Something nobody's talking about: OpenAI's losses might actually be attractive to certain investors from a tax perspective. Microsoft and other corporate investors can potentially use their share of OpenAI's operating losses to offset their own taxable income through partnership tax treatment. It's basically a tax-advantaged way to fund R&D - you get the loss deductions now while retaining upside optionality later. This is why the "cash burn = value destruction" framing misses the mark. For the right investor base, $10B in annual losses at OpenAI could be worth $2-3B in tax shields (depending on their bracket and how the structure works). That completely changes the return calculation. The real question isn't "can OpenAI justify its valuation" but rather "what's the blended tax rate of its investor base?" If you're sitting on a pile of profitable cloud revenue like Microsoft, suddenly OpenAI's burn rate starts looking like a pretty efficient way to minimize your tax bill while getting a free option on the AI leader. This also explains why big tech is so eager to invest at nosebleed valuations. They're not just betting on AI upside, they're getting immediate tax benefits that de-risk the whole thing.

There is a pretty big moat for Google: extreme amounts of video data on their existing services and absolutely no dependence on Nvidia and it's 90% margin.

> AI is a world-changing technology

As stated in TFA, this simply has not been demonstrated , nor are there any artifacts of proof. It's reasonable to suspect that there is no special apparatus behind the curtain in this Oz.

From TFA: "One vc [sic] says discussion of cash burn is taboo at the firm, even though leaked figures suggest it will incinerate more than $115bn by 2030."

Google’s moat:

Try “@gmail” in Gemini

Google’s surface area to apply AI is larger than any other company’s. And they have arguably the best multimodal model and indisputably the best flash model?

> AI is going to be a highly-competitive, extremely capital-intensive commodity market

It already is. In terms of competition, I don't think we've seen any groundbreaking new research or architecture since the introduction of inference time compute ("thinking") in late 2024/early 2025 circa GPT-o4.

The majority of the cost/innovation now is training this 1-2 year old technology on increasingly large amounts of content, and developing more hardware capable of running these larger models at more scale. I think it's fair to say the majority of capital is now being dumped into hardware, whether that's HBM and research related to that, or increasingly powerful GPUs and TPUs.

But these components are applicable to a lot of other places other than AI, and I think we'll probably stumble across some manufacturing techniques or physics discoveries that will have a positive impact on other industries.

> that ends up in a race to the bottom competing on cost and efficiency of delivering

One could say that the introduction of the personal computer became a "race to the bottom." But it was only the start of the dot-com bubble era, a bubble that brought about a lot of beneficial market expansion.

> models that have all reached the same asymptotic performance in the sense of intelligence, reasoning, etc.

I definitely agree with the asymptotic performance. But I think the more exciting fact is that we can probably expect LLMs to get a LOT cheaper in the next few years as the current investments in hardware begin to pay off, and I think it's safe to assume that in 5-10 years, most entry-level laptops will be able to manage a local 30B sized model while still being capable of multitasking. As it gets cheaper, more applications for it become more practical.

---

Regarding OpenAI, I think it definitely stands in a somewhat precarious spot, since basically the majority of its valuation is justified by nothing less than expectations of future profit. Unlike Google, which was profitable before the introduction of Gemini, AI startups need to establish profitability still. I think although initial expectations were for B2C models for these AI companies, most of the ones that survive will do so by pivoting to a B2B structure. I think it's fair to say that most businesses are more inclined to spend money chasing AI than individuals, and that'll lead to an increase in AI consulting type firms.

Anthropic is building moat around theirs models with claude code, Agent SDK, containers, programmatic tool use, tool search, skills and more. Once you fully integrate you will not switch. Also being capital intensive is a form of moat.

I think we will end up with market similar to cloud computing. Few big players with great margins creating cartel.

Like railroads, internet, electricity, aviation or car industries before: they've all been indeed the future, and they all peaked (in relative terms), at the very early stages of these industries future.

And among them the overwhelming majority of companies in the sectors died. Out of the 2000ish car-related companies that existed in 1925 only 3 survived to today. And none of those 3 ended up a particularly good long term investment.

The railroads provided something of enduring value. They did something materially better than previous competitors (horsecarts and canals) could. Even today, nothing beats freight rail for efficient, cheap modest-speed movement of goods.

If we consider "AI" to be the current LLM and ImageGen bubble, I'm not sure we can say that.

We were all wowed that we could write a brief prompt and get 5,000 lines of React code or an anatomically questionable deepfake of Legally Distinct Chris Hemsworth dancing in a tutu. But once we got past the initial wow, we had to look at the finished product and it's usually not that great. AI as a research tool will spit back complete garbage with a straight face. AI images/video require a lot of manual cleanup to hold up to anything but the most transient scrutiny. AI text has such distinct tones that it's become a joke. AI code isn't better than good human-developed code and is prone to its own unique fault patterns.

It can deliver a lot of mediocrity in a hurry, but how much of that do we really need? I'd hope some of the post-bubble reckoning comes in the form of "if we don't have AI to do it (vendor failures or pricing-to-actual-cost makes it unaffordable), did we really need it in the first place?" I don't need 25 chatbots summarizing things I already read or pleading to "help with my writing" when I know what I want to say.

I, personally, use chatGPT for search more than I do Google these days. It, more often than not, gives me more exact results based on what I'm looking for and it produces links I can visit to get more information. I think this is where their competitive advantage lies if they can figure out how to monetize that.

This will remain the case until we have another transformer-level leap in ML technology. I don’t expect such an advancement to be openly published when it is discovered.

>That doesn't mean AI is going to go away, or that it won't change the world - railroads are still here and they did change the world - but from a venture investment perspective, get ready for a massive downturn.

I don't know why people always imply that "the bubble will burst" means that "literally all Ai will die out and nothing will remain that is of use". The Dotcom bubble didn't kill the internet. But it was a bubble and it burst nonetheless, with ramifications that spanned decades.

All it really means when you believe a bubble will pop is "this asset is over-valued and it will soon, rapidly deflate in value to something more sustainable" . And that's a good thing long term, despite the rampant destruction such a crash will cause for the next few years.

> The simple evidence for this is that everyone who has invested the same resources in AI has produced roughly the same result. OpenAI, Anthropic, Google, Meta, Deepseek, etc. There's no evidence of a technological moat or a competitive advantage in any of these companies.

I think this is analysis is too surface level. We are seeing Google Gemini pull away in terms of image generation, and their access to billions of organic user images gives them a huge moat. And in terms of training data, Google also has a huge advantage there.

The moat is the training data, capital investment, and simply having a better AI that others cannot recreate.

I don't see how Google doesn't win this thing.

> AI is a world-changing technology, just like the railroads were

This comparison keeps popping up, and I think it's misleading. The pace of technology uptake is completely different from that of railroads: the user base of ChatGPT alone went from 0 to 200 million in nine months, and it's now- after just three years- around 900 million users on a weekly basis. Even if you think that railroads and AI are equally impactful (I don't, I think AI will be far more impactful) the rapidity with which investments can turn into revenue and profit makes the situation entirely different from an investor's point of view.

> The simple evidence for this is that everyone who has invested the same resources in AI has produced roughly the same result.

I think this conflates together a lot of different types of AI investment - the application layer vs the model layer vs the cloud layer vs the chip layer.

It's entirely possible that it's hard to generate an economic profit at the model layer, but that doesn't mean that there can't be great returns from the other layers (and a lot of VC money is focused on the application layer).

This is different because now the cats out of the bag: AI is big money!

I don't expect AGI or Super intelligence to take that long but I do think it'll happen in private labs now. There's an AI business model (pay per token) that folks can use also.

I think we'll find that that asymptote only holds for cases where the end user is not really an active participant in creating the next model:

- take your data

- make a model

- sell it back to you

Eventually all of the available data will have been squeezed for all it's worth the only way to differentiate oneself as an AI company will be to propel your users to new heights so that there's new stuff to learn. That growth will be slower, but I think it'll bear more meaningful fruit.

I'm not sure if today's investors are patient enough to see us through to that phase in any kind of a controlled manner, so I expect a bumpy ride in the interim.

Your premise is wrong in a very important way.

The cost of entry is far beyond extraordinary. You're acting like anybody can gain entry, when the exact opposite is the case. The door is closing right now. Just try to compete with OpenAI, let's see you calculate the price of attempting it. Scale it to 300, 500, 800 million users.

Why aren't there a dozen more Anthropics, given the valuation in question (and potential IPO)? Because it'll cost you tens of billions of dollars just to try to keep up. Nobody will give you that money. You can't get the GPUs, you can't get the engineers, you can't get the dollars, you can't build the datacenters. Hell, you can't even get the RAM these days, nor can you afford it.

Google & Co are capturing the market and will monetize it with advertising. They will generate trillions of dollars in revenue over the coming 10-15 years by doing so.

The barrier to entry is the same one that exists in search: it'll cost you well over one hundred billion dollars to try to be in the game at the level that Gemini will be at circa 2026-2027, for just five years.

Please, inform me of where you plan to get that one hundred billion dollars just to try to keep up. Even Anthropic is going to struggle to stay in the competition when the music (funding bubble) stops.

There are maybe a dozen or so companies in existence that can realistically try to compete with the likes of Gemini or GPT.

I like to tell people that all the AI stuff happening right now is capitalism actually working as intended for once. People competing on features and price where we arent yet in a monopoly/duopoly situation yet. Will it eventually go rotten? Probably — but it's nice that right now for the first time in a while it feels like companies are actually competing for my dollar.

very few software has commoditized, doubt it will be the fate of AI tech stack.

Have you thought about what happens if we get a new improvement in model architecture like transformers that grows the compute needs even further

The "Railway Bubble" analogy is spot on.

As a loan officer in Japan who remembers the 1989 bubble, I see the same pattern. In the traditional "Shinise" world I work with, Cash is Oxygen. You hoard it to survive the inevitable crash. For OpenAI, Cash is Rocket Fuel. They are burning it all to reach "escape velocity" (AGI) before gravity kicks in.

In 1989, we also bet that land prices would outrun gravity forever. But usually, Physics (and Debt) wins in the end. When the railway bubble bursts, only those with "Oxygen" will survive.

Or the airlines. Airlines have created a huge amount of economic value that has mostly been captured by other entities.

Um meta didn't achieve the same results yet. And does it matter if they can all achieve the same results if they all manage high enough payoffs? I think subscription based income is only the beginning. Next stage is AI-based subcompanies encroaching on other industries (e.g. deepmind's drug company)

This is so obviously right.

I may add that investors are mostly US-centric, and so will the bubble-bursting chaos that ensues.

Just in time for a Government guaranteed backstop.

Massive upfront costs and second place is just first loser. It’s like building fabs but your product is infinitely copyable. Seems pretty rough.

Deepseek has invested the same amount as OpenAI?

I’m waiting to get an RTX 5090 on the cheap.

Also that open source models are just months behind

Eh, I wouldn't be so sure, chips with brain matter and or light are on its way and or quantum chips, one of those or even a combination will give AI a gigantic boost in performance. Finally replacing a lot more humans and whoever implements it first will rule the world.

for me its clear OpenAI and Anthropic have a lead. I dont buy Gemini 3 being good. it isnt. whatever the benchmark said. same for meta and deepseek.

People seem to have the assumption that OpenAI and Anthropic dying would be synonymous with AI dying, and that's not the case. OpenAI and Anthropic spent a lot of capital on important research, and if the shareholders and equity markets cannot learn to value and respect that and instead let these companies die, new companies will be formed with the same tech, possibly by the same general group of people, thrive, and conveniently leave out the said shareholders.

Google was built on the shoulders of a lot of infrastructure tech developed by former search engine giants. Unfortunately the equity markets decided to devalue those giants instead of applaud them for their contributions to society.

If performance indeed asymptotes, and if we are not at the end of silicon scaling or decreasing cost of compute, then it will eventually be possible to run the very best models at home on reasonably priced hardware.

Eventually the curves cross. Eventually the computer you can get for, say, $2000, becomes able to run the best models in existence.

The only way this doesn’t happen is if models do not asymptote or if computers stop getting cheaper per unit compute and storage.

This wouldn’t mean everyone would actually do this. Only sophisticated or privacy conscious people would. But what it would mean is that AI is cheap and commodity and there is no moat in just making or running models or in owning the best infrastructure for them.

AI is capital intensive because autodiff kinda sucks.

I still don't understand how it's world-changing apart from considerably degrading the internet. It's laughable to compare it to railroads.

This is why I think China will win the AI race. As once it becomes a commodity no other country is capable of bringing down manufacturing and energy costs the way China is today. I am also rooting for them to get on parity with node size for chips for the same reason as they can crash the prices PC hardware.

"AI is going to be a highly-competitive" - In what way?

It is not a railroad and the railroads did not explode in a bubble (OK a few early engines did explode but that is engineering). I think LLM driven investments in massive DCs is ill advised.

Did railroads change the world though?

They only lasted a couple of decades as the main transportation method. I'd say the internal combustion engine was a lot more transformative.

  > There's no evidence of a technological moat or a competitive advantage in any of these companies.

I don't use my Google Pro subscription much. Gemini 3.0 Pro spends 1/10th of the time thinking compared to GPT 5.2 Thinking and outputs a worse answer or ignores my prompt. Similar story with Deepseek.

The public benchmarks tell a different story which is where I believe the sentiment online comes from, but I am going to trust my experience, because my experience can't be benchmaxxed.