Interesting that the `brew-rs` experiment has concluded and didn't find much of a performance increase. I suppose that is expected though with a lot of the bottleneck being network IO?
These are terabyte sized files (realistically a multi hour transfer) that you're unlikely to have access to in the first place. Every organization has exfiltration checks these days. You may succeed but you'll want to be on a plane to a non-extradition country no more than hours after you kick off the transfer.
I assume they’re encrypted/DRM’ed when deployed on inference hardware, so only core researchers/sec admins would potentially have some access to unprotected weights, and they are far too well paid to risk it leaking the model
Incentives matter on the average, but people are too unpredictable for categorical statements like that. They can always have other reasons beyond personal gain to leak secrets.
There was no shortage of spies and defectors leaking American nuclear secrets to the USSR during the Cold War.
Newer NVidia cards (H100 and up) support both in-memory model encryption and ‘trusted’ execution environment/remote attestation, not sure how widely used in frontier model deployments, but at least vendor claimed perf overhead is ‘3%’ [0]
The employees are hoping to become very very rich after the IPO and after they are allowed to sell the shares given to them - risking a likely multi-million dollar pay back to leak a model that will be superseded by publicly available models in a couple of years is not a likely decision.
Apologies for the naivety, but, why is SpaceX valued so high? Starlink? Are rockets really a lucrative business? Don’t get me wrong, being able to send objects up into orbit is cool, but is it $1.8T cool?
No. Space is not lucrative or profitable. The SpaceX profit story rests entirely on Starlink. Starlink has a plausible moat servicing ships at sea and extreme remote areas. The big problem for Starlink is that they are trying to grow into a shrinking TAM, as terrestrial wireless expands with ever cheaper equipment ever farther into the countryside that Starlink is counting on for their TAM.
Elon's visions border on self parody. If I told you that humanoid robots were going to be digging tunnels for the Boring Company you'd have to stop and think if I was pulling your leg.
To be clear, I don't support SpaceX specifically, but the amount of resources available to us from beyond our planet are quite literally infinite, only bounded by our ability to move fast enough to get it.
Comets that routinely pass by our planet have rare-earth metals in quantities that we don't even have on the planet at all. Hell, that's where our rare earth metals came from in the first place. Getting access to 100 million tonnes of platinum could totally change how we use the metal, right now it's most effective use is probably within catalytic converters to reduce emissions from cars.
Helium-3 and Deuterium in high quantities can be used as clean fusion fuel, basically clean atomic energy.
I struggle to see how these can't be lucrative in the long term.
I can't improve on how unlikely it is that any of that happens. Space is for exploration and the advancement of science, and to a certain extent engineering, if you don't mind the inefficiency of obtaining those advancements in engineering.
How many decades ago were people hyping space manufacturing? Where are the space factories? Where are the profits?
Because of the Musk reality distortion field. The claim is that all data centers will move into space, and that SpaceX will completely own that market.
The datacenter thing is mostly just a meme that billionaires say because it makes them feel smart and gets them media attention, it doesn't seem to move stock significantly.
The actual distortion field is around Starlink. Which is the main product and the only one that's (nominally) profitable. It's the one all the hype centers around. xAI is barely even notable in the AI space.
This also makes it possible to judge the size of the distortion field, as Starlink is just an ISP, for which we have accurate valuations. And for what it concerns shareholders, a strictly worse one than a conventional ISP. Space infra is much more expenive than putting some glass in the ground, once.
Comcast is a behemoth of a company doing far more than just ISP. Worth a "mere" $90 billion. Charter Communications is a similarly sized "pure" telecom. Worth $20 billion.
Both of the above ISP companies have roughly 30 million subscribers. Starlink has 10 million. Yet they want $2 trillion at IPO.
A 20x to 100x overvaluation. And what do you get beyond an ISP?
* A private aerospace company that's not doing notably better than the space divisions of old aerospace. (Remember: Starlink is already accounted for so doesn't count here)
* An AI company that has so little demand it's currently handing a bunch of compute to Anthropic for such a deep discount the latter has claimed to become profitable.
* Twitter. Which is worth either $33b if you count Elon's internal buyout valuation, or $10b if you count realistic valuations.
While there is some hype around "The future of space!", the reality is that the long term growth for that is fairly dead in the current geopolitical climate. Nobody's saying it out loud yet but US Aerospace is being replaced. Fewer and fewer US launches will be bought. The EU is even building their own Starlink equivalent.
Starship is going to make whole entire industries viable that were not viable previously. It might even take a significant chunk of air freight which is going to be a big deal with rising oil prices.
Is that supposed to be a joke? There is no plausible scenario where SpaceX gets any significant fraction of the air freight market. Even under the most optimistic scenario the costs for suborbital launch are much higher than regular airplanes.
In a few decades there might be a small market for carrying passengers long distances really fast. Initially for the military to insert special ops troops in a crisis, and eventually maybe for wealthy consumers after safety improves.
Starship in its current incomplete form (v3 fully expendable ship and booster) already has the lowest cost to orbit in $/kg of any launch vehicle ever. It's around $400/kg to orbit fully expendable.
Add in booster reuse, which SpaceX has already demonstrated on test flight 9, and the cost to orbit drops to $200/kg.
A fully reusable Starship has a launch cost of around $75m - $90m and the last V3 launch managed 44 tonnes of payload on a sub-orbital flight of not even 200km (Starlink satellites have an orbit of around 550km). That's an optimistic launch cost of $1.700/kg for a rather meaningless altitude and assuming a fully reusable Starship that doesn't keep blowing up.
I have no idea where you pulled your $400/kg number from, but it's complete and utter nonsense. To be economical at all, Starship needs to reach its target capacity of 100 tonnes to orbit, which is simply never going to happen. But even if it somehow does, it's physically impossible for Starship to ever make it further than the moon, at extreme costs, due to the refuelling requirements and fuel boil-off in orbit.
> A fully reusable Starship has a launch cost of around $75m - $90m
No, that's the Starship build cost, i.e. the cost of an expendable Starship. A fully reusable Starship currently does not exist, but reusable launch cost be around $5m/launch (amortized).
> the last V3 launch managed 44 tonnes of payload
Intentional, Starship wasn't fully loaded.
> on a sub-orbital flight
Intentional, test flights are sub-orbital.
> of not even 200km
Intentional, done to target the landing site in the Indian Ocean.
> That's an optimistic launch cost of $1.700/kg
You can do basic math, but you are intentionally using incorrect numbers. Garbage in, garbage out.
> I have no idea where you pulled your $400/kg number from
$80 million / 200 tons = $400/kg to orbit (fully expendable).
This number is already exaggerated, the booster is already proven to be reusable.
If the current Starship is mass produced, this improves to $50 million / 200 tons = $250/kg to orbit (fully expendable).
> To be economical at all, Starship needs to reach its target capacity of 100 tonnes to orbit
You do realize the Starship + Booster stack weighs 5,000 tons, and that a 100 ton payload is only 2% of the rocket mass? And that 2% is an achievable fraction, both Falcon 9 and Falcon Heavy have a payload fraction >4%. The Starship upper stage alone weighs 1,600 tons.
> refueling requirements
In terms of problem difficulty, orbital refueling is a minor engineering challenge to solve.
> fuel boil-off in orbit
I hope you are being facetious at this point. How do you think LNG is transported around the world? You realize this problem was solved decades ago?
That $75-90 million figure is the current cost to launch Starship. It is also the approximate cost to build Starship.
Both of these things are true.
$5 million per launch is an Elon Musk wet dream that's never going to happen. You know like Tesla FSD?
Sure 44 tons to orbit was intentional. You can report back when they intentionally launch 100 tons to orbit. Until then it's just another worthless Musk promise.
The fact that the Falcon 9 and heavy can launch more than 2% of their mass into orbit has no bearing on Starship's capability to do the same.
You're comparing apples with oranges.
Neither of those rockets is fully reusable like Starship. They don't have to carry a return supply of fuel for landing, or a heat shield, landing legs, aerodynamic wings and everything else that's required for full reusability.
And refuelling in orbit is only a minor engineering challenge? That's just hilarious! Try reading something other than Musk's X feed.
I hope you are being facetious at this point.
Please don't tell me that you think Starship is fuelled with LNG propellent.
Fuel boil-off in orbit is real, just like it is on the ground. Except in orbit you don't have a big cryogenically cooled tank of propellant to top it up with sitting only 100m away.
Have you ever watched any rocket launch ever? Serious question.
Ever wondered what those huge clouds of vapour are? You know, the ones streaming out of the rocket while it's sitting on the pad ready for launch?
> The datacenter thing is mostly just a meme that billionaires say because it makes them feel smart and gets them media attention, it doesn't seem to move stock significantly.
A significant portion of their valuation is based on this. The spacex private stock price moved significantly based on this data center narrative.
> And for what it concerns shareholders, a strictly worse one than a conventional ISP.
This is ignorance. There is absolutely zero meaningful competition to Starlink in the maritime, aviation, and remote internet markets. 150mbps down with <80ms latency isn’t impressive in a city but it’s mind blowing on an airplane 1000 miles from land.
> The EU is even building their own Starlink equivalent.
No they aren’t. The only somewhat credible competitor so far is Amazon Kuiper(Leo) and they are still nascent.
> There is absolutely zero meaningful competition to Starlink in the maritime, aviation, and remote internet markets.
There are roughly 100,000 ships at sea. There are roughly 15,000 planes in the sky.
The remote internet markets are remote because either A) exceedingly few people live there, or B) exceedingly poor people live there. (And usually, both at the same time)
This just isn't a big market. That's why the telecom giants haven't bothered. To justify a trillion dollar valuation you're gonna need a billion users. SpaceX would be better off putting fiber into the ground in Africa.
> There are roughly 100,000 ships at sea. There are roughly 15,000 planes in the sky.
That’s pretty great for price range of $500-$2000/mo.
> The remote internet markets are remote because either A) exceedingly few people live there, or B) exceedingly poor people live there. (And usually, both at the same time)
This is incorrect. Its usually just places people live that are difficult to reach with good telecom infrastructure because of lower income and/or lack of a good business infrastructure for internet. This includes the US that was frequently over capacity on Starlink in essentially the entire southeastern US for over a year when I was trying to get it there early on.
I suspect you don’t realize that “cell phone coverage” != “good internet”. You usually only need to go about 10-20 miles out of town before all fiber/DSL/cable evaporates. The cell coverage in an area like that isn’t the good kind you get in the city. You’ll get 5-10mbps down and brutal data caps.
Starlink is popular in the Philippines, Indonesia, New Zealand, Australia, and even Europe for a reason.
> That's why the telecom giants haven't bothered.
Telecom giants can’t bother because their costs don’t scale the right way due to tech limitations. A LEO sat can provide coverage in remote/sparse places AND coverage in denser money making places all in one orbit. A fiber on the other hand can’t serve the Aleutian chain and then the Congo 15 minutes later.
> To justify a trillion dollar valuation you're gonna need a billion users.
No you don’t. I think you ignored the part of my message about a significant portion of the valuation being from datacenters in space (a yet unproven market).
> SpaceX would be better off putting fiber into the ground in Africa.
No they wouldn’t. This doesn’t work. I recommend you look up why this has failed every time so far and why Africa is served by undersea cables to coastal cities.
Fiber is terrible for places with poor infrastructure. If there aren’t people adequately maintaining a power grid, there sure as hell aren’t people to maintain even more delicate fiber and required last mile infrastructure.
IRIS is a 290 sat constellation that doesn’t have any proven sats in orbit or a user terminal. The capacity even if they deliver what they claim on paper is not a Starlink equivalent.
The tyranny of being in MEO/LEO is that your sats are not overhead at least half of the time.
Lower capacity, higher latency, and worse coverage makes it more of a viasat improvement than anything like a starlink equivalent.
They fumbled by not trying to provide global coverage.
I find it amusing to read comments like these, because they remind me of the massive awareness gap between people who understand SpaceX's product line, and those who don't.
In your world, you only see and interpret SpaceX's existing products. You then see SpaceX's eye-watering valuation, and then are confused where this comes from.
Meanwhile, people who understand SpaceX's product line, and the implications these products in five or ten years, can analyze the situation more accurately.
I can tell you are in the unaware group, since you don't mention nor analyze two of SpaceX's world-changing products (Starship and Starlink Mobile).
Rocket launch ex-Starlink is a small $N-billion market; a few dozen flights at $50M per flight. Starship is revolutionary and I can easily believe that it will expand the market by a remarkable order of magnitude. Multiple tens of billions. How does that justify a valuation over $1T?
Starlink Mobile is more significant, but it's still unlikely to double Starlink revenue -- most mobile traffic will always be transited by local cell phone towers.
P.S. I think somebody is going to make a lot of money from Starship. The money in space is not from launch but from the services it enables. Starlink >> Falcon9. But I don't think SpaceX is going to be the ones to find the next Starlink. It's much more likely to be a third party who launches on multiple providers to keep costs down.
I'll mention the ones no one else is talking about.
The Golden Dome buildout will deliver $200-300 billion in revenue to SpaceX through 2040 https://en.wikipedia.org/wiki/Golden_Dome_(missile_defense_s... . Golden Dome is only viable with low-cost Starship, and SpaceX will build the satellites housing the radars, IR detectors, interceptors, and backbone communications network. The interceptors themselves will likely be built by existing players i.e. Lockheed Martin and Raytheon. SpaceX revenue here is astronomical, potentially $500+ billion if a full buildout occurs.
Starlink Mobile is atrociously underestimated, with Starlink V3/V4 satellites and the $17 billion Echostar spectrum, Starlink will deliver 4G speeds direct to cellphone for thousands of customers per cell across the globe. It will never match the bandwidth throughput of terrestrial towers but will be extremely cost competitive in rural areas. A single cellular tower costs ~$250k to build and tens of thousands per year to maintain. It will be far cheaper for mobile operators to partner with SpaceX than do their own costly buildout. Assuming quick adoption, revenue will be $50-100 billion per year by 2040, with high margins.
Not to mention Starship, which in its incomplete form of V3 already has the lowest cost-to-orbit of any launch vehicle ever, will usher in a wave of space exploration. The moat is huge, Starship is 15 years ahead of the competition.
Just to be very clear about Starship: We have a very limited amount of payload we are sending up in space every year.
The biggest jump in payload is starlink itself. Starlink though doesn't scale very well. V2 can only handle a certain amount of customers and has only a lifetime of 5 years.
Space-X has to build Starship to even being able to send v3 up to increase the margin of this setup. But even then, every 5 years that thing has to be replaced and new build.
Every mobile tower, fiber cable etc. underground has a lot higher lifetime than that.
Starlink also has the issue of latency handover. Every few minutes you have to do a handover which leads to package loss. I can't do a Teams Call through Starlink fyi.
And Starlink already exists and is relativly affordable despite that, they only have 9-10 Million customers and they had to increase the price.
And while all of this 'magic no one gets' is happening, Starship hasn't profen non leo orbit with proper payload AND reusability. Without reusability, they will not get the costs down that much anymore. Its already relativly cheap.
And in parallel all of this 'trillion dollar future margin magic' gets opposition by other companies like eutelsat and Amazon.
Ah yes the world changing product of starlink mobile. Which doesn't get booked in the USA, is slow and needs a lot of energy. Whatever you think this is, 500km mobile range is 500km and this on a planet were normal people already have a very very well working mobile setup for at least 10 years by now.
Is space-x some kind of business gap? yes sure. Will they make billions with this? Depending on other companes, yeah sure. Is it a trillion dollar business? No
Yes yes i'm fully unaware of this.
Btw. Musk def sells you the story of Mars and dyson sphere and stuff to keep the magic but while he does all of this, he rents out colossus 1 and 2 to his competitors because he is unable to sell his OWN AI product.
"You just don't understand bro" has been the trite handwaving of criticism for a over decade now. It already wore out when the bitcoin bros kept saying it to all their critics.
> I can tell you are in the unaware group, since you don't mention nor analyze two of SpaceX's world-changing products (Starship and Starlink Mobile).
Just because I did not mention Starship by name does not mean it's not in the reply.
And Starlink Mobile is still an ISP. "It's worth a trillion dollars because it's mobile!" Haven't heard that one since the Dotcom bubble.
But more to the point:
> Meanwhile, people who understand SpaceX's product line, and the implications these products in five or ten years, can analyze the situation more accurately.
They are looking 10 years forwards. I am looking 10 years back.
This exact same "just you wait, in 5 years there'll be a miracle technology that generates infinite profit" rhetoric has been used for those 10 years.
Still waiting on the miracle self-driving that was supposed to justify Tesla's $1.6 trillion.
> They are looking 10 years forwards. I am looking 10 years back.
And that's your flaw. Companies are priced by looking ahead and projecting future revenue and earnings, not by looking 10 years into past. Your analysis is fundamentally flawed for this reason. Neither of SpaceX's major future revenue drivers (Starship and Starlink) existed a decade ago. This explains your confusion regarding SpaceX's valuation.
This sounds amazing until something needs replacement. Until data centers on earth has a 99.99% (or higher) level of autonomous operation with very minimal requirements to maintenance and part replacements, they're not sending anything into orbit...
Why tungsten? In terms of thermal conductivity, It’s way worse than silver and copper and on par with good aluminium alloys. Those are cheaper and much lighter (so again much cheaper to put into orbit).
could it be that those people don't actually believe him and just appreciate the genuine shit-show he puts up enabling them to laugh at people who get sad about what he says, and their eagerness to believe
Where did they say that all data centers will move into space? I thought the claim was that it's going to be more and more feasible and profitable to have DC's in space.
His Scifi ideas are probably a min of 100 years too early.
And while he doesn't even need his own compute (renting out colossus 1 and 2), he thinks we will send server racks full of expensive hardware into space in no time.
Why?
Because his Space-X Trillion evaluation doesn't make sense if he doens't has payload for Starship.
So how much payload do we send to space? Actually not that much, starlink itself is the biggest change by far. So he builds Starship which he needs for starlink v3 but what then? Yeah Datacenter in space...
I doubt it'll be traditional server racks. This will be a paradigm shift, for sure. I think he's probably ahead of his time here, but that's usually how it goes.
One must also consider the proximity between musk and the trump administration, the market pricing this proximity is the market pricing power, access and aligning its interest with the blatant collusion between political power and business in the US.
SpaceX already dominates the global industry, there's barely anything left to grow into at current $/kg, but of course reducing $/kg by 10x means 10x more cargo to launch
According to SpaceX's own documentation more than 80% of their value comes from xAI and their data centers. Starlink is where they are making the money, but they are pouring it into AI.
Big picture: Nice-sounding economic theories claim that stock market valuations are rational, but those theories are mostly bullshit. As soon as the actual humans in the real-world stock market get excited, or scared, or otherwise emotional, they mostly stop caring about all that stupid boring gotta-do-math "rational" stuff.
Yes, eventually, the humans have to sober up, and stock market valuations return to approximately what the economic theories say they should be. But the dangers of betting on that "eventually" are very well known: https://quoteinvestigator.com/2011/08/09/remain-solvent/
Starship makes space access 1000x cheaper than before, with cost-to-orbit under $10 / kilogram. Also much larger payloads become possible. That's an insane economic unlock, because space contains unlimited amounts of resources and energy. Space-based manufacturing, building space hotels, mining asteroids etc. becomes viable. Larger satellites and probes for commercial or scientific use becomes possible.
All this might not make sense in a standard business sense. Real profits might be decades away, who knows. Anyway, people are willing to throw their money at it, because they think it's important.
If they succeed in the long term, it'll easily be the most valuable company on Earth.
The earth is finite, and space, for all intents and purposes, is not, and expansion into it would thus be required to sustain any super linear growth of the economy.
Well, and rockets are cool. Perhaps people would much rather invest in something with the veneer of furthering space exploration (and the promise of infinite riches) than buy into some crypto blockchain startup.
And, its not as if other current valuations are sane.
SpaceX isn't opening space to everyone. They're are preparing for the select few to be able to escape once the earth is no longer sustainable due to their own efforts.
Actually given that the first colonists on Mars will live pretty miserable lives before dying early of radiation poisoning Musk and Co are trying to recruit other people to move there.
Musk, Thiel, Bezos etc. none of these guys have ever said they want to move there.
No they aren't. They're trying to design robots that can build the habit for them before they arrive and pamper them once they are there. Maybe they'll send a few human guinea pigs first to work out the kinks, but they ain't trying to go to Mars, or anywhere, for the good of the human race as a whole.
Musk has repeatedly talked about how he wants to go to Mars.
Opus 4.8 is the first tangible improvement since Opus 4.5. And it doesn't seem to have the personality problems of the last release -- I've been enjoying using it.
That'll populate over the next couple weeks -- those are the live games on the spectate tab which take a while to generate statistically worthwhile data. I'm curious how it does. From using it all day, I can say Opus 4.8 is my new favorite model, hands down.
What would happen if some regular Joe, completely unrelated to any other user, watched for bets that could signal insider information and made significant profit themselves, too?
That's called "copy trading", which has been around for decades.
If there's no connection between you and the trade you're copying, there's nothing you can be charged with. Normally there's a natural latency between the "signal" trade (i.e. the trade to copy), and the copy trade, which obviously can alter the profitability. This latency can range from sub-seconds if there's some public ledger, to days/weeks/months if the info is due to disclosure. Obviously when it comes to crypto and public ledgers, we're on the former.
But as soon as you place such trades based on insider trading, that's insider trading.
fair number of people do this, bunch of free tools to filter by this criteria (brand new account, more than $xx,xxx size), vulnerable to spoofing obviously
This is a trade that is already done by trading businesses. It doesn’t even need to be this explicit; you could for instance dispose one side of a trade to reduce exposure to insiders.
This is where a lot of the crypto scammers on x have moved, selling ai/bots to do this. It seems like the odds slip significantly after these large bets are filled.
Not sure if Polymarket does the same. Also not sure if I really trust these people to be the arbiter of morality and adhere to their own rules when it doesn't benefit them.
> A registered entity shall not list for trading or accept for clearing on or through the registered entity any of the following:
> (1) An agreement, contract, transaction, or swap based upon an excluded commodity, as defined in Section 1a(19)(iv) of the Act, that *involves*, relates to, or references terrorism, *assassination*, war, gaming, or an activity that is unlawful under any State or Federal law.
Polymarket is now two separate entities. Last year they were banned from the US and stopped taking on new US customers and made their site withdrawal only for anyone based in the US. You could bypass it with a VPN, but they did enough to at least look like they were complying.
They've now created a separate company that I believe is called Polymarket US which has it's own version of the site/app which they are using to re-enter the US market. This switched from invite only beta to open sign ups a few months ago.
Polymarket US is mainly just sports betting with a small number of crypto, commodities, and political election based markets, although they have said they plan to expand their market options. Their US based site almost certainly doesn't violate those rules. The version that is used outside the US is the one that has markets that violate that, but since they're not in the US, there's not much that can be done.
So if you kidnap them, hold them in a bunker for a month, then release them, it will pay out. That's probably a positive thing for the world somehow, right?
Sounds like this could easily be worked around by placing a bet on something that will definitely happen as a result of a death, i.e. "bet $X on whether Iran will close the strait".
reply