "retail salesperson" experience.

We put specific data related to a problem or task and have the AI recommend a course of action.

In HR: Put in a list of candidates resumes and LinkedIn profiles and cross reference against the job opening and skills you want, the AI makes a suggestion.

In Ag: Put in weather data and crop scouting notes and the AI recommends products to spray or apply.

None of these models have any special analysis going on other than those set by the humans controlling it. The task for AI companies monetizing it revolves around creating a narrative that the AI doesn't miss details and can improve user quality of employment. Convincing CEOs that it makes their employees more productive while the actual employees do the same work with less mental effort making your workforce dumber/lazier.

The secret to "AI" is when a physical process can be eliminated or automated. Thats when things will get real. A fully automated tractor that doesn't require babysitting will change a farmer's life more than a million more datapoints and dashboards.

fightingfarmer09 said:

---

The secret to "AI" is when a physical process can be eliminated or automated. Thats when things will get real. A fully automated tractor that doesn't require babysitting will change a farmer's life more than a million more datapoints and dashboards.

---

Agreed. But you wrote this as if this hasn't started happening. It's started and things are getting real.

Like much innovative technology, it's difficult to tell what will actually win in a money-making fashion. However, I suspect the big money in AI will be made by using AI to create new "things" that don't currently exist.

Example #1: Merk uses AI to create and manufacture new organic chemistry that cures cancer. Trillions of dollars made by Merk on cancer cure. How much does AI get "paid" for this? Difficult to quantify, but gets credit for much of the profit of new cancer drug.

Example #2: Dupont uses AI to create new pyrolysis process that really does turn plastic back into hydrocarbons cheaply. Dupont makes trillions of dollars on their new pyrolysis plants.

Example #3: Chinese AI develops something that can de-encrypt all secure data encryption used for your direct deposit paycheck, so Oracle uses AI (and possible quantum computing) to create a more secure encryption that the Chinese AI system can't crack.

Rinse and repeat.

Can AI play linebacker?

We rolled out Copilot within Teams a while back and I found it rather useless. But, last week a coworker showed me an AI bot the company has built with Copilot that is being trained on all emails and trades around a certain asset, and it was pretty impressive. So my judgement of Copilot is definitely evolving

chris1515 said:

---

We rolled out Copilot within Teams a while back and I found it rather useless. But, last week a coworker showed me an AI bot the company has built with Copilot that is being trained on all emails and trades around a certain asset, and it was pretty impressive. So my judgement of Copilot is definitely evolving

---

Copilot is fine you just have to know what it is and isn't good at. It's definitely not good at deep reasoning and decision making. And don't rely on it for quantitative analysis. It's very poor at that. Also know there are different levels to Copilot.

It's agentic AI is good at smaller scale automations and summarizing things. I utilize it on calls when I can. Does a great job summarizing meetings and taking actions down then you can set up an agent to sit on top of those things and remind you of what you need to do or what others owe you. That's been nice considering how many actions seem to get naturally lost.

Principal Uncertainty said:

---

Like much innovative technology, it's difficult to tell what will actually win in a money-making fashion. However, I suspect the big money in AI will be made by using AI to create new "things" that don't currently exist.

Example #1: Merk uses AI to create and manufacture new organic chemistry that cures cancer. Trillions of dollars made by Merk on cancer cure. How much does AI get "paid" for this? Difficult to quantify, but gets credit for much of the profit of new cancer drug.

Example #2: Dupont uses AI to create new pyrolysis process that really does turn plastic back into hydrocarbons cheaply. Dupont makes trillions of dollars on their new pyrolysis plants.

Example #3: Chinese AI develops something that can de-encrypt all secure data encryption used for your direct deposit paycheck, so Oracle uses AI (and possible quantum computing) to create a more secure encryption that the Chinese AI system can't crack.

Rinse and repeat.

---

This. I've said for a long time there are no new ideas but AI is changing that. I have a client I can't discuss quite yet that's doing exactly this type of project. Literally everyone we talk to reacts with a "wait, you can do what?" expression. From potential consumers to corporate to high level government officials everyone is watching him closely right now for a product that isn't possible without AI.

As to the frontier models that's a great question. There's too many of them right now. Betting on Google and MS is betting one of them will get good at it or acquire one of those that is. I'd LOVE for MS to acquire Anthropic because Claude is significantly better than Copilot for what my company uses it for. I'm really not sure where this all lands for those entities. I'm not convinced any of the big software companies will get good on their own. Apple is objectively terrible at AI to the point where I could see them having to acquire someone just to remain relevant. Their walled garden approach may be a problem here. I used Claude daily to manage my email and company tasks which is something you genuinely can't do on Apple. Right now I feel like their "slow but reliable" approach is doing more harm than good. We'll see if that plays out in the long run.