Nvidia says two mystery customers accounted for 39% of Q2 revenue
Posted by DataLore19@reddit | hardware | View on Reddit | 241 comments
Posted by DataLore19@reddit | hardware | View on Reddit | 241 comments
Tower21@reddit
Well, I guess Nvidia earnings are the canary in the cole mine on the AI bubble.
This current AI bubble, while at least providing a much better model than the last 5 or 6 AI bubbles, depending on your count, doesn't get us to our AI overlords.
Constrained within a limited scope, there is a definite use case, and we will feel the ripple effect of this for many years.
I just don't see the path to artificial general intelligence, with this latest bubble. There is the chance to test run what an AGI would mean to our society, while not actually getting there.
I hope we will learn lessons from this that positively affects society, but I fear we won't.
advester@reddit
The internet is a real transformative lasting thing, but dot com still popped. This AI bubble can pop even if current AI tech is the real deal.
Strazdas1@reddit
dot com bubble popping didnt sink the companies manufacturing fiber optics.
Glum-Position-3546@reddit
It almost killed Cisco (they never recovered to their 90s stock price), which is probably the closest dotcom analogue to Nvidia.
Companies making fiber optic cables can pivot to another industry, or make different cables all together (it isn't a stretch to switch from making fiber optic cable to making copper cable for utilities, for example). Nvidia however makes GPUs, they can't exactly pivot away from GPUs.
TheImmortalLS@reddit
ive run starlight mini, which is a diffusion upscaler, but i'm convinced it's doing the equivalent of the enhancement meme but only but with actual math like motion vectors and differentials because look at the results
NSA and other big governmental players like china (in general) need more horsepower to spy on their citizens and research/make sense of junk data
ResponsibleJudge3172@reddit
AGI is not the "main goal". Only LLM producers care about that.
Specialized domain AI is. A medical AI developer cares not for AGI fantasies
Sillypugpugpugpug@reddit
It's just a fucking algorithm, it's a leap forward but it's still shit.
one_who_goes@reddit
And do you think our brains are anything other than an algorithm?
Strazdas1@reddit
Yes. Our brains is a neural network.
maelstrom51@reddit
If you simplify things enough every thought and feeling you experience is an algorithm carried out via chemical processes.
Being an algorithm doesn't mean much.
Strazdas1@reddit
Of course. Humans are just bio-computers with a lot of redundancies built in to make us keep running multiple decades.
Strazdas1@reddit
can you list what you think is the last 6 AI bubbles?
got-trunks@reddit
Always kinda weird looking at things like this, a trillion dollar business can run into existential problems if some nerd works out a bit of math that renders the need of a datacenter stuffed with MW of compute obsolete.
I mean, probably not but it's always interesting to think about past sorta similar examples. I've only just started playing with local models and stuff and it's really interesting in general, but I've also found it helpful just to sort of help me collect and format ideas.
I really dislike some aspects of it, I think for day to day use my need to automating tasks is pretty limited. But if I had the same ease of access to these tools a just over a decade ago when I was starting out in IT, I could see how my days would have been a lot more efficient.
twnznz@reddit
When your customer base is a handful of whales your revenue risk is enormous. NVDA is a terrifying stock and you should stay the hell away from it.
fumar@reddit
Nah. They can't keep up with demand. If Microsoft says they can't use any more ai chips, there are plenty of customers who would take more
puffz0r@reddit
that's not how supply and demand works, if 20+% of the demand goes away the price drops because there's not as much competition for hardware.
fumar@reddit
That's how it works if supply was previously met. But demand has greatly exceeded supply for generations now.
puffz0r@reddit
That's why nVidia can get away with charging $40,000 per GPU. If one of the big players pulls out, they won't be able to charge the same amount.
BavarianBarbarian_@reddit
You sell all your bread to 20 people for 10€ each bread, and there are a hundred more people who would've bought at 10€, but you were already out of bread. 10 of those 20 buyers don't come back the next day. Do you now lower the price of your bread to 5€ out of the goodness of your heart, or do you remember that there were 100 people yesterday who would've paid 10€, who haven't found any other bread as good as yours in the meanwhile?
Animewaifylord@reddit
See you got it wrong if there's a 100 people then they'll sell bread so that the person in top 20 can afford it, if 10 of those rich folks goes away price falls to the next top 20 level
BavarianBarbarian_@reddit
That's what I'd think normally, but it seems like Nvidia hasn't priced the cards like that. Perhaps they're afraid that if the "top 20 richest" fail to create anything profitable from AI, their entire structure could collapse, and are therefore trying to get cards into more hands?
Animewaifylord@reddit
Companies aren't buying graphics card like 5090s for Ai datacenters, that's the gaming side and entirely different market, they're buying the 20 thousand dollars blackwell cards. In high demand they price these cards upwards of 40 thousand usd and their new chip is going to launch above that price. Ai Companies aren't buying your thousand dollars graphics cards
_Meru@reddit
Why were you pricing your bread at only 10€ to begin with, knowing that this price leads to a quantity demanded exceeding that of your constrained supply? You have 20 loaves of bread, yet you set the price low enough that you have 120 people willing to buy at that price. I think that's the source of confusion here.
The person you are replying to was assuming Nvidia's prices are the equilibrium/market-clearing price, which seems to not be the case, there is a demand-excess. Jensen Huang has told AI customers to "pace themselves", and that they plan on scaling up. I don't know enough to comment on why Nvidia's prices are set the way they are now. But maximum short-term profit extraction doesn't seem to be their goal here. Prices can reflect many complex factors including customer relationships, regulatory constraints, etc...
Strazdas1@reddit
Because if i make the bread 11€ other people start baking bread too.
_Meru@reddit
Fair enough
BavarianBarbarian_@reddit
Hell if I know. Maybe they think that if they make it more expensive that would encourage their customers to instead invest in the competition to try and get them up to the same performance?
aggthemighty@reddit
This isn't bread, it's AI chips. If 20% of Nvidia's orders are canceled, that probably signifies something happened to the AI bubble, and I'm not sure the small fries would still be eagerly lining up for their chips.
Animewaifylord@reddit
Yes demand does exceed supply and that's factored into the price of the commodity, the current buyers are paying more than whatever the next guy is willing to pay that's why they're getting chips, one of the two big company decided to make their own chips or use a competitors, then others will obviously come it to buy the chips but they'll do so at a price they're willing to pay
_Meru@reddit
Any idea why they haven't raised prices in spite of this?
csprofathogwarts@reddit
They still have to strike a balance. Ask too much and market share of AMD as well as specialized equipment manufacturers - Cerebras, Groq starts rising up.
HiSilicon's Ascend 910C chip based Cloudmatrix 384 cluster has already beaten Nvidia's GB200 based NVL72 cluster in performance. But it requires more chips (thus, probably more expensive) as well as more power. There is a price point where operating that cluster might become cheaper than buying Nvidia's - and Nvidia would definitely like to avoid that.
yflhx@reddit
This is also currently priced in however.
luuuuuku@reddit
In what? In the market cap? Yes, but that doesn't really matter to nvidia. In revenue? No, how would it.
Animewaifylord@reddit
Its priced in the price of the chips themselves, basic demand and supply, the current buyers are paying whatever the next guy is willing to pay, but if one these companies like say Google decides to make their own chips instead, demand falls by 20% now the other smaller players will obviously cover the demand but they'll get lower prices because they're not being priced out by Google anymore
luuuuuku@reddit
Do we know that? From nvidias perspective the demand didn't really change because they're not doing direct sales. By the time the cards actually arrive on the market, nvidia already made their revenue. Prices for individual GPUs didn't really go up that much either.
NVIDIA doesn't really sell graphics cards, their partners do. Apart from some smaller exceptions, nvidia only sells the GPU die and memory, not full graphics cards. For their Gaming cards that's Asus, Gigabyte etc. and for the datacenter/workstation segment the actual PCIe cards are built buy PNY, who is probably one of the "mystery" customers. Supply and demand matters moe for retailers which nvidia is not.
Animewaifylord@reddit
If they weren't doing direct sales then they wouldn't be saying their revenue comes from two companies, and tweeting Elon made a record 100k purchase and whatever. It makes zero sense to not price the graphics cards at what the top 20 in the market can pay for them, theyre not going to keep their margin same as before just for the good in their hearts
luuuuuku@reddit
You didn’t understand what I said. Please read it again.
Animewaifylord@reddit
See you got it wrong, you're talking about graphics cards sold asus, gigabyte and the like when Ai chips are completely different, corporations aren't buying 5090s for AI and Graphics card prices is totally different market entirely. Gaming is small portion of Nvidia's revenue most if comes from their blackwell cards for high end data centers not gaming chips they sell to asus
yflhx@reddit
Yeah but market cap is like one of the most important things for investors. And the discussion was about stock.
surg3on@reddit
When big customer #1 stops buying the driver will be impacting big customers #2 to #20 as well
tomsrobots@reddit
If Microsoft drops demand for chips, the whole industry will follow suit.
luuuuuku@reddit
That’s normal for computer hardware. It’s the same with CPUs, like 60-70% of all x86 CPUs from AMD and Intel are sold to Lenovo, HP and Dell. Those sales are distributors for the most part. The hardware companies don’t really do direct sales, so their individual customers never appear in any statistics.
tomsrobots@reddit
I mean, this is really different. Selling to distributors is healthy. Selling to a couple of companies building data centers is shaky.
azn_dude1@reddit
They're building data centers so that they can distribute computing power (they're CSPs). How is it different?
advester@reddit
Dell has proven customers and isn't stockpiling. The AI datacenter is spending investor bubble money and hoping it may pay off.
azn_dude1@reddit
It's basically guaranteed that somebody will find a profitable use for computing power. That's been the case since computing power existed. You don't need a known proven entity to be your customer, the entire industry is your customer.
Animewaifylord@reddit
No it's still risky, like what if even one of those companies is like: "Hey you know instead of buying so many expensive chips, let's make our own or let's switch to that cheaper manufacturer" Now Nvidia's demand falls by 20% in one announcement and chip prices come way down
azn_dude1@reddit
That's already happened lol. Google, Meta, Amazon have all done so but still continue to rely heavily on Nvidia. Yet what you said would happen didn't happen hmmm
Animewaifylord@reddit
I'm not saying it has happened already but it can happen in the Future. Google, amazon, meta etc don't make their GPUs right now but the same thing happened to Qualcomm, Google has been using their chips but decided to make its own chips, IF they decided to make their own GPUs then demand WILL fall, that's why it's risky Even it didn't happen in the past doesn't mean it can't happen, people said the same about Qualcomm and Intel, Ai isn't even that old
azn_dude1@reddit
Google has been making their own TPUs to accelerate AI workloads for 10 years. Amazon and Meta have also announced similar efforts to make their own AI chips. Where was this 20% dip in Nvidia demand you prophesized?
Animewaifylord@reddit
Chatgpt and Ai bots aren't that old, those chips could still be in the pipeline. They're buying nvidia cause right now they want the best chips, everyone is competing for general super intelligence and no one can wants to give their guys second grade chips, when they think it'll be more profitable to use their own chips they will simply switch
azn_dude1@reddit
It's easier said than done. It's not "a 20% drop in demand when there's only an announcement" kind of thing. Like I said, Google has been making TPUs for 10 years and hasn't made a complete switch. There's not that much volatility among the big companies, hardware isn't the kind of industry where a big company can immediately manufacture a competitive product and software stack and turn on a dime to switch over.
Strazdas1@reddit
Azure/AWS also has proven customers
luuuuuku@reddit
what makes you think those two customers are single datacenters? There is no evidence for that.
jv9mmm@reddit
That's not what he said
Trzlog@reddit
At least with Lenovo, HP and Dell, you know there's structural demand that won't disappear overnight. Dell isn't going to decide to stop selling PCs tomorrow. They aren't going to decide to stop using Intel tomorrow.
luuuuuku@reddit
There is "structural demand that won't disappear overnight" for nvidia GPUs too
Trzlog@reddit
Which is irrelevant to what Nvidia is making their money with today.
luuuuuku@reddit
Source for that?
tessatrigger@reddit
trailing stop limit and my profits are locked in forever if the stock suddenly tanks.
leoklaus@reddit
When the AI bubble pops, NVIDIA will be absolute fucked. If their datacenter revenue dropped to before bubble values, they’d lose almost 40 billion in revenue (~80%) and that’s disregarding a market that is completely oversaturated with datacenter cards.
NVIDIA is probably one of the worst ticking time bombs in market history.
Raikaru@reddit
There’s quite literally no reason why all AI demand would completely stop.
andrerav@reddit
While Nvidia's customers are ponying up the cash for all this AI compute, a highly significant portion of that compute power is currently failing to generate ROI while putting a huge stress on power grids and the environment. Unless that turns around somehow, a strong market correction is a scenario. The current upward trend is being driven by (as of yet) unsustainable business models.
Zaptruder@reddit
The rationale is simple; if they get to the point where AI can and does indeed replace massive swathes of the economy, you'll wish you had some money in AI stocks or something.
Whether they can or can't pull it off... well, that's a matter of technical understanding that evades most people (including many Reddit commentators on either side of the discussion).
Strazdas1@reddit
AI/robotization already replaced massive swathes of the economy. Look at things like Dark factory/warehouse/store. Even if we dont fully robotized, half the service workers are already repalced by software. Even family diners now ask you to scan a QR code and order online while you are sitting in the diner.
leoklaus@reddit
Again, none of this is the type of AI that is currently being pushed.
Literally not a single one of your examples is using large amounts of GPU compute during any part of its life. None of these things will sell even a single $40k datacenter GPU.
The only thing that requires these insane amounts of compute are model training and inference for general purpose models like the ones OpenAI is developing. And these models are generally unsuitable for productive work.
Even if a factory needed image recognition, they wouldn’t need a model that can recognise the difference between a bonobo and a chimpanzee, because all it needs to recognise is two types of screws. Apart from being much more expensive to run, a general purpose model would also be much less accurate than something trained for that specific use case.
Strazdas1@reddit
Sounds like you only look at the surface for the chat LLM models and completely ignore all the other AI being developed.
And those factories are absolutely using datacenter GPUs to run their models. 40k is cheap when you compare it to cost of employing people.
leoklaus@reddit
What AI are you talking about here? What AI does a factory need that would warrant a datacenter GPU or even outsourcing their compute?
Strazdas1@reddit
Depends on what they are manufacturing. Ive seen AI that figures out the most optimal way to do laser cutting of wood based on a 3D software model.
andrerav@reddit
For that to happen, solving the energy and environmental issues would be prerequisite, and that is a much more complex problem than the replacement itself.
AI can't help us if it ends up consuming the earth.
Strazdas1@reddit
solving environmental issues isnt a problem of intelligence. Its a problem of culture/politics. We know what needs to be done, we dont have the balls to do it.
andrerav@reddit
While your english is broken, I understand what you are trying to say. And the most simple response is that no-one ever said otherwise.
And so the bubble grows.
Strazdas1@reddit
Im sorry if my english is hard to understand, i am not a native speaker.
andrerav@reddit
That's fine, you got your point across, albeit a moot one.
leoklaus@reddit
I'm by no means an AI expert, but I can tell you with absolute certainty that what most people refer to as "AI" and what is pushed so hard right now is fundamentally incapable of replacing or even approaching human intelligence and is therefore extremely unlikely to have any noteworthy impact on the economy.
LLMs like ChatGPT cannot think or reason. Simply put, they generate responses based on the likelihood of the next word following the input prompt and the already generated output. With the huge sets of training data these models have, the responses they give are often correct by chance and (this is a really big issue) will pretty much always seem to be correct, even if they are completely wrong or "made up".
No amount of training, development or technological progress will ever change that because these models are simply not designed to think or reproduce knowledge.
There are branches of AI that are actually very interesting and absolutely capable of replacing or at least assisting humans, like those used for image or pattern recognition (for example in the medical field), but those have been around for years before the big hype and are not the driver of this bubble.
Strazdas1@reddit
You dont need human intelligence to replace most jobs. A factory worker sorting batteries into packpaging isnt using his peak human intelligence for the job, just image recognition and hand movement. AI can do that and does now.
leoklaus@reddit
Apart from this being a task that would be much more suited to be handled by a "stupid" robot, AI could do that for over a decade and the models you need for tasks like these can easily be trained and run on mid-end consumer GPUs from 10 years ago.
An entire factory could easily run on a few consumer GPUs costing way less than $10k.
This is neither the technology that OpenAI, Microsoft and co. (primarily) work on, nor is it the technology that warrants building data centers for billions of dollars.
Strazdas1@reddit
AI can do jobs of stupid robots are surprising speed. Like, the same tasks but AI finds a way to do it faster if you dont need complete accuracy. Image recognition is a great example. AI can tell objects in an image tens of times faster than dumb algorithms can. Sure, it gets it wrong 0.05% of the time, but thats good enough for the speed benefits.
The technology that OpenAI, Microsoft and o thers are working is very diverse with different leaders. Just taking LLM models for example Google has the best visual models, Microsoft has the best coding models, OpenAI has the best communication models, Meta has the best physics models, etc
Also dont forget things like research. There isnt any modern drug research lab that is not using AI heavily. And yes, its done locally, on locally build server racks. But the AI has to be trained first.
Zaptruder@reddit
Even if what you state is true - the difficulty is convincing everyone else of that. But for the most part, given that this sort of criticism has been around for years, even as capabilities continue to improve in a variety of arenas - it remains unconvincing to a sufficient number of people that AI investment remains at an all time high.
Professional-Tear996@reddit
The AI bubble will pop because people will have to confront the energy/environmental costs of keeping the party going much sooner than they had anticipated - unlike the lethargy that has been induced in them due to the sheer scale of replacing fossil-fuel infrastructure.
Zaptruder@reddit
Projected power usage for AI datacenters in 2030 is ~1000 TW/h. Total global power usage in 2030 30-35k TW/h.
So it's very significant - but hardly the only factor in pushing us into our environmental barriers.
I expect more finger pointing and more kicking the can down the proverbial road in 2030.
Professional-Tear996@reddit
It is not the proportion of power usage by datacenters in relation to the overall power usage, but the manner in which datacenters are turned on whenever they need to fire up the GPUs for training purposes.
It will become unsustainable unless every datacenter is energy harvesting during peak hours for regular use and only turns on at off-peak hours for datacenter use.
leoklaus@reddit
At some point investors will want to see a return, and there are very few niche AI products that actually have the potential to generate meaningful profits.
They can’t keep burning money forever and once the bubble pops, it will be insanely hard to find investors for future AI projects.
Raikaru@reddit
Uber started in 2009 and has only been profitable once in 2023. Investors are willing to be “irrational” for way longer than you think.
leoklaus@reddit
Uber had a business model and a clear path to profitability…
pet_vaginal@reddit
AI too.
leoklaus@reddit
If there is a clear path to profitability, you should be able to lay it out for us.
Strazdas1@reddit
work automation has been very profitable source of revenue.
leoklaus@reddit
What work can ChatGPT automate?
Strazdas1@reddit
Ill give you one example from my company. We have GPT4 do consistenty checking of invoices we recieve to note if they dont match contracts and may be scams. Then a person looks at them and decides if GPT4 was right or not. This way a single person + GPT4 has saved the company nearly a million euro this year and did a better job than 3 humans used to do before it. It automated two people out of the job and did it better.
leoklaus@reddit
Your company employed three people to check invoices for their legitimacy full time? And these people each made 500.000€/year?
Or is the million euro saved the result of your company not getting scammed out a million this year (I’m assuming in contrast to the years before)?
Sounds like your business has more pressing issues than looking for ways to automate using AI…
Strazdas1@reddit
no, these people cost about 20k euros per year. nearly a million was savings from not falling for fake invoices.
pet_vaginal@reddit
Not everyone offer inference at loss. It’s assumed that API usages can have descent margins. Otherwise OpenRouter would be a lot emptier for example.
Same for subscriptions. I wouldn’t bet that every subscribing user is a net positive but that’s also a valid business model.
Now the game is to gain a good user base at any cost, accumulate data, and build the best infrastructure.
Path for profitability is not a secret, enshittification, and better margins over time.
leoklaus@reddit
That’s all fair and good but who’s paying for that? Cheating students aren’t going to cut it.
How are they gaining a paying customer base in the first place? Enshittification is the endgame.
pet_vaginal@reddit
I think you should look a bit into AI usages. Indeed people don’t send all their money to NVIDIA to help students cheat at exams.
And yes they do have a paying customer base. It’s a bubble and the projections are likely non sense, but it’s not based on nothing either.
leoklaus@reddit
Can you give me some examples of AI usages?
nanonan@reddit
That's selling AI as a novelty, not an actual economic use of AI.
andrerav@reddit
When you buy the business plan on Temu \^
pet_vaginal@reddit
Do you think it would be an AI bubble if no one believes that money can be made ?
andrerav@reddit
No, that would contradict the very definition of a tech bubble. I firmly believe that large swaths of incompetent people firmly believe that money can be made.
pet_vaginal@reddit
Plenty of money have been made by companies who survived the .com bubble.
andrerav@reddit
Sorry, I don't respond to straw man arguments.
pet_vaginal@reddit
I'm sorry my comment didn't match your expectations.
mposha@reddit
Uber (and AWS for that matter) had utility though and didn't lose money with each user.
Raikaru@reddit
comparing AWS to Uber is literally 100x dumbest things ever. I’m just wondering by this logic does google not have utility?
mposha@reddit
Unless I'm misunderstanding, since you drew the comparison of "AI" and its lack of current profitability, to Uber's initial lack of profits, I extended that same comparison to AWS as they were also considered "unprofitable" initially (at least in their reporting) and are now obviously profitable.
Raikaru@reddit
AWS became profitable way sooner than Uber. Uber is 4 years away from 2 decades into being around and still isn’t profitable yet it’s mc is near its peak
mposha@reddit
Yet neither lose more money as they scale.
Raikaru@reddit
idk if you’ve looked at uber’s financials but they actually did lmfao. Their biggest struggle years were 2017-2022. They finally managed to turn it around in 2023
LickMyKnee@reddit
Delusional.
jocnews@reddit
No, the people who paid for the GPUs are going to get screwed over, or people they borrowed from / got investment funds from.
Nvidia will laugh all the way to the bank even if the bubble burts because those suckers were already parted from their money which is now Nvidia's.
Nvidia made more net income in 2-3 years than Intel has in a decade, they are going to be more fine than any other company, unless they ship GPUs against not yet paid contracts etc. I doubt they do.
leoklaus@reddit
I‘m not talking about NVIDIA the company but NVIDIA the stock, as was the comment I replied to…
NoleMercy05@reddit
Short it hard. You can get rich!
Octane_911x@reddit
Have you seen the insane demand from OpenAI? They don’t have enough data centers, so they rent from Microsoft, Google, and Amazon, while also building their own. Even with that, it’s still not enough, so they had to make ChatGPT-5 more efficient to handle more customers. They also have more features to release but are constrained by compute power. Imagine all the competitors, all hungry for compute. There’s so much room for growth in AI and its potential. And guess what—it’s not like you can patent your idea. Your code is private, and what you provide to customers can generate huge revenue. GPT already charges $20, and even more for pro customers.
If blizzard charges WoW customers for $9 a month with a massive amount of customers and they made huge revenue, now imagine AI.. people are paying $20 and more. It did not even explode yet
Sofar coding AI is on the front followed by graphics, their is no ceiling
tessatrigger@reddit
trailing stop limit and my profits are locked in forever if the stock suddenly tanks.
W0LFSTEN@reddit
Is it though? Are Meta GPU purchases going to drop off a cliff?
Plank_With_A_Nail_In@reddit
They still extremely profitable without these two customers.
_extruded@reddit
Melon Husk and Zark Muckerberg
Zookeeper187@reddit
It’s Microsoft and Meta
blaktronium@reddit
MS and Amazon is my guess.
RevolutionaryCoyote@reddit
From the article:
So Microsoft and Amazon are likely driving the demand, but they are not the actual direct customers.
Exist50@reddit
That doesn't sound right. All the major CSPs roll their own infrastructure, top to bottom. They don't just buy servers from HP or Dell. They should absolutely be buying directly from Nvidia.
fratopotamus1@reddit
That's not true - they go through ODMs like Quanta, ZT, Super Micro, Foxconn, etc.
Exist50@reddit
ODMs are different than OEMs. If that's the argument, then Nvidia has basically no direct customers to begin with.
KeyboardG@reddit
Why would they be a secret? Microsoft is shouting from the rooftops how much they are investing in AI.
ApprehensiveCook2236@reddit
You mean Macrohard and Off-Meta
GenZia@reddit
I thought Suckerberg was busy with his "Metaverse" boondoggle in a desperate charade to become a real-world James Halliday?
dev_vvvvv@reddit
Whatever happened to that anyway? All I heard about for like 6 months was everything would be in the metaverse, it was the future, etc. And then it disappeared into nothingness.
Animewaifylord@reddit
Should rename his company again
ResponsibleJudge3172@reddit
Nothing. The internet got tired of hating on it and moved on. Quest still sucks billions every year and they have expanded to wearable glasses now
Strazdas1@reddit
Metaverse exists and multiple vendors offer it. I think it shows great failure in communication just from the fact that people think Metaverse is Zuckerbergs and not a group effort of which Zuc is only one member.
ch1llboy@reddit
I don't like all the weight of the headset on my head.
tecedu@reddit
Got rolled into the meta products, so the meta glasses, whatsapp, instagram and facebook. all of them slowly got absorbed into the metaverse, its not the same as what we all predicted
Kepler_L2@reddit
Same thing as 3D TVs, VR, AR, Web3, Crypto, NFT...
GenZia@reddit
Been wondering the same thing.
The whole failure of the thing is a good enough reason for his "impeachment."
Perks of being a chairman, I suppose.
HCharlesB@reddit
Had no legs.
BlueGoliath@reddit
Imagine wasting that much money on something so stupid and failing so badly and not going under.
TenshiBR@reddit
The amount of press he got was insane. All major networks doing long expositions about it. There were a variety of companies in other areas doing big events promoting it. The Tech Bros types were all over it, saying it was the future...
Kudos for the more conservatives pointing out it was just another NFT similar crap for them to get more money
cashew76@reddit
Sunk Cost.
glguru@reddit
I have heard that XTX are also buying lots of nVidia chips as well.
oh_woo_fee@reddit
Can you decode this? Couldn’t tell who they are
Mandelaa@reddit
NSA (USA) & IDF (ISR) ?
Professional-Tear996@reddit
The latter has MSFT welcoming them with open arms, so unlikely. Though Nvidia is hiring there, so you never know.
DistanceSolar1449@reddit
Ji Xinping
SeepDeek
StellarJayEnthusiast@reddit
Russia and China most likely
dkizzy@reddit
Mystery at that percentage? Lol
shroudedwolf51@reddit
I wonder if it has to do with the smuggling operations that NVidia has been turning a blind eye to.
ExynosHD@reddit
Probably not.
Strazdas1@reddit
"Military does not use modern chips"
hm___@reddit
Should be illegal with sums that big
luuuuuku@reddit
why?
Getherer@reddit
Wdym why? Because it could be a war mongering government for instance? Trading transparency exist for a reason
luuuuuku@reddit
Because it's irrelevant and probably even illegal. But I think you have no idea what you're talking about
Beginning_Month_1845@reddit
The US has exports restrictions to certain countries, especially China for AI GPUs, which so happens to by nVidia’s biggest profit source. “Mystery customers” sounds very similar to not saying who you are selling to when it is obvious there are some parties you regulated on how and what you can sell
luuuuuku@reddit
Are you stupid? "Mystery Customer“ is a made up term from the press that tried to click bait you into an irrelevant article. Just because NVIDIA doesn’t make it public doesn’t mean that us authorities don’t know that. Nothing relevant happend. NVIDIA has to file quarterly reports according to the security exchange act. And in there, basically in one line in text NVIDIA mentions that most revenue comes from two customers, Customer A and Customer B. That’s all that happened. This is normal and even though I got plenty downvotes for that, no other publicly traded company discloses the names of those customers in these reports. This report is for investors, not for authorities. If anything it’s expected, most customers will probably not publicly disclose how much they’re paying for their products. This is absolutely normal and was it there for a while now. The same is true for many other companies. Don’t fall for click bait just because it fits your narrative.
Kezika@reddit
Quite the opposite actually, it is illegal to not to do.
ASC 280-10-50-42
evernessince@reddit
Record and transparency requirements for companies.
It's also particularly important in this case because Nvidia deals restricted products. Not disclosing your biggest customers is immediately suspicious. It would be akin toif tsmc was like "Our biggest customers are Apple, AMD, Nvidia, oh and some mystery party that accounts for 50% of our sales".
luuuuuku@reddit
Name one company that does that. And link a source.
evernessince@reddit
I named one right in my comment, tsmc.
https://www.guru3d.com/story/amd-might-move-to-samsung-for-3nm-fabrication-due-to-tsmcs-preference-for-apple/
Gotta love that lightning down vote you gave me ,you have no intention of having engaging argument here he's just here to downvote people who disagree with you.
You can't deny the facts buddy.
luuuuuku@reddit
that's for one node, not entire company revenue.
Jonnny@reddit
Why are you so entrenched into a position on such an obscure, who-gives-a-shit-to-most-people topic?
sultan-of-ping@reddit
Cop out, take the L gracefully at least g
BioshockEnthusiast@reddit
Yea that guy's behaviors screams "I'm capable of grace" lol
SilentHuntah@reddit
Yeah, that shit's so childish haha. It's like you know that I can see you downvoted my comment not even 15 seconds after I posted my reply?
Mattbird@reddit
AND use MLA formatting
SkitzMon@reddit
Not disclosed to the public does not mean not disclosed to the federal regulators and their legal team.
wdeezy@reddit
For what it’s worth, it is. But not until the end of the year. It will be a risk disclosure due to concentration of revenue sources in their annual report; this is not a required component of the quarterly filings, though.
Balance-@reddit
This sounds very interesting, could you share some more details / things to read up upon?
wdeezy@reddit
ASC 280-1050-42 is the US Generally Accepted Accounting Principles topic and coupled the SEC Regulation S-K 101(c) public disclosures the practical result is companies disclosing their major customers explicitly. They can technically sidestep naming, say Amazon, but in doing so (especially with orders of this size) are just obtuse as they will have to say “a leading publicly traded company with large operations of direct to consumer retail sales, cloud infrastructure, and diversified hardware interests” that gives up the game anyway. Most just say “Amazon” and move on with their day.
Kezika@reddit
Doesn't it just require the naming that it's one customer though, at least according to Deloitte's article on the topic and their example on not specifically naming the customer they could just say "Revenues from one customer of nVidia's electronics segment represents approximately 39% of the company's Q2 revenue."
https://dart.deloitte.com/USDART/home/codification/presentation/asc280-10/roadmap-segment-reporting/chapter-5-entity-wide-disclosures/5-7-information-about-major-customers
Although I could just be misunderstanding the example.
whiskeytown79@reddit
A lot of their tech is subject to export restrictions. They might not want to reveal who the customers are, but they certainly know, and so does the government.
TheImmortalLS@reddit
NSA and china backdoor #1 and #2 customers lol. they need more compute horsepower to spy on citizens and make sense of junk/noisy data
ive run starlight mini, which is a diffusion upscaler, but i'm convinced it's doing the equivalent of the enhancement meme but only but with the AI eqv of actual math like motion vectors and differentials because look at the results
TheImmortalLS@reddit
NSA and china backdoor no doubt
Vushivushi@reddit
The two customers are major manufacturing companies like Hon Hai (Foxconn) or Quanta Computers.
The customers we're all thinking of, Nvidia considers indirect customers. Meta, X, OpenAI, Microsoft, Google, Amazon, Oracle, Coreweave, etc., they buy from those manufacturers.
For these indirect customers, Nvidia will only say that there are 2 which represent 10% or more revenue attributable to compute & networking (datacenter).
They also disclose that cloud service providers are responsible for 50% of their revenue.
peterking2000@reddit
I work a contract job for Quanta and I can confirm this. Our number one customer at the moment is Google. We are wrapping up the GB200 line, and GB300 are coming up. Lots of money are involved.
jocnews@reddit
The biggest hyperscales now basically act as server vendors for themselves, they just contract factories to build. Their GPU orders are likely direct, they try to sidestep as many other sub-contracting companies as possible to cut expenses. Still, Nvidia rips them off, the amount of venture capital that will be lost when they all realise AI doesn't pay is going to be nasty.
Will be harsh for those whose pension funds will be left as bagholders.
pixel_of_moral_decay@reddit
This is the right answer.
NVIDIA consults with downstream customers in terms of wants/needs for product line planning, but the companies you think about aren’t customers. They are downstream.
NVIDIA makes very specific chips, which without a bunch of other stuff they don’t make are useless. Those manufacturers sitting in between are what make it useful.
T1beriu@reddit
Microsoft, Meta, Amazon or Google.
Andr0id_Paran0id@reddit
Elon and _____ (im guessing thiel or altman)?
asdfzzz2@reddit
and Definitely_Not_Chinese_From_Singapore_I_Swear.
SirActionhaHAA@reddit
According to nvidia's q2 earnings.
ProfessionalPrincipa@reddit
Nvidia also said in 2018 that crypto had little to do with their revenue numbers. Those statements subject of a shareholder lawsuit which is now proceeding forward again.
ResponsibleJudge3172@reddit
Now their revenue has eclipsed even back then
xternocleidomastoide@reddit
the tax reason being not having to pay sanction penalties...
It's basically impossible to track those units past Singapore ;-)
SirActionhaHAA@reddit
They were ordered by american companies so if they are breaking the law they should be punished by the administration :-)
Most of them were never sent there, they were redirected and that was the problem. Ya know that right? :o)
fumar@reddit
The American company might just be a shell corp for a smuggler.
tecedu@reddit
Dont think a smuggling company would have so 39% of revenue
Strazdas1@reddit
thats 39 smuggling companies in a trenchcoat.
fumar@reddit
Definitely not.
Soggy_Association491@reddit
At that point, why not a different country company instead of America?
xternocleidomastoide@reddit
You don't know how any of this works, right? :o)
Vb_33@reddit
No u
SkitzMon@reddit
Singapore is a trans-shipment hub. There isn't enough affordable land there to host the datacenter hardware being bought via Singapore.
RawbGun@reddit
It's not actual individuals but companies. The obvious answer is Microsoft and Meta
xAI is tiny compared to those
constantlymat@reddit
Sam Altman is running his entire company of Microsoft Azure credits due to a licensing cooperation with Microsoft.
He hasn't got that type of liquidity.
BigBananaBerries@reddit
Theil would be a good shout now that he's dealing with the whole US Government civilian database.
5553331117@reddit
No one thinks it could also be Google?
MairusuPawa@reddit
No, Google is working on their own ASICS now.
ResponsibleJudge3172@reddit
They have 6 gens of TPUs. Does not stop them from buying billions worth of Nvidia chips
clearlybreghldalzee@reddit
now? as in 10 year ago when google TPUs were introduced?
Strazdas1@reddit
google runs their own TPUs now dont they?
DistanceSolar1449@reddit
Go look up Google TPU V6
luuuuuku@reddit
So, what? What does it have to do with NVIDIA GPUs?
DistanceSolar1449@reddit
https://simple.wikipedia.org/wiki/Substitute_good
luuuuuku@reddit
So, what? They cannot replace NVIDIA GPUs.
DistanceSolar1449@reddit
What rock are you living under?
The Google TPUs very much can replace Nvidia GPUs. Google TPUs are the only things that are competitive with Nvidia GPUs on the market.
They trained Gemini 2.5 Pro on their own TPUs. https://www.vincentschmalbach.com/the-best-ai-model-doesnt-run-on-nvidia-chips
luuuuuku@reddit
Google is still using lots of nvidia GPUs. Don't act stupid. Running Gemini is not the only use case for GPUs at Google. What rock are you living under? Ever heard of AWS, Azure or Google Cloud?
Kryohi@reddit
Their biggest models are all trained and run on TPUs. They have zero need for the vast amounts of Nvidia hardware referenced in this article, zero.
luuuuuku@reddit
But Google is also the third largest cloud provider that lets you rent GPU nodes. It doesn't matter what Google uses for Gemini. Does Gemini generate profits? No. Google Cloud is the fastest growing business and when they rent out nvidia GPUs they cannot replace them with their TPUs. Doesn't work that way. Google Clouds generates significant profits.
DistanceSolar1449@reddit
Uh, they barely own any Nvidia GPUs, they literally try not to rent out the few Nvidia GPUs they have and instead tell customers to use their TPUs- that's exactly what they do. https://cloud.google.com/tpu?hl=en#how-it-works
They own only 169k H100s (from 2022) and barely purchased any B100/B200s (from 2024). For reference, Nvidia quarterly revenue is $46.7 billion, so 169k H100s at $25k each is a mere 9% of Nvidia's quarterly revenue. Or about only 8 days worth of revenue for Nvidia.
Jesus, learn to take an L.
UsernameAvaylable@reddit
lol. Because google is not selling them to anybody. Google does not run their large scale stuff on Nvidia GPUs.
luuuuuku@reddit
They still buy lots of nvidia gpus.
Dreamerlax@reddit
Google has their own FPUs.
3legs1bike@reddit
article says that's unlikely, because they are an indirect customer. more likely it's an OEM manufacturer or something like that, where other companies like cloud providers buy from
wave_action@reddit
Supermicro?
Not2plan@reddit
Isn't supermicro a vendor for NVDA?
wave_action@reddit
No, they basically make the servers that the NVDA cards go into. So when the comment mentioned that it was likely an OEM manufacturer that sounds like something Supermicro would do. Like if Microsoft wanted 1000 AI servers for their datacenters, they would go to Supermicro who probably gets first dibs on buying cards so that the big buyers can buy servers.
Creative-Expert8086@reddit
dell?
fumar@reddit
No. They have their own custom AI GPUs for internal use. The do sell Nvidia GPU use on GCP though. It's Microsoft and one of Amazon or Meta.
Plank_With_A_Nail_In@reddit
Remember Nvidia design the GPU's, they do not make them, the cards with them on they don't make those either, their servers....they don't make those either...its all outsourced. Nvidia doesn't make finished products...their founders cards are made by unnamed manufacturers.
The customers have to be companies that put them onto PCB's and make finished products.
ResponsibleJudge3172@reddit
Remember, TSMC doesn't make wafers, or EUV machines, or the treating chemicals, clean romss, electricity, or non of the essentials for the business
Strazdas1@reddit
didnt they use PNY to do the founder cards this time?
nanonan@reddit
You're entirely correct, the fact that this is downvoted is a poor reflection on the userbase here.
LongjumpingBaker5041@reddit
I’ve stopped following Nvidia news and data already. In the short term, it’s just going to keep going up and up anyway.
DrSpreadOtt@reddit
I worked e-commerce. There was nothing I didn’t know about you with a simple purchase. Any purchase. $1-$1,000. I have your address. Your credit card details. Explain how two massive customers are a mystery?
luuuuuku@reddit
NVIDIA doesn't have to disclose that.
And no, nvidia doesn't know where their GPUs actually go to.
The same is true for AMD and intel too.
Kezika@reddit
ASC 280-10-50-42 says otherwise
DrSpreadOtt@reddit
What does this have to do with who bought them? I don’t care if you stick whatever you buy where the sun don’t shine. I know who you are though and that’s what I’m trying to say.
luuuuuku@reddit
Only if you're doing direct sales. Obviously, nvidia knows who these customers are, they just don't want to disclose that.
But no hardware company does direct sales but sell to big integrators like Supermicro, Dell, HP, Gigabyte, Lenovo etc.
Liesthroughisteeth@reddit
Don't worry, they are probably just cutouts for the Chinese and Russian military. Nothing to see here.
AnxiousJedi@reddit
I'm guessing that the 'mystery' customers are China and China.
No_Hornet_1227@reddit
Two mystery customers aka ``Jensen is helping china avoid the sanctions and smuggle gpus through other countries and doesnt do shit about it because money``.
Seems like Jensen along with the entire nvidia board of ceo should be going to prison. And singapore and hong kong should be banned from importing gpus for at least 5 years + sanctions for helping china avoid the law.
wind0wlicker@reddit
Gamers Nexus did an episode on one of these mystery customers, maybe both :)
hanumanCT@reddit
If it’s not a big tech player, it’s probably some hedgie doing high frequency trading with AI. Market has been fishy for years and I’ve always suspected a shadow group manipulating the market.
selfstartr@reddit
Note to investors - your stock could crash if just 1 big customer leaves Nvidia.
luuuuuku@reddit
The same is true for AMD and intel
Vo_Mimbre@reddit
They account for the companies we know about.
If they’re hiding the names it’s gotta be controversial to investors, or borderline problematic. I don’t say “illegal” because especially in the U.S., what’s legal and not changes at the speed of Twitter.
So my guess is China and Taiwan. China is an issue due to above (legal at the speed of Twitter) and selling to both together is a geopolitical optics challenge.
kimi_rules@reddit
Yup, obviously for Amazon with their AWS racks and Meta that has millions of those.
Archaon0815@reddit
CIA and Mossad
Archaon0815@reddit
Palantir
hitsujiTMO@reddit
Elon and Zuck would not be "mystery" customers.
Those buying them to import them into China are the mystery customers.
CaptainDouchington@reddit
The mystery is that it was probably China