Why is Nvidia so valuable?

In My Humble Opinion
41

Yeah personally I still find the AI technology very impressive but ultimately it has to deliver return on investment on a large scale to a lot of companies to justify AI companies investing perhaps hundreds of billions of dollars which in turn is what is generating the incredible rise of nVidia’s revenue, profits and stock price. And it’s really not clear that companies are finding that much return from their AI spending as yet.

In the medium term nVidia also faces a significant threat of their biggest customers like Google and Amazon backward integrating into designing their own AI chips something they are already doing to some extent and also the threat of Chinese chip manufacturers over the next 5-10 years.

And none of these threats have to be fatal. The markets are pricing a lot of growth in NVidia’s stock and it would only take a moderate slowdown in growth to cause a big correction in the price.

42

All fair points.

In their defense, Nvidia isn’t really strictly an “AI company”. They don’t just sell AI chips, but “big parallel number crunching chips” that can be used for everything from gaming to crypto to, yes, AI. Their next horizon project is robotics: NVIDIA Jetson Thor Unlocks Real-Time Reasoning for General Robotics and Physical AI | NVIDIA Blog

In their in-house R&D, they use GPUs to create virtual worlds and scenarios (warehouses, etc.) to run millions of ML iterations to pre-train robots in those virtual worlds to learn their physics, maneuvering, etc., all in software, to give the physical robots an edge before they’re even “born”. That’s just stuff they’ve publicly announced. And the consumer product they announced can help run on-device ML vision etc. models.

I don’t own any Nvidia stock, so I’m not trying to hype them up or anything… they’ve just been around for quite a while and have successfully out-competed many peers in several fields, not just AI. That takes more than fast chips and good timing. Not many folks expected Intel (or IBM, for that matter, who’s now trying to make a comeback) to become irrelevant and Nvidia to become so dominant… it’s a ruthless industry, and Nvidia had to take many big risks along the way.

None of that disproves that their stock is immensely over-valued right now because of the crazy bubble. It’s just that I hope Nvidia the company survives even after an AI crash or two… if only because GeForce Now is so life-changing.

43

Their sales, revenue, and profits have increased more than 50% in just the last few months alone: https://www.nytimes.com/2025/08/27/technology/nvidia-earnings-ai-chips.html

But their stock has decreased 2%.

44

I haven’t read the entire thread so perhaps I am being redundant, but the reason is simple. The market believes that Nvidia has a lot of rich customers (AI companies) who will pay high prices for the products. So they expect Nvidia to charge a premium and make lots of money. Since the barriers to entry in this business are so high, the market thinks the high prices will continue for a while. Hence, high stock prices.

45

But what about the carpet-pissers?

Stranger

46

I think the market reaction to the latest quarterly statement shows how insanely high the expectations for Nvidia are. 50+% growth for a company that large is sensational and yet the initial market reaction was negative.

47

What’s even more interesting is the apparent concentration of their customers. While it’s just one quarter, this article highlights this. It also comments on some lack of transparency in their financial reporting.

48

Customer A” made up 23% of total revenue, and “Customer B” comprised 16% of total revenue, according to the company’s second-quarter filing with the Securities and Exchange Commission.

Wow, that must be one hell of a gaming PC they’re building.

49

Or maybe the farmers are using them to laser weedwhack…

50

The first step in AI is to match the number crunching capability of the human brain. The Intel X86 CPU architecture has been the basis for PC’s since the 1980’s. It has been tweaked and enhanced for the last 50 years or so. It has so much baggage after all that time that it is just not up to the task of brain emulation. NVDA has a CPU that can do it.

The next step in AI is to raise a newborn baby that is different from everybody else in a society that does not really like it. Once it becomes a responsible adult, you have artificial intelligence.

The outlandish power requirements we see these days show that the number crunching capability is just barely there.

51

I’m going to predict things again. Here’s how I think the bubble collapses:

  • First, the smaller AI companies will go down. They’ll just run out of cash and won’t be rescued. We might see some desperate mergers so their investors can pretend all that imaginary book value is still there.
  • The last one to go down will be OpenAI. It’ll run out of money so thoroughly that not even everyone put together whose accounts are held up by OpenAI equity will be able to pump in money fast enough to keep the company propped up. They finally have to admit their losses.
  • Without the flagship, the whole thing falls down.
  • Google, Microsoft, and Amazon can finally admit this never made any money. They have an excuse to account massive losses and go “oh well.”
  • Nvidia will be fine — they’re the ones who got everyone else’s money.

Stranger

52

NVidia stock will suffer - that share price is being propped up by expectations of huge future earnings - but NVidia the company will be fine, as long as they don’t announce some kind of massive special dividend or stock buyback for their investors, something that makes actual cash go off their balance sheet.

53

Nvidia’s market cap today is at 4.45T, so essentially flat since August. (The market sank today because of Trump’s new tariff threat against China but those have all been blips.) Trump’s investing federal dollars to take a 10% stake in Nvidia hardly changed its stock price at all, which means the market shrug what should have been an epic move.

This reminds me of the days when everybody online was forecasting the bankruptcy of Netflix. That’s usually a good time to buy.

54

Are you confusing that with Intel? I’m not seeing news about the government buying part of Nvidia.

55

Intel, not Nvidia:

Reuters: “US to take 10% equity stake in Intel, in Trump’s latest corporate move”

Nvidia’s stock valuation may suffer an adjustment as the rest of the tech sector recorrects but Nvidia is still a fundamentally solid company producing a needed product for which they have essentially cornered the market through genuinely more useful technology.

FWIW, I’m not convinced in the Pivot-to-AI scenario in its entirety; there will certainly be fallout from collapse of the ‘AI bubble’, if just because the tech industry is so invested in AI as the as the future of profitability that they’ll keep floating it as long as there is the appearance of something there. Unlike Enron, which was dependent upon the delivery of a vital resource (energy), most of the AI-oriented technology development is really predicated on specious and nebulous claims of ‘efficiencies’, and setting aside all of the prognostications of AGI in 2025/6/7, there was never a concrete promise of any particular capability. But the capex to keep expanding ‘compute’ is going to dry up (and even if it didn’t, the energy sector wouldn’t be able to scale to provide it), and OpenAI is going to face some kind of reckoning of the exaggerated bombast that Sam Altman likes to slide into every presentation while GMA are going to necessarily scale back their investments to something more inline with what the technology of LLMs and other transformer-based architectures can do, which frankly isn’t all that much from a practical standpoint.

Stranger

56

While AI may be in a hyper-bubble right now, transformers are still a fundamentally novel and useful technology. This idea that you can statistically model everything from language to music to images and videos, just by throwing sample data and compute at it, is a pretty fundamental shift in how the world deals with information aggregation and conversion. There’s no going back from that new reality; even if it’s not the final step in AI (and nobody thinks it is), it is still a major leap forward.

Even if all of today’s AI companies go bankrupt, this architecture isn’t going to disappear overnight. It’s already too useful for things like transcriptions, translations, images, ideation, etc. Truthiness is still a hurdle, but there are many outlets and fields where that isn’t essential (sigh, ads and marketing for one). Transformers will keep evolving, and other data-dependent architectures will be experimented with.

This is like the “steam engine” of our times; even if a few early “railroad barons” don’t end up making it over the long term, the digital “industrialization” will still continue.

And since what Nvidia ultimately sells isn’t packaged AI anyway (unlike OpenAI and its competitors) but hardware for parallel computing — the shovels — they will keep making money as long as some process still requires a lot of compute. A decade ago that was video games, then crypto, now AI, and (according to Nvidia), robotics next.

There aren’t that many major GPU companies in the world to begin with. There’s Nvidia, AMD, Intel (up and coming or past its prime, depending on who you ask), Apple… maybe Qualcomm and Samsung and Google and Amazon, at smaller scales. Any time an industry needs a lot of parallel compute, it’s up to one of those companies to provide it. And at scale, in data centers, it’s overwhelmingly Nvidia except for the few FAANGs that can afford to experiment with their own in-house chips. Everybody else buys Nvidia or rents Nvidia.

No doubt their stock will drop quite a lot if the AI bubble goes bust — and I don’t see how the AI bubble could not bust — but Nvidia existed long before that bubble and will hopefully exist after it. They’re just an impressive hardware company with good sustained leadership over decades; that’s really quite different from your average hype-of-the-day techbro startup. Their valuation may be nuts right now, but their product is a fundamentally sound (and essential) infrastructural part of modern life.

57

Gah. Just ignore that sentence. The rest makes sense. I hope.

58

I have been extremely skeptical but am becoming less so. It’s hard to look at capacity like this, an LLM trained to work through difficult diagnoses, and not be impressed at least a bit.

https://hms.harvard.edu/news/ai-system-detailed-diagnostic-reasoning-makes-its-case

It is limited still. Mainly by the history and findings entered which is part of a skilled diagnostician’s toolkit. But still.

59

It is less impressive if it doesn’t consistently give reliable diagnosis or medical guidance consistent not only with the data it was trained on but also the local medical culture and available treatments, and is also cost effective and provides a measurable improvement in patient outcomes or reducing physical workload. IBM developed a health care version of the “Watson” system as a non-LLM AI ‘expert system’ for “clinical decision support systems” which, while not performing primary diagnosis, reportedly does provide accurate guidance for treatment plans and guidelines as well as tracking electronic patient medical data, notes from healthcare providers, pertinent clinical studies, and journal articles, and other administrative functions. It reportedly worked pretty well compared to expectations but the cost and lack of adaptability to local practices made it unprofitable, and IBM spun it off and sold their health care division to venture capitalists.

Now any really revolutionary technology is going to take a while before applications that hew to its strengths and aren’t undermined by the weaknesses winnow their way through the market but the extent to which LLM (and multimodal) based AI has been hyped as a cure-all to every need, desire, and ailment would be impossible for anything short of literal sorcery to live up to, and in the case of critical medical applications the requirements for high reliability and minimizing factual errors are pretty obvious failure points for this application. A medical LLM-based tool providing diagnoses or treatment plans which makes basic errors or provides inaccurate information even a small fraction of the time is still unacceptable (at least, in a rational system, although health insurance companies apparently have no problem whatsoever in using these systems to make insurance denials even when they are seriously flawed). If you have to double check every single thing that your ‘AI medical assistant’ gives you because it is inherently unreliable, it is worse than useless.

Stranger

60

Let’s focus on this specific system. It isn’t Watson. It isn’t what insurance companies are using to maximize denials. It is not at this point, for many reasons, replacing any humans. It is specifically trained on a predominantly United States data base and is going to be no more or less cost effective in its diagnostic and treatment plans than human physicians in tertiary centers.

And it is impressive. Not perfect. But as meaningful to my read as when a computer match performance with chess master, or more so, with a Go master.

These tools will increasingly be used by doctors on tough cases, at first to see if the system comes with ideas they missed. Then they will start to trust it and use it first just checking its work. And then, I fear, they will rely on it, and become detrained.

To be real, most cases doctors see don’t need this tool. Most cases are straightforward and most of the diagnostic skill is recognizing when it is not the straightforward, not recognizing the pattern, but recognizing that it doesn’t fit the usual pattern and digging more. Nothing presented yet convinces me these systems are good at that.

And it is still impressive.