Replay 2000 - 2002 market

[btresist]

@Kingran

I am not sure why you keep saying AI revenue doesn’t justify the investments. $NVDA is having steller revenue growth, OpenAI, in just 2 years achieved $13B revenue run rate and has 800 million Monthly Active users. If those numbers don’t impress you, I can understand. But they are some of the fastest growth the world has ever seen.

Nvidia makes software and hardware used for AI. Why is it surprising that they are having stellar revenue growth during an AI bubble? The question is what happens to NVDA when the bubble bursts? Cisco made hardware for the dot.com industry and was also flying high until that bubble burst. 25 years later it still hasn’t reached its 1999-2000 highs. Deja vu all over again?

OpenAi in just two years has achieved an operating of loss of $7.8B during that $13B revenue run period. That’s burning a lot of cash with no clear prospects for profitability any time soon.

@Kingran

DO you have any numbers behind your assumptions??

MIT (a group you would think would understand AI) found 95% of AI projects fail. I think 95% is a pretty significant failure rate. https://complexdiscovery.com/why-95-of-corporate-ai-projects-fail-lessons-from-mits-2025-study/

I think a big part of the problem is data. We have lots of data, but most of it is not AI-ready data, which has to be of high quality and accurate. AI is only as good as the data it is trained on. The more uncertainty and variance, the less useful it is. Data Scarcity: When Will AI Hit a Wall?

In other words, a data infrastructure has to be built in order for AI to optimize its value. It’s like cars before the national highway system. Until that happens, AI is definitely in a bubble. But who knows how long that will take.

In the meantime, Walmart and Amazon will primarily use AI to improve e-commerce, hoping for incremental gains in sales by suggesting products. That’s great, but hardly revolutionary.

[bjurasz]

@btresist

Nvidia makes software and hardware used for AI. Why is it surprising that they are having stellar revenue growth during an AI bubble? The question is what happens to NVDA when the bubble bursts?

We don’t have to go back very far in time. I remember it well. I’ve been with Nvidia just over three years. When I hired in my RSU’s were priced at $18 per share (split-adjusted). Within 6 months we lost half our value, I believe due to the crypto bust. It recovered, and then some, within another 6 to 9 months. (a friend hired in 6 months before, at a higher strike price, what lousy timing!). But that was a fast, rapid decline.

Crypto was a big catalyst for us, then it burst. We were lucky to have another catalyst come along.

[Kingran]

@btresist

OpenAi in just two years has achieved an operating of loss of $7.8B during that $13B revenue run period.

I see in 2 year of ChatGPT launch they are doing $13 B run rate, and you are seeing $7.8 B operating loss… that’s what makes market. I guess there is nothing more to talk here…

[dividends20]

When people talk about “market”, this means they are clueless and have no idea.
They are scared, go into cash and think they can time the movements. They cannot.

On the other hand long term investors invest in specific companies or DCA in S&P 500 index. They sleep well, get rich slowly and enjoy life.

Ignore the noise.

[Kingran]

@btresist

MIT (a group you would think would understand AI) found 95% of AI projects fail.

Frankly, I would throw out that report into trash can. MIT did 52 interviews and based on that they prepared that report… say for example, Citibank has 100’s of AI projects in pilot, POC, being developed, etc. Do they expect all 100 is going to production?? No. On the other hand, they have already implemented many AI projects, where they are seeing significant gains. Even if 10% of their projects succeed, their ROI is great.

@btresist

We have lots of data, but most of it is not AI-ready data, which has to be of high quality and accurate.

The days of labeled data is required for training is over, that is so 2022. The current models have the ability to learn from RAW data. Simple example, I have only yesterday to analyze a company uploaded last 4 conference calls as PDF’s and then their analyst package as excel sheet, and my notes from the past conference calls in word document and link to google docs, and the LLM is able to process all of them and prepare the report I asked in the structure I asked.

@btresist

In the meantime, Walmart and Amazon will primarily use AI to improve e-commerce, hoping for incremental gains in sales by suggesting products.

I have already talked about one use case at Wal-Mart, where using AI for treasury management, definitely product suggestions will happen, but that is already happening today. If you think that is what $WMT will use AI for, then “you know nothing, Jon Snow” !!! Some of the projects at $WMT is supply chain simplification, order processing, automatic ordering, document reviews, invoice processing, fraud detection, … in each of these projects they have 100’s of people working.

$WMT has one of the leading edge implementation of Blockchain for traceability, today, they can identify a product all the way from source, to the store to the consumer. When a “produce” recall, they can trace what used to take 7 days they can do it now in seconds. Why I am mentioning it is, most of us don’t understand or underestimate how pervasive technology implementation in a big box retailer like $WMT. $WMT understands technology deeply.

When you make a statement that they will just use it for product recommendation, that shows you are deeply under estimating how pervasive they think the technology will become for them.

[JimA759s]

@Kingran

When a “produce” recall, they can trace what used to take 7 days they can do it now in seconds.

I’m a bit confused - any reasonably coded database would be able to do that in seconds and many years ago. Do you really need AI to do this?

JimA

[albaby1]

@Kingran

That’s your thinking. The capacity build is not really a bubble.

That wasn’t my point.

If we’re not in a bubble, then obviously the software guys will be fine. But if we were in a bubble, I don’t see how the software guys would be able to avoid the blowback.

It doesn’t matter that a big chunk of this is taking place within big public companies rather than smaller, newer firms getting IPO’s. If it turns out that AI isn’t going to actually provide an earthshaking return on investment, then those firms are going to stop funding their AI projects - and that’s going to be brutal for the software folks just as much as the chipmakers.

For example, when Meta finally realized that they needed to stop trying to make “Metaverse” happen, they stopped throwing all that money into the “Metaverse” furnace (though didn’t bother renaming the company) - which ended up causing a bloodbath for the software folks that were working on that project. If AAPL/GOOG/MSFT were to come to a similar conclusion about AI, and those firms stopped shoving hundreds of billions of dollars at the AI project, it would not just be bad for engineers working at NVDA - it would be also brutal for all the software folks at those companies that are involved in AI research (although some of them are getting big enough bonuses upfront that they’ve already got their generational wealth, so weep not for at least those folks).

So that’s what I don’t understand. The industry wide project of AI is sucking up all these resources. If that ends up being economically justified, then obviously that will all work out fine. But I don’t understand why you intimated the software side might be able to avoid as much of the downside if it ends up being a bubble after all.

[Kingran]

@JimA759s

I’m a bit confused - any reasonably coded database would be able to do that in seconds and many years ago

No, not really. They are not using AI, they are using blockchain, for complete traceability.

Traceability was driven by FDA’ Food Safety Modernization Act regulation. For someone like Kraft, the complexities involve, ingredient complexity, supply chain fragmentation, inventory management. Many retailers have implemented for direct products like Palm Oil, but still struggling with complex products, where they are focusing on high value items, or items that have high exposure like e-coli.

For ex: Costco has traceability for many products, not for all products.

[Kingran]

@albaby1

If we’re not in a bubble, then obviously the software guys will be fine.

We are talking two different things. What is a capacity overbuild? Lots of retail space was build, and e-commerce picked up, resulting in excess capacity of the malls, strip malls, etc. They owners of the property cannot lease their properties to the capacity, when they can lease the rent is low, resulting property valuation going down, and unable to service the debt. Eventually bankruptcies follow, and the market adjusts, this is capacity overbuild.

The stock market valuation bubble is, $PLTR market cap is 100x of revenue… when AI slows down, the stock price will crash, but $PLTR may continue to enjoy double-digit growth in revenue, and profits. This is excess valuation coming down. The share price will not have any impact for $PLTR’s ability to continue to innovate, develop new products, increase revenue, profits. Even the early investors who bought the stock in $10’s, $20’s are going to be fine, but those who bought in this later stage will suffer. There are many index buyers, who while thinking they are not owning $PLTR are owning indirectly and actually buying at very high price.

So when AI bubble breaks, reality sets in, the valuation of these companies will come down, not necessarily demand for their product, their ability to continue to invest.

These are two different things. That’s why I said software/ Application developers may be fine. But other infrastructure players like $NVDA may see significant revenue erosion, because the demand for their new chips will come down, other infrastructure players may suffer.

So, stock market valuation correcting is very different from AI capacity build.

[albaby1]

@Kingran

The share price will not have any impact for $PLTR’s ability to continue to innovate, develop new products, increase revenue, profits.

Why wouldn’t it, though?

Again, if we’re in an AI bubble and eventually all these firms come to realize that AI is not going to provide anywhere near sufficient economic return to justify all the effort being put into it, not only would Palantir’s stock price fall, but their business would change. Right now they’re devoting a lot of resources into developing AI products, expecting that there will be a really big profitable market for those AI products. If they discover over time that there won’t be a really big profitable market for those AI products, but rather a more modest one, then they’re going to stop devoting as much of their resources to making so many new AI products.

Again, the model is Meta. Meta didn’t fire all those software engineers that had been working on the Metaverse because they suddenly lacked the ability to keep shoveling money at developing the Metaverse. They fired them all because Meta realized that the Metaverse wasn’t going to be a very profitable thing, and decided to stop funding activities in that direction.

If we are in an AI bubble and it bursts, it will be exactly because demand for the end products isn’t as robust as all these firms believes it is. Either because the product ends up costing too much to deliver, or just isn’t as economically valuable to end-users, or perhaps both. How could that not devastate the software folks who are currently commanding massive salaries in order to work on making those products based on the assumption that they’ll be Golden Geese?

[intercst]

@albaby1

If we are in an AI bubble and it bursts, it will be exactly because demand for the end products isn’t as robust as all these firms believes it is. Either because the product ends up costing too much to deliver, or just isn’t as economically valuable to end-users, or perhaps both. How could that not devastate the software folks who are currently commanding massive salaries in order to work on making those products based on the assumption that they’ll be Golden Geese?

I think that’s it’s more likely that a group of uber-wealthly venture capitalists continue to fund each other’s AI firms to maintain and expand current stock market valuations.

Does Private Equity have any laws similar to “check kiting” in banking?

Also, many of the top AI researchers who are being offered 9-figure pay packages by Meta, Google, etc. seem to think that AI is inherently dangerous in the “for-profit” form.

intercst

[dividends20]

OpenAI wraps $6.6 billion share sale at $500 billion valuation

[btresist]

@Kingran

Frankly, I would throw out that report into trash can.

That is a convenient way to deal with stuff you disagree with. However, multiple analysts have come to the same ball park conclusion that most AI generative projects disappoint. https://venturebeat.com/ai/why-do-87-of-data-science-projects-never-make-it-into-production

@Kingran

On the other hand, they have already implemented many AI projects, where they are seeing significant gains. Even if 10% of their projects succeed, their ROI is great.

Show me the money. Lots of companies have run pilot AI projects that if they extrapolate forward will save money. But really, that’s what most of the current hype is about. You make a big deal about WalMart. Walmart began implementing AI methods in about 2018. Here is Walmart’s operating costs during that time period.

Where is the tangible AI impact on cost reduction?

@Kingran

Simple example, I have only yesterday to analyze a company uploaded last 4 conference calls as PDF’s and then their analyst package as excel sheet, and my notes from the past conference calls in word document and link to google docs, and the LLM is able to process all of them and prepare the report I asked in the structure I asked.

I don’t think you understand. A generative AI model can use data as you describe. However, the way you develop that generative AI model is through extensive training using data that is of AI-quality in accuracy and format. Big difference. You should try to understand the field before making big claims about investing in it. AI-quality data is the bottleneck with real negative results. https://www.davydovconsulting.com/post/newer-ai-models-are-trained-on-its-own-ai-data-deliver-lower-quality-result-than-older-ai-models-tra

@Kingran

When you make a statement that they will just use it for product recommendation, that shows you are deeply under estimating how pervasive they think the technology will become for them.

My comments have to do with investing decisions today, not 10 years from now. I believe that AI has been used on a large scale for sales recommendations by Amazon and Walmart and have generated significant increases in sales income. I don’t see similar data for other endeavors. One sees claims that elaborate AI models being developed will (for example) make shipping more efficient or reduce personnel. But left unsaid is whether similar savings could have been done by human consultants being assisted by cheap off-the-shelf AI products for less cost than the current massive AI investment.

@Kingran

I see in 2 year of ChatGPT launch they are doing $13 B run rate, and you are seeing $7.8 B operating loss… that’s what makes market. I guess there is nothing more to talk here…

Those numbers mean despite a $13B revenue run rate ChatGPT still lost $7.8B in operations. Interpretation: they spent a lot more than $13B to make that $13B.

If you believe that is the sign of a good investment perhaps saying nothing more is a good idea.

Amazon also went through a phase where they were also growing revenue rapidly but losing money in operations. However, they had a business model that generated huge positive free cash flow. The latter is why I invested in AMZN at that time, thought it was worth the risk. In contrast, ChatGPT isn’t expected to be cash flow positive until 2029. https://finance.yahoo.com/news/openai-does-not-expect-cash-191503509.html

The survival of OpenAI is dependent on the charity of others, like Nvidia.

[DrBob2]

@bjurasz

I remember it well. I’ve been with Nvidia just over three years. When I hired in my RSU’s were priced at $18 per share (split-adjusted). Within 6 months we lost half our value

I’ve owned NVDA since 2017 and seen three 50% drops. It comes with the territory. My $30K is now worth $1500+K. (I did sell $100K, so even if TSMC is nuked by the ChiComs I will still have come out ahead on NVDA.)

DB2

[dividends20]

@btresist

The survival of OpenAI is dependent on the charity of others, like Nvidia.

hmm, amusing view of OpenAI

[Kingran]

@btresist

Walmart began implementing AI methods in about 2018. Here is Walmart’s operating costs during that time period.

First of all I didn’t say they are investing in AI since 2018. I mentioned about Blockchain implementation as an example of their deep understanding of technology.

The below image, from Morningstar, shows the margins have not eroded, and not increased. $WMT primary operating cost increase is due to higher employee cost and you could see that in 2023. Saying AI is not delivering for $WMT by showing operating expense is disingenuous. For an Investor to discern whether AI has ROI or not by looking at income/ balance sheet is not possible.

@btresist

However, the way you develop that generative AI model is through extensive training using data that is of AI-quality in accuracy and format.

I hear you where you are coming from. I provided an example of RAG and that is how the enterprise going to interface with AI. There are tons of data sitting within enterprise which are not used for the training but still can be used during inference and as they infer the model is also learning.

Now, those who claim AI training on data generated by AI is of lesser quality are not understanding few things. For ex: the real world data has fewer instances of edge cases, and often enterprises need the model to handle the edge cases appropriately and not hallucinate. The synthetic data is very useful in this. There are other areas like specialized model training while retaining data privacy, scaling, etc… but they are too deep technical. The firms who were making money by selling data, or labeling data, are making a big deal of this. The model’s are getting better and not degrading.

@btresist

Amazon and Walmart and have generated significant increases in sales income. I don’t see similar data for other endeavors.

You are not going to find those data outside, except may be in few industry presentations. Typically, either you need to be plugged in through the firms consulting/ working on those efforts or you know someone in WMT who is working on those areas. Recently I spend a week in Bentonville, where I learned $WMT’s effort on AI, we had pretty lengthy discussion like hours, because other topic of discussing Trump policies was getting too dangerous. and he leads their data division.

I am more familiar with AWS efforts on AI.. because I get to hear about that everyday

@btresist

Those numbers mean despite a $13B revenue run rate ChatGPT still lost $7.8B in operations…
Amazon .. had a business model that generated huge positive free cash flow.

OpenAI is investing huge sums, sums that they don’t have, and expect to raise from outside sources. Those funds can dry quickly, if AI is not delivering the growth they are expecting. If and when that happens OpenAI will adopt and live within its means.

But OpenAI will survive. You don’t understand the power of 800 MAU, pretty soon it is going to be 1 Billion.

Thanks for the conversation, it is getting too long. We all have our believes. There is no set rule for making money. Just make money

[btresist]

@Kingran

First of all I didn’t say they are investing in AI since 2018.

Never said you did. Ironically it was Google’s AI that did so. Perhaps one can’t trust the accuracy of AI…?

@Kingran

Saying AI is not delivering for $WMT by showing operating expense is disingenuous.

I’m just suggesting that the evidence out there is more negative or absent than positive. Somewhat surprising for a tech that is presumably revolutionizing all human endeavors.

@Kingran

Now, those who claim AI training on data generated by AI is of lesser quality are not understanding few things.

I’m no expert so that’s not my conclusion. It is the conclusion of guys in the link who presumably are experts. It is also not a minority opinion. Just do a google search on “the AI data problem”.

@Kingran

You are not going to find those data outside, except may be in few industry presentations. Typically, either you need to be plugged in through the firms consulting/ working on those efforts or you know someone in WMT who is working on those areas.

This scarcity of public data is in itself interesting.

@Kingran

But OpenAI will survive. You don’t understand the power of 800 MAU, pretty soon it is going to be 1 Billion.

The 1Billion MAU will survive. Not so sure about OpenAI.

Every new iteration of ChatGPT has increased energy and CPU costs. Increasing costs is the opposite trend of most successful technologies like computers, batteries, solar panels, etc. Makes one wonder.

[Kingran]

@btresist

This scarcity of public data is in itself interesting.

There is no scarcity. Stop saying that. LLM’s have trained on giant human generated data. But, what if you want to avoid human bias in the data, for ex?

@btresist

It is the conclusion of guys in the link who presumably are experts.

Those guys are not experts, not even by a mile… they are just web site developers using wix studio… you call them experts??? This is from their website, they describing themselves.. do you see anywhere AI??? Everyone has an opinion, that doesn’t make them expert.

Our expertise includes:

• Web design and development on platforms like Wix Studio, Wix Velo, and Shopify, focusing on creating visually stunning and highly functional websites.
• Development of complex iOS/Android mobile apps using a versatile tech stack to address unique challenges and requirements.
• SEO, SMM and paid ads strategies that enhance visibility, drive traffic, and boost conversions.
• Branding solutions to craft a distinct identity for your business that resonates with your audience.

Depending on your query, Google search is going product the results. If I query for UFO and aliens I will get 1 million links, how many UFO’s and aliens you have encountered in your life??? Those who are building frontier models are not complaining and folks who complain are not even fringe players, except couple of well-respected AI pioneers who are bit skeptical.

I will stop with this. This is not a serious conversation, rather we are talking our believes here. Thanks for the conversation.

[McLovin1981]

@GDavenport

Now that I’ve used AI and presented it, how many will:

1. Disqualify everything I mention in threads from this point forward?
2. Not bother to read the content - BECAUSE it was generated in partnership with an AI
3. Discount further statement as not relevant or not useful, implying there is no value here (or in AI in general)

Interesting times require an open mind.

4. Conclude you are using AI wrong.

[Leap1]

The first probable AI killer app MIGHT be search. Depending on how ads are offered to the public.

OpenAI is trying to catch up to Google’s ranking system, hardware, and software infrastructure. The company can do that, but won’t own search, not much of it.

adding

OpenAI is to Westinghouse as…

…but AI is not to electricity as…