For last 5 quarters, Nvidia has sold much more GPUs than get installed. On 3 year depreciation schedule, 8xh200s are $1.50/hour per card just sitting there. B200s are $2/hr. (Nvidia changes network cable each generation). Very expensive to do what you are suggesting is happening, but only other explanation is smuggling into China, or secret sovereign/military purchases.
It’s also possible that they are lent to tier 2/3 datacenters in revenue share deals.
The fundamental lie behind the AI fraud is that compute is scarce relative to demand. OpenAI, Grok, Meta did overinvest in hoarding GPUs far ahead of their usage. Last 2 are now competing on hourly GPU market, and AI tokens/revenue has fallen 20% since spring peak, which is a bubble pop compared to 10x/year perpetual growth expectations behind the hoarding. While hourly gpu rental market is stable rather than declining, it is stable at very low prices, mostly for accounting reasons of not actually losing money intentionally. Deployed GPUs are abundant proven by accounting floor based pricing. A b300 can deliver 8x the tokens of an h200 but only rents for 2x more.
Im no saxophologist, but doesnt the sound come out of the buthole?
The senior government official pointed to busy naval activity in the region around Greenland
This is an extreme lie. Pathetic europe deserves to be destroyed for tolerating it. There is no russian or chinese need to passage close to greenland, unless to deliver goods to us or canada.
Only reason usneeds greenland is to threaten europe.
This is a pro Israel force deployment. Not a force to help prevent Israel from more abuse of Lebanon. Puppet Lebanese government is not allowed to have its own army.
most of the fuel weight required is to lift the rest of the fuel. Fuel costs is about $1m for full load. Rest of cost is huge staff, maintenance, and capital cost of rocket.
The $200/kg launch price target is based on 150 ton capacity. That’s a $30m launch costs target. Volume/foldability matters the most because that is the actual constraint that limits datacenter launch to a single NVL72 size.
It will never be an economic thing. Only unpluggable skynet military thing. The weight is not an issue. though. It’s volume.
Incredibly bad, too lenghty, mostly irrelevant criticisms of the fraud of space datacenters, followed up by the Skynet military justification of being unable to unplug skynet.
Space datacenters from SpaceX are a fraud because they have a 5 year lifecycle with deorbiting of entire unit. The costs compared to 30 year lifecycle of terrestrial solar/battery powered datacenter energy is thus 6x higher (costs of shell/shield, solar, radiators is about the same but 6 replacements). Terrestrial building costs are $20/watt. SpaceX ambitions are to get $30m/launch costs. To be only 2x the terrestrial costs, launch costs need to be $1m (just the fuel costs) with deorbit being to fly off into space instead of a salvage trip.
At 12x the costs, the competitive GPU rental hurdle has to be 12x more expensive than earth. Only military skynet applications would pay for this, and specifically, only permit mechahitler to decide if skynet is doing a good job.
Headline is innacuralte. Customers who purchase a $200/month subscription, can save up to $13800 on token costs compared to “a la carte” token pricing. Open AI does lose over $1 per $ in revenue, but this is more a function of them having too much compute, and very high training costs, rather than a loss per gpu hour on tokens served, it should stilll be a loss for OpenAI:
Open source model pricing per gpu hour, ranges from 80% margins at high batch saturation per user (low tps per user) to 50% at concurency of 8, to loss at fewer user requests per gpu. https://inferencex.semianalysis.com/compare/deepseek-r1-b300-vs-h200
OpenAI has very high prices, and too much compute. They would likely lose $6000 per user who maxes out their $200 plan, and certainly over $1000 per user.
It’s a good deal for world for reopening SoH. No other alternative for opening it. US/GCC would have incurred too much permanent damage. While Trump is only president stupid/compromised enough to have fallen for Netanyahu bait to start, he’s also only one with reality distortion power to call a stop, victory.
World/Americans being liberated from USrael empire is closer today. There is crazy congressional bills to add $700B (70%) in military spending this year with $500B permanent, and to put Mossad in command of US intelligence. This gets harder, if military goals are limited to Cuba/Canada, or recession hits, and US power/“pretenses of idealism in abusing power” loses credibility. There’s a far better world with Iran an influence peer.
I’m optimistic this will stick. Quiet control/tolls over SoH. Sanctions/funds relief. US leaves. 60 days talks likely to extend past mid terms, with Trump declaring victory the whole time for stopping nuclear program, before finalization. GCC likely to line up investments in Iran.
Israel can likely be contained. American assets will be forced to say how great Trump’s victory is. Israel has limited supply of US missiles, and needs cooperation from air refueling services to bring planes to Iran.
what do 4 small stars on flag represent?
Monarchies over the ages are a bad example, because there was a sustainability mandate to pass off a good country as inheritence to heirs. A candidate for worse countries would be Idi Amin, and US allies, but for US allies, their alliance is forced due to CIA coups/colour revolutions, and the scale of Idi Amin’s pillage is a million x lower. The utter oligarchist/corporatist supremacist nazi economics and geopolitics of US, together with the pure Israel/Zionazi capture of every elected official is a corruption that is beyond any other in both scale, and in fundamentally voluntary complicit bribery scheme for pure bribery value, excluding Mossad operations such as Epstein. The late stage empire pillaging stage is next level. That Congo arrests officials for corruption proves low levels of extended corruption. ie. punishments happen.
China does build datacenters in desert next to solar farms. Water access is overblown, because it can use a closed loop system where water is trucked in from somewhere it is more abundant. Cooling in desert is more expensive, but China has abundant cheap power. It is unfair to say the US is only place that cares about it’s people, when it is the most corrupt country in history that cares the least about popular welfare. China’s drive for abundance is a populist drive. US scarcity profit maximizing mindset is the opposite.
One of the reasons Huawei is far ahead of Nvidia on cost competitiveness of cluster systems is both their last (384 matrix) and current (950 based) clusters are containerized and don’t need buildings. Nvidia B200 and later systems require liquid cooling with heavy duty single story floors purposed built with embedded liquid cooling pipes in them. This is usually a $10b premium per gw of compute in building, not including the ability to reuse any other building/datacenter with older nvidia systems.
The need for more GPUs/datacenters was based on xAI using them for their own models, instead of competing with new datacenters/GPUs in the rental market. GPU rental prices went down to below cost levels for H series after Anthropic deal. It has bounced back this last week, but rates are still in the “very well supplied” range. On demand rates half of Anthropic/Google rental deals. GPU rental rates are the best measure of AI bubble point. Concerted fraud is still possible to provide the illusion of “this is fine”. The google deal only starts in September, while Anthropic got some free GPU time when deal was announced.
In a way, pumping more money into AI (through IPO) delays the bubble pop, even if NVidia doesn’t sell as many GPUs if their customers add them to rental market. Bubble can pop from 3 directions. Less GPUs/TPUs sold than forecast; datacenter overcapacity ; understanding poor economics/tokenomics of high competition LLM labs.
The sad part is that the more money invested, the greater the funding for the fraud, and longer lasting fraud.
There are signs that IPO is too big for market. 30% retail, and $135/share fixed price are because demand is not present. The illegal private/SPV market is trading below $129, where SPV entity that was not a fraud about to go bankrupt, and confident in the IPO would buy back shares up to close to $135.
This BS by google is desperation to pump their existing stake. Anthropic’s headline $1.25B/month deal is also a fraud. Big discounts until after the IPO, and same cancellation rights.
xAI got a made up (no one putting cash in) $250B valuation in its merger. Including Twitter it has barely any revenue, and without having to rent gpu datacenter space, was still losing $10b/year for a not particularly competitive Grok model.
What’s withholding SpaceX from selling more stock with an even shorter time until they can be dumped?
97% of the stock is already issued. 3% goes to new IPO bagholders, shortly later, passive index forced buying from the 3% market. The 97% are the ones with accelerated dump rights, that kick in a percentage shortly after fraudulent accelerated index inclusion date.
training data is as important as the source code here to replicate the end result
this is the nature of this flame war. Perfect replication of the end result, which is extremely opaque in how it works, is not nearly as important as the weights, that you can post train for any domain specific/general improvement with any other dataset. Which is how the authors would improve/change the weights further as well.
This seemed more alarming news: AAI is tracking IPs/proxies/vpn, timezones on requests.
While it is explained as “only to accuse China of evil”, its tracking everyone, and empowers US government crackdowns on anyone either AAI or US doesn’t like.