Elon Hints At Huge AI Breakthrough (MASSIVE implications)

Added: 2026-05-30

[00:00:00]Ladies and gentlemen, I present to you the future of transportation. Tesla Cyber Cab autonomously driving itself out of Tesla's Texas factory navigating the parking lot. And as Tesla's lead of AI commented after Musk posted this footage, soon it'll be driving itself to Austin City reporting for duty. In other words, straight out of the factory, straight off a production line, straight into the city, immediately beginning operations as an autonomous vehicle. The roads are about to change forever. And as I've warned, if you're a Tesla investor, don't fall prey to shiny object syndrome and ignore what's happening. The mass production of Tesla's dedicated robo taxi, which is almost certainly going to be dramatically overshadowed by the biggest IPO in history, SpaceX. Speaking of, >> there's no human being on the planet that is better at turning electrons into tokens than Elon Musk. Right? Let me say that again because we focus a lot on the launch business. We focus a lot on the communications business, right, which is Starlink, which we all benefit from. But Elon is building the biggest data centers on Earth and eventually in space of anybody. And nobody is better at turning those electrons into tokens.

[00:01:11]That is an extraordinarily valuable place to sit. He just struck a huge deal with Anthropic. I think more deals are to come. And that from my perspective totally changed the game on the SpaceX IPO. I knew that the launch business was great. I knew that uh Starlink was great. But that alone would not justify the valuation. Him building the data centers which are going to power all of AI and him then using those to train up his own AI with incredible team uh that he acquired from Curser I think is a total gamecher for the IPO and puts him in a really good position.

[00:01:46]>> Brad is completely on point here. I'll share more details with the anthropic deal soon, but you'll note Brad is describing the ability for Space Xai with a massive advantage in terms of both their ability to scale AI compute faster than anyone else, but also cheaper and the proofs in the pudding.

[00:02:04]The recent Anthropic deal, one of likely many in the pipeline. Anthropic currently paying SpaceX AAI $1.25 billion per month to access a portion of their compute. Brad points out that SpaceX AI can also use their compute for their own products. The implications to me seem quite obvious. All of the world's serious AI companies will be throwing close to infinite piles of cash in the direction of SpaceX AAI to not only access portions of their training compute because these companies will see an incredible almost immediate return on investment, access more compute, improve your models, increase number of tokens available to consumers, massively increase revenue. these things pay for themselves almost instantly. But I think the next phase is not just going to be companies paying for access to existing compute, but also to help Space XAI build out additional AI infrastructure.

[00:02:55]It might sound absurd for a company to help pay Space XAI to build AI compute and then pay for the privilege of accessing that as well. But if SpaceXI can build much faster, they can. And more affordably, they can. But even if they couldn't, that's not relevant.

[00:03:08]Speed is everything here. Everyone is massively constrained by access to compute. If these companies can throw money at SpaceX AI and save themselves a year or two and bring in billions of dollars, they'd be crazy not to throw essentially unlimited cash in the direction of SpaceX AI. And this doesn't just apply to Earthbased compute, but orbital AI as well. On this note, John posts this today. Elon replies John's post. Space X potential compute business. XAI spent 3 to 4 billion on Colossus 1. Uh it's probably more than that, but whatever. They are leasing it out for 45 billion over 3 years to Anthropic. Blah blah blah. Let's get to the actual point of this. Musk's reply.

[00:03:49]SpaceX has not committed to leasing Colossus for years, although it's possible that may be what happens. This is a 180day lease with 90-day notice mutual cancellation thereafter. The short term was our request, not anthropics. We won't leave them hanging and will provide a reasonable off-ramp, but if compute gets super tight, I said we might need it back at some point.

[00:04:13]Now, this is extremely important, although it's not new information. The S1 made it quite clear. In fact, at the time I explained why this was such an enormous high-risk, highreward piece of leverage or SpaceX AI. Either party can pull out on 90 days notice. But given the enormous constraints on AI compute access globally right now, the odds of Anthropic finding a better deal somewhere else or not needing the comput access is relatively low. It's not zero, but it's relatively low. Whereas the odds of SpaceX AI requiring access to some of that compute for additional training for new models, new projects, etc. is actually relatively high. SpaceX AI has enormous leverage here. Yes, it is high risk cuz anthropic could go, "Well, we're out." But I think the downside here, SpaceX will probably find a use for this computer. I put out the time. This shows what an incredible position of power SpaceX finds themselves in. Unsurprisingly, at least this is new information. M does confirm the short term was our request.

[00:05:09]So, Enthropic weren't wanting a potential out. It was Space Xai. In the future, should hypothetically SpaceX AI acquire a company called Cursor almost immediately after IPO, hypothetically, of course, or perhaps envision a new product or an opportunity or maybe even hypothetically again of course totally not going to happen. Totally not inevitable. Maybe SpaceX were to acquire Tesla. It would certainly be handy to have an enormous amount of compute capacity to for example train new models or improve existing models whether it be an AI coding model agent autonomous software humanoid robots etc. It's also possible that another AI company out there wishes to outbid anthropic and say bro we'll pay $2.5 billion a month. What are you going to do? What are you going to do is well if you're SpaceX AI say Anthropic so we got a better offer. We don't want to [ __ ] you guys over. So, we are giving you the opportunity to match or exceed that. Otherwise, see you later. 90 days. Bye. Again, extremely strong signs of enormous leverage in SpaceX AI's favor. Some important posts today from Alon Musk. Extremely nerdy.

[00:06:12]And if you're an autistic AI engineer, you'll probably bust a few nuts. Pause and read the full thing. I'm not even going to bother cuz everyone's eyes will glaze over and go, "What the [ __ ] GB300's and they were." I will get to the bottom line and the point here. Musk post today that SpaceX has almost finished writing version one of an in-house AI training stack. The end result of which is a potential speed improvement for AI training versus what most of the industry uses of more than an order of magnitude. So the bottom line here this system could make training huge AI models. The kind of models that could bring in tens hundreds of billions of dollars of usage from users. the most useful large models could be trained 10 times faster or more. He wasn't done adding in a follow-up post. Next, we'll be writing the inference stack in C for simultaneous high-speed reinforcement learning across a large block of GB300s.

[00:07:07]And a third post clarifying note, I posted this to encourage those who enjoy getting incredible performance out of hardware to join SpaceX, which explains the extremely nerdy jargon. Bottom line here though is a very substantial improvement in training large AI models.

[00:07:25]And when I say very substantial, I mean abs are [ __ ] lutely gigantic. More than an order of magnitude. If they pull this off, holy [ __ ] [ __ ] A user on X repost Musk's recruiting effort and writes, "There's a ton of headroom to better optimize AI training and inference." Elon responds, "Yes," adding, "It's not that we've discovered some magic bullet, but rather than Jax, which is the industry convention, the standard at the moment, or at least the open- source version of it is mostly optimized for small to mediumiz training runs on Google TPUs, which is Google's own AI hardware, whereas we need massive training runs on Nvidia GPUs. This is basically what the rest of the industry is using. more. A user writes, "They made an in-house AI training stack from scratch in C as a hobby project." Musk very candid in his response, "It's essential or we simply cannot be competitive." This follows a post 2 days ago where Musk writes, "I just left the Space XAI office at 2:45 a.m. and people were still going. I think we might be close to a significant breakthrough in training." And as we've since learned, that significant breakthrough is roughly an order of magnitude improvement in the training of large AI models. And for those of you unaware, while we're on the subject of large AI models that accurately describes full self-driving, as I said in the intro, I have a strong suspicion that many investors will be suffering from shining object syndrome.

[00:08:52]What's happening with Space Xai right now leading up to and then after the IPO. The amount of news updates, leaps forward, the hype justified, I'll add, is going to be next [ __ ] level.

[00:09:04]Tesla's CyberCap now in production is about to change transportation forever and completely transform the financials of Tesla's business. Make sure you're paying attention. Want more content, early access, bunch of perks? Click the links in the pin comment. AG1 has been part of my daily health protocol for almost half a decade. It's packed full of vitamins and minerals and helps me fill in nutritional gaps while supporting energy, digestion, and immune function. It also has prebiotics and probiotics to promote gut health. I take my health very seriously and so should you. Try AJ1 today by visiting drinkag1.com/smr or click the link in the pin comment and enjoy a free welcome kit with vitamin D3 plus K2 and AG1 travel packs. That's drinkag1.com/smr.