SoftBank sells its entire stake in Nvidia
cnbc.com
cnbc.com
> SoftBank said Tuesday it has sold its entire stake in U.S. chipmaker Nvidia for $5.83 billion as the Japanese giant looks to capitalize its “all in” bet on ChatGPT maker OpenAI.
They are switching gears, not exiting, folks.
This is Softbank though, they're not really noted for great investment decisions (apart from Alibaba really early on).
Like, my prior is that when Softbank invest in something, the growth is done (but then I am, very much, a cynic).
It's just baffling they are still getting billions to spend.
[1]: https://www.wsj.com/articles/softbank-sells-entire-nvidia-st...
When you realize that most investors at that level are degenerate gamblers, and that they're even worse about thinking "the Generals were due!" than your buddy who can't look away from the DraftKings app for more than five minutes, it's not so baffling anymore. The only difference is that when you have that much money, you own the casino.
https://news.ycombinator.com/item?id=21799821
Bill Hwang of Archegos Capital Management is a close second (lost $20B in two days, and basically brushed off margin calls while his highly overlevered bets were collapsing, incurring $10B+ of losses at the investment banks he worked with).
https://www.bloomberg.com/news/features/2021-04-08/how-bill-... | https://archive.today/lM0SU
Almost everything he has done since then has lost money. When I read that he is investing in something, I just assume it will end badly.
They came back telling us he was basically asleep during the pitch meeting which was scheduled for only 10 minutes anyway.
Our business/product really had no chance of succeeding at this point and most knew it. We got some money from Softbank anyway - forgot how much. Our management was basically laughing about how easy it was to get funding from Softbank.
I jumped ship a year later or so and that was good timing.
And if that is the VC model, then are the VCs smarter than everybody else?
Mayoshi Son isn't playing the VC game like most traditional VC funds. He's operating on a massive scale and his LPs are sovereign wealth funds, who can have other geo-political priorities than pure profit.
Some people don't know that the usual VC game is often still be profitable for the VC even if their fund isn't profitable. VCs are investing the money of their LPs (Limited Partners), who are usually very large institutional investors with billions under management (think state pension funds, Harvard endowment). Most of the LP's funds are invested in a diversified blend of safer, lower-return vehicles but they take up to 5% and invest it in high-risk, high-return things like venture capital and hedge funds. But they spread it across a dozen or more firms with different strategies.
So each VC is playing a portfolio bets and their LPs are playing a portfolio of porfolios. The LPs just need one of their 12+ VC funds to be a lead investor on a unicorn win. This math usually works out in their favor (there's now >50 years of data). VC funds charge their LPs a yearly management fee of a few percent of the invested capital - whether the fund makes money or not. Over the 10 year life of a fund, this adds up and covers the VCs overhead and very generous salaries - usually >$500K at larger firms. In the VC's view, $500K/yr isn't getting long-term "rich" but it'll pay for a pretty lux life. Even with the VC taking out fees, the LP's math still works thanks to only needing one VC firm to win and if one or two more of their VC funds just return 2x or 3x. It maths up even better.
When your personal worst-case downside is $500K/yr minimum with substantial upside, it's not a bad gig. However, these VC types are generally top-of-class Ivy League grads, who are clearly very sharp and ultra high-potential - the type who'd expect $500K earning opptys on Wall Street, consulting, investment banking, etc.
I think this model and portfolio investment strategy is the result of the fact that VCs can't pick winners, though. They are forced to take this "high-risk, high-reward" position in the portfolio because they can't predict the result of any of their investments, even over a whole fund's worth of investments.
Imagine a world where startup success had nothing (or very little) to do with luck, where you could plug a bunch of standard metrics into a complex algorithm (probably a spreadsheet) and it would spit out an utterly reliable set of statistics on likely return. Something that an actuary could come up with, as they do for insurance/assurance.
In this world, VC isn't high-risk, high-reward. There would be categories of VC fund, some would be lower-risk, lower-reward, others higher risk, and yet others that would be low-risk, high-reward (but priced accordingly).
The fact that we don't have this, and all VC is high-risk, high-reward (and often no reward at all, even across an entire fund), is testament to the fact that we can't pick winners. And the reason we can't pick winners is that startup success is very non-deterministic, i.e. mostly down to luck.
https://markets.businessinsider.com/news/stocks/charlie-mung...
but you don't make the same type of deduction for lottery winners - surely a good lottery number picker should get a second win sooner or later!
Therein lies the problem. People do take it seriously, particularly on the idea that he's now due, and keep feeding him money to make these bets with, instead of tossing him out of the boardroom, like he should have been with his bet on WeWork.
If they skip a beat, or it becomes impossible to increase performance (TSMC node stumble or P4-style dead-end), then it won't take long for someone to catch up.
If they keep executing on better hardware performance + software to support it? They keep the money spigot on.
They don't have a moat... because their moat is their performance growth rate.
If they stopped today, competition would flood in. But by the time competition catches up to Nvidia today, Nvidia has moved several steps ahead.
You’re also wrong. TSMC produces but a small part of the products that Nvidia sells: the silicon. They don’t produce any of the tens of thousands other components that go in a server.
And calling the actual processor a “small part” of where Nvidia’s value is doesn’t jibe with reality.
Neither is OpenAI's.
SoftBank have holdings in Arm and likely to know what is coming down the pipeline.
If they are privy to information that we don't know that led to this discussion, isn't that insider trading?
They may know more, but not that much more.
> SoftBank has been a repeat investor in Nvidia. It sold off its investment before the AI boom took off and then bought the chipmaker's shares again before divesting in October to double down on its biggest investment bet, OpenAI.
By proxy, having the strongest frontier model becomes less and less necessary for them and instead building a strong product by properly layering medium-strong models in a cost-efficient way is the priority.
In the end, regardless of technical understanding, people will always shop around on price if the feature set is similar enough I suppose.
If I want an LLM answer to "is erythritol bad for you", I'm not firing up ChatGPT. I'm just typing it into Google, and the LLM answer it spits out is pretty good.
ChatGPT needs to be significantly more compelling for most people to use it for one-shot LLM answers over Google Search. And the minute Google removes the one-shottedness of its search answers, it's over for ChatGPT.
Imo ChatGPT is just "a feature not a product", in the search engine space, as the adage goes.
The techies are the ones who are steering clear and sticking with search engines in my experience.
And much of the MS Office suite is actually a worse copy of another product, merely copied and forced via network effect. There are more than enough far better alternatives to e.g. Teams and OneDrive. The latter even makes Excel far less reliable to the point of unusability.
Not to mention Windows is easily replaced for the truly average person who spends their entire computer use inside a Chromium browser nowadays. Outside PC Gaming (which is far from the average if we are honest), the vast majority of people never touch software that isn’t already on mobile or more likely exclusively in the browser.
Familiarity and brand strength, not need or compatibility, keep the customer facing offerings of MSFT going. That isn’t inherently unreasonable either, for these people Windows is fine as Chrome works well enough on it and they do not have any needs beyond that, so why switch. Same with search, Google still finds flights, restaurants, Wikipedia, etc.
Being better is worth little if the vast majority are sufficiently well catered to by the incumbent. Even if another OS or search may improve their experience somewhat (better reliability, longer battery life, etc.) that doesn’t trounce the established player as long as the few basics aren’t utterly broken even to them.
You can actively antagonize your users, be worse than the competition, more expensive too, you only loose most if you make the few things most have purely become reliant on and are expecting worse, basically regress.
Reason all the bugs in Windows are fine, after all the memories of most people are dominated by BSODs and just not crashing daily is enough to be considered pretty good, despite the overall experience far worse than most other OS/DE combis.
I wish this was true
More subjectively, but if you want to ignore the "AI overview" and consider that not results (then why are they on my results page?!), just look for a specific, slightly older video on Youtube, a site Alphabet themselves run. You can provide the exact title down to the letter and are in many instances going to struggle to find it with Youtube search or Googles Video Search, even when wrapping in ", etc. Kagi and DDGs video search tabs meanwhile yield the desired results. Again, this doesn't mean they are amazing or massively far ahead as Google was able to reliably provide such results in the past. It just means that Google Search has regressed, yet users stuck with it because, as I've been saying, quality (or lack thereof) has a hard time overcoming brand strength.
Maybe I'm out of touch but I don't know anyone for whom that's true. They all use either games or Office or Photoshop type stuff.
I mean you can replace it but all the options - mac, linux or chromebooks have friction.
And friction on MacOS, Linux, ChromeOS, Android or iOS, I don’t see that considering the vast majority still switched to 10 despite the departure from a decades long established environment and, lest we forget, quite a lot of XP/98 software also did not work well on more modern Windows, again, not really a problem for these either. There is no reason not to switch despite the brand, which I again must emphasize, is ok. They are fine enough with the status quo, even though alternatives would be more stable, secure, have better battery life, etc. I just dislike the idea that some in the space cling to that better alternatives are surely going to be adopted and the only reason they aren’t by the masses is compatibility, slight differences in UI, etc.
Regarding Microsoft it’s not as easy to switch as going on another website, OpenAI doesn’t have that ecosystem lock in.
But it didn’t, Google users beyond a few initial super users who switched early on in the late nineties/early two thousands, track directly with the massive growth in new active web users. These new first time web users, far more than switching made up the growth between 2002-2006 that made Google dominant over Yahoo. It also should be kept in mind how much better Google was over Yahoo, whereas I am doubtful LLMs will have such deviations in quality, looks more like the opposite.
Point remains, OpenAI will, in my opinion, retain the strongest consumer share of LLMs, regardless of output quality, introduction of ads, etc.
This is not ecosystem lock in, it is solely branding. What do you think most have in mind when you mention search? What when it comes to online shopping? Now do the same for “AI”. Only one brand comes to mind for most.
People use Google search because they have (or rather, need) a Google account. Also, there are no alternatives really, because search is very difficult.
Microsoft has alternatives, but it doesn't matter, because you're artificially restricted from using them. All your friends (MBAs who cut your check) are using Excel, so why don't you just use Excel too?
Both Google and Apple (ie any searches in Safari make money for Apple based on the Google deal) said there was no slowdown on Google searches and the “AI overview” by Google is “good enough”.
Just like Blackberry, but instead of having to buy an new device to switch to the competition, the customer just needs to click on a different website.
OpenAI is already the consumer-first AI platform. And, in my opinion, in the minds of the average consumer, the biggest one that is "an app" with memory and folders, vs. something like Gemini which is perceived as an extension of Google Search and thus doesn't have the same "knowledge database" UI.
They don't really, because of Gemini and Grok.
If it was OpenAI vs Claude then yeah, ChatGPT is known whereas Claude you have to be an enthusiast to know.
But everyone using Google runs into Gemini and everyone using X runs into Grok, and people talk. Now even the laypeople know there is more than one AI, and they're from big brands that they trust. Which means that people will window shop for the AI with the best performance per cost. Bye bye brand power.
I mean, even just the name - 'X'. You can tell that's some Elon Musk bullshit because it's the stupidest fucking name I've ever heard in my life and nobody with even a quarter of a brain would ever suggest that as a brand name.
When that's literally the brand of your company, naturally I question if everyone has the screws tightened up there, you know?
We do not know for certain as of yet, but I’d be very surprised if Apple felt the need to loudly communicate to their user base that the upcoming Siri improvements aren’t Apple Intelligence. They won’t lie of course, but Googles brand in the space isn’t nearly as strong with the layperson, whereas they more than happily shouted their use of GPT from the rooftops initially.
Apple is in the business of contracting out new-hotness until their polish is done.
Then they put their name on the thing, pretend it’s new technology – and depending on the named feature, sometimes it is!
The numbers are just mind boggling even in the optimistic scenario.
I'm personally skeptical that they are able to pull off ads, at least on the short timescale they likely need to. They more or less have to nurture/disrupt a whole SEO industry, and the way big corps allocate advertising budgets are very slow to shift (many are still struggling to find their footing in social media advertising).
Ads are not just a switch one can turn on. Firstly, you need to build a decent ad serving/targeting/pacing engine. Secondly (and more importantly) you need to hire a shed ton of sales people (in many, many geos) and then ramp them all up (difficult if you're building the product at the same time).
And then you need to keep at it for 3-5 years minimum before you'll finally get the bigger/more conservative brands/agencies/etc to buy in properly.
At that point, you'll make decent money, after accounting for all of the costs. I'm not sure that you'll make enough money, but it would definitely stanch their bleeding a little.
tl;dr if they haven't already built this ad product, it's unlikely to make a material difference before 2030.
The person/department/agency that is responsible for doing SEO is the same one that is responsible for spending ad money on Google. It's in their best interest to internally sell "search" (no matter if organic/inorganic) as a good advertising channel and that creates great stickiness for Google's ad business.
Microsoft sells services pretty much everywhere on earth and has a near-monopoly on PC OSes, enterprise email, productivity suites plus a giant cloud computing service....and their annual revenue last year was just $245 billion.
OAI can't get anywhere close to that, not least because they have multiple competitors and several large corporations (MSFT, Google, Meta) would rather dogfood their own solutions than use a rival's.
I agree with you that OpenAI seems much more risky in terms of it's actual true viability as a business, but the risk:reward must be there for Softbank.
> “This should not be seen, in our view, as a cautious or negative stance on Nvidia, but rather in the context of SoftBank needing at least $30.5bn of capital for investments in the Oct-Dec quarter, including $22.5bn for OpenAI and $6.5bn for Ampere
> That amounts to “more in a single quarter than it has invested in aggregate over the two prior years combined,” Bulk said.
> … Following the recapitalization and SoftBank’s $22.5 billion investment into the ChatGPT maker, the Japanese firm’s 4% ownership of OpenAI will increase to 11%.
It sounds like Nvidia was played out for them and they’re aggressively trying to get in on OpenAI so they can exit sometime after retail investors get in
Even if you're all-in on OpenAI, does it not make financial sense to have _some_ stake in Nvidia considering they are the only ones with an actual moat?
Unless there are CUDA alternative breakthroughs we will hear about in the next few days.
https://finance.yahoo.com/news/amazon-says-anthropic-will-us...
> Anthropic is using 500,000 of Amazon’s Trainium2 chips for its Claude AI models as part of the tech giant's Project Rainier
They did not. They sold MOST of their stake in Nvidia long ago before anyone paid attention to Nvidia vs AI. It's one decision as part of a complex cloud of decisions.
Also this "stake" was around one thousandth of Nvidia. Minor. Most companies have lots of owners with much larger stakes.
AI inference on the other hand is basically just very large floating point tensor matrix multiplication. What does an ASIC for matmul look like? A GPU.
> What does an ASIC for matmul look like?
A systolic array, and ultimately quite different than a GPU. This is why TPUs et all are a thing.
In general with a systolic array you get a quadratic speedup. For example with a 256x256 array, it takes 256 cycles to shift operations in and out, but in doing that you accomplish 65k MACs in 512 cycles for a speedup of 128x over serial.
In the big picture, TPUs are systolic arrays. They don't have threading, divergence, or similar.
GPUs in the big picture are SIMT, a hybrid of SIMD and multithreading where individual data streams in SIMD are relaxed to allow them to diverge somewhat.
Memory Wise the TPU can keep the partial products right in the array. Parameters and weights are held in large on die scratch memories, and backing those are streams coming from the HBM. The TPU acts as a single giant CISC coprocessor, and has much more predictable memory and communication patterns vs a GPU, which its design exploits for higher efficiency at inference vs GPUs.
So even if they use the word "tensor" and both have HBM based memory systems, how those are actually architected is very different.
Most of the hardware we are using was designed for computer graphics not AI. Now that China isn't buying Nvidia any longer and actively trying to get their own companies to produce hardware, what happens to all these datacenters when a company produces a device that has 80% of the performance of the current Nvidia hardware but 20% of its power consumption?
Softbank can also choose not to.
https://www.reuters.com/business/media-telecom/cantor-nears-...
Masayoshi Son is BFF with Trump and Lutnick:
https://asia.nikkei.com/politics/international-relations/how...
There is more going on here than is publicly known. Investing in hot air is of course a specialty of the Trump admin.
California's residential real estate market is $10 trillion.
All it would take for the stock to double is every last person in California selling their house and buying NVDA. /s
OpenAI just completed separating it's non-profit and for-profit restructuring: https://www.cnbc.com/2025/10/28/open-ai-for-profit-microsoft...
This probably means the for-profit structure will be going public in 2026, and there is probably a last private round happening.
Softbank is known for doing it the other way around...
No AI player will admit to be positioning for a sector collapse but it would be naïve to believe that anyone close to ground zero isn't planning contingency plans and are just coasting in the golden age of AI. And those who have the most skin in the game are certainly the ones most incentivise to do so while denying their true intentions.
LLMs are going no where but the reality of them not being the civilisation shifting money makers people have hyped them up to be is undeniably setting no matter what the AI astroturfers hiding in HN tell you. People have been jumping off the train or positioning to reach the closest chair when the music stops. Not long ago 2023, the world bubble was just whispered here. 2 years later it has hit mainstream and has been uttered by people holding the money bags. It might be a dream or it might not.
The difference is that, if you were in denial, now you are seeing signs that are much harder to deny. And we should expect the upcoming signs to be harder and harder, until the number of people denying reaches to zero right as the bubble fully bursts and the AI startups layoffs begin.
(That doesn't necessarily mean sell it all. But HODL can get you broke. It's prudent to take some juicy gains while they're available.)
Masayoshi Son has form when it comes to calling the top of the market with this particular company.
Really living up their reputation? Or, am I way off-base here? I am just an outside observer and I would love to be educated.
Did it makes sense for Softbank to be in so little on the company "selling shovels" to all their "gold mining" investments?
Microsoft (which has >50B into OpenAI)->Anthropic for biz use cases
Softbank->OpenAI. Yet again, the dumbest money ever?
There are many, far worse Softbank fails. How does this keep happening? Who is advising this team?
I am available. 80% sure that I could do better. I just genuinely hate to see capital being destroyed like this.
But also selling ALL is a bad move, says one who sold all of Nokia at 1000 markka in 1996.
Not nothing, for sure, but low enough that you can buy it back in a couple of months.
By purchasing OAI shares now, Softbank will probably be able to sell them to Nvidia at inflated prices later.
Also, I highly doubt they sold those 5bn of shares in any exchange. Most likely they did it over several weeks on a dark pool. Because again 5bn is massive and it would cause a crash in any open exchange.
But hey, maybe the market crashes tomorrow and this is seen as the best piece of timing ever seen, maybe it doubles in the next year.
In any case, you never lose by taking profit.
Until it wasn’t. Then it was again.
Either way, cash flow != stock price, not that SoftBank is in a ‘business’ that likely needs a lot of cash.
Maybe they just needed a new yacht?
Or need to bail out one of their investments?
The bubble is starting to crack.
There are 3 facts here:
1. Softbank's nvidia stake sale has happened
2. Softbank has not put it all directly into the AI gold rush
3. 2 other investment entities have also sold most/all their stake in Nvidia
Do what these 3 facts as you will but these are ripples in the surface of the tech world
The Gpt-5 release just confirmed what most knew. And we know that the best time to leave is when the market can still get slightly hotter.
> SoftBank said Tuesday it has sold its entire stake in U.S. chipmaker Nvidia for $5.83 billion as the Japanese giant looks to capitalize its “all in” bet on ChatGPT maker OpenAI.
-3% straight after opening of the stock market 20 min ago.
The question is: now that Nvidia is the new Intel, is OpenAI the new Microsoft?
Softbank clearly thinks so.
A smaller scale version of this might be happening here. Or it could be a bubble. Either way it's interesting!
Additionally, for the AGI producer (or platform host), there would have to be some way of price function that would cover the costs of operating or providing the AGI service.
Thus, a more realistic look at this is “purchasers of AGI would have no reason to purchase “more” of it beyond a certain threshold, after which it is no longer useful to their ends, however purchasing shares of the AGI producer would expose them to value from new customers who could use it better”
Hahah! Fucking hell, openai is going to fail big time, they have nearly a 1,5 TRILLION USD in debt. There is no way they can handle that, their only option is to cozy up with trumo and make the tax payers pay for all that.
Seems your 1.5Tn figure is in "spending commitments" and not current/hard debts. I.e. budget earmarked for datacenters and research from '25-'35 or so.
Nvidia is the leading shovel maker.
...
I don't understand the rationale here.
I'm on investing subreddits all the time, and I'd expect it there.
But I don't see how this is "anything that gratifies one's intellectual curiosity."
If there wasn't an AI bubble narrative, then sure, this this would gratify my curiousity. But now I don't see it, not even in the most charitable way.
I'm curious what the line of thinking is on how this does, in some way, gratifies one's intellectual curiosity.
Edit: I figured that I'd get all the downvotes. I've been here long enough to understand the social dynamics of the site. Funnily enough, I was more in the "Hacker News" demographic between 2015 and 2023. Since then, it has shifted a little. Nowadays, I have to force myself a bit to sound more positive than I actually am in order for my comments to be appreciated (in terms of upvotes, as I do view that as a form of social feedback), and that is fine.
I understand that this post gives bad vibes or sounds perhaps a bit mean? I am not intending it that way. I really just don't get it. Look at my comment history, I sometimes ask questions like this, but not that often. I suspect I'm not the only one in this.
They don't know the future, just like the rest of us.
If we're talking indicators, if 5 Softbanks would do it in rapid succession one after another it just mentions to me that the "smart money" is showing signs of moving out.
In terms of whether AI will or will not fuel growth, I think it will fuel growth. Self-driving cars seems to be a solved problem for cities at least fairly soon (e.g. Waymo, anti-example: Tesla, camera's is not the way).
It's a question whether LLMs state of the art models will grow more, but what hasn't been done that well yet is integrating it into current software. I know, because in part, that's my job. There's still a huge productivity unlock there, also in ways that people can't fully imagine.
Right now, LLMs seem to be an enabler for software engineers, especially software engineers on smaller projects (I can't find the research at the moment, it was a while ago that I read it). It seems to be an enabler for many people, but they do need to put time into prompting it in a way that works for them.
Fixing the context window issues and others I think will be really hard tasks, because I suspect we then need to know what goes on inside the black box.
If an LLM could continuously learn, so somehow continuously keep updating its weights such that it learns better, that would be a breakthrough.
I wouldn't be surprised if they reinvest some of that money into AMD or similar, as those companies play catch up. AMD is bringing ROCM up to speed with AI models, and their consumer hardware is having pretty good press despite some PR fumbles.
Selling Nvidia stock isn't a sign of lack of faith, it's just an exit after strong results. Softbank think this is a peak or near enough, good for them.
As you say, until there's some breakthrough, there's little point keeping money in the dominant company.
Also, Nvidia appear to have delayed their 60-series GPUs, and this will cascade to their entreprise models, due to the 3GB module memory shortage. The next gen will be VRAM heavy, but you can't do that during a shortage, you need the market to correct tor for supply to increase. So any 'breakthrough' powered by new hardware is now about a year away.
This won't stop people secondguessing on the bubble front though. Personally, I think bubble popping is going to come from a lack of faith from investors in the downstream companies like OpenAI who are struggling to make money from their resource intensive products. Nvidia are already profitable with the hardware, MS/Google/Amazon will always make money with the servers. And if AI bursts, other sectors can soften the blow for those companies. It's the massively inflated AI model makers that need to worry about who makes it out of this profitably.
Stock news is barely on HN.
Oh, wait, I guess I see it now. It is on HN way more frequently when it involves Big Tech. And NVidia is increasingly seen as part of that. It used to be FAANG but now it's the Magnificent 7. The bias shifted.
That'd make sense.
Quality of technical content and discussions has visibly gone down since. More emotional/low-value/political replies everywhere. Slowly turning into /r/programmerhumor...
most of HN bought their houses on the back of softbank throwing money at everything that moved