Breaking Down Trump's AI Order Delays  🔗 **Source:** [Breaking Down Trump's AI Order Delays](https://share.snipd.com/episode/fd0c037d-3dba-4076-ac64-825d3fec54c4) ## Highlights - **SpaceX Is Building Its Own AI Power Stack** - SpaceX is vertically integrating AI infrastructure by buying gas turbines to power XAI data centers. - Jaeden Schafer notes a $2.8B turbine purchase through 2029 and $15B/year Colossus leasing to Anthropic, plus NAACP lawsuits over unpermitted generators. Transcript: Jaeden Schafer The first thing I want to talk about is the SpaceX gas turbines. They've committed $2.8 billion to purchase turbines through 2029, which are going to power XAI's data centers. This is all according to their IPO filings. This basically solves the immediate power shortage, but there's a whole bunch of legal lawsuits coming from the NAACP. They're suing over 27 allegedly unpermitted generators at the Colossus 2 data center in Mississippi. Well, that lawsuit in particular is a little bit controversial because sure, you can sue a company for emissions, but the NAACP typically is launching lawsuits that are defending People of color. And the lawsuit basically says that the data centers are disproportionately harming people of color. And a lot of people are basically saying this is sort of a politicized lawsuit. We'll see where that goes. SpaceX apparently has signed $805 million for a turbine deal in March of this year. And they have a $2 billion mobile turbine agreement in the late april that they just signed both of those are covering equipment all the way through 2029 colossus suit two now powers 46 portable turbines and the naacp lawsuit alleges that 27 of them lack required air permits impose public health risks spacex right now is important to know that they're currently Leasing their Colossus server to Anthropic for $15 billion annually. And they plan additional third party leasing deals to drive revenue. So this is something that not only are they using to train XAI, but when they're not training XAI with it, and I think Colossus two is what they're using for XAI. But Colossus one was kind of the older one. So they have, you know, that's not currently being trained. They're just giving this out to other companies like Anthropic and, you know, who knows, maybe OpenAI or Google or other players. Although Elon Musk and OpenAI don't have the friendliest relationship, but still $15 billion annually to give the Colossus one plant over to Anthropic is, you know, definitely a major Driver of revenue for them. This is all important because SpaceX right is betting that vertically integrating the infrastructure, right? So owning the data center, owning the power, owning the AI units is going to be renting from cloud providers like Anthropic and OpenAI currently do. Google, Microsoft, AWS, all of them are making a killing on this. Meanwhile, SpaceX has basically built their own to train their own models and then they're renting it over to Anthropic for $15 billion a year. But in order to do that, they had to get these up very quickly. Gas turbines were the fastest way to do that without having to, you know, plug into the grid and get all the permits to be grid powered. So that's kind of the big constraint. Now there's lawsuits that might make that all get shut down. ([Time 0:01:04](https://share.snipd.com/snip/767fc0a5-de57-4d5e-9df3-f03be468617a)) - **Google Locks Agents Behind Premium Gemini Tier** - Google is launching agent products like Spark, Halo and Daily Brief but gating them behind a Gemini Ultra $100/month tier. - Jaeden Schafer argues this fragments Google's historical free consumer play and may confuse users with many siloed products (Flow, Gemini, etc.). Transcript: Jaeden Schafer Google is now pitching an AI agent ecosystem. Google IO just happened, which is super exciting. And I'll be talking to someone at Google tomorrow for more news on everything that was announced there. But for this particular AI agent ecosystem, they've announced they have information agents, they have one called Spark, Halo and Daily Brief. But all of those while they sound super useful and interesting are locked behind a $100 a month Gemini Ultra paywall instead of, you know, offering free access or even the $20 a month Access, you got to pay 100 bucks a month. It's pretty, it's hard. I feel like we're seeing Claude made just so much money with their $100 and $200 a month tier. And ChatGPD always had a $200 a month tier, but there's no way I was paying for it mostly because their product wasn't good enough. Actually, I did pay for their $200 a month tier last year at some point to try their agents, their agents weren't good enough, I canceled it. But now that Anthropics Cowork and Claude Code are good enough to do everything I ever wanted, why would I ever cancel and I'm happy to pay the $200 a month. It looks like Google wants to get in on this kind of like higher tier $100 a month thing. But these agents are kind of use agents that have one particular use or a few different particular uses. It's not like Claude Cowork that can literally do anything and everything and do it all well. I even have Claude Cowork going looking at wave files and editing, you know, music projects on on something like logic or going and going to xcode on Apple and building iOS apps like it's Just so insane what it can do. And so for me, I'm thrilled to pay that, but I'm not sure I would go and pay a hundred dollars to use Spark, Halo or Daily Brief. So anyways, I'm excited for what the direction Google's moving in, but they're going to have to build something comparable to, um, and maybe they, maybe they will and they do, but like, That's what I would, something's got to be comparable to cowork. That's the only thing I'm going to be paying that much money for. I don't really care about people's little agents. I could probably build my own with co-work if I wanted to pay that extra money. And it is tricky because, you know, well, $200 a month, I see I get such an insane amount of value out of it, but I'm not going to stack a bunch of $200 a month subscriptions on top of each other. So Google really has to differentiate themselves and make themselves better than co-work. And I will cancel co-work and move to them if they have something better. That's kind of the place they have to be at. So this is definitely reversing Google's kind of historical playbook, right? They've always been releasing consumer products that are free, and they're trying to do that to gain adoption. What these particular agents do, the information agents, they're replacing Google alerts with a 24-7 AI monitoring for market trends, price tracking, weather. Spark is actually going to integrate with Gmail and Docs as kind of a personal assistant. Halo is going to sit as an Android notification layer. Gemini Ultra Pro and Plus subscribers are going to get access starting this summer, but Google gave no timeline for when free user rollout was going to happen even though they said that You know agentic features are going to come quote when the time is right okay whatever the heck that means personally I think this is fragmenting uh their kind of audience a little bit Into their free and paid tier and who gets what there's a bunch of messaging for startups by the way too like poke or wingman and they already have these sort of tech space agents. And they're cheaper and I mean, easier to use than some Google's multi product approaches. Google does have a ton of different products. This is kind of the biggest criticism I'm seeing lobbed at Google after IO, which is just how many different platforms you have to access all of their different tools. You got to use flow if you want to get their Google VO video generator, you got to, I don't know, there's just tons of different platforms to get access to all of their different tools. They keep changing things. They keep depreciating things. Their API is accessible for different things in different places. Anyways, people think it's all a little confusing. It would be nice if I would just go to Gemini.com and could talk to anything and everything from Google. Maybe there's a little tab at the top that just says text, audio, image, video, and done. I don't need to have to go to Google flow and all of the different products that they've kind of rolled out. I get they're kind of testing stuff, but I think they, uh, they're probably confusing the market a little bit and they have so many users. They could probably consolidate this a lot. ([Time 0:03:41](https://share.snipd.com/snip/1e4d6740-38d1-414c-a850-626713ba5457)) - **White House Pauses Prerelease AI Security Order** - The US delayed an AI security executive order citing competitiveness with China and industry pushback. - Jaeden Schafer explains the order would have required pre-release sharing of advanced models 14–90 days before launch and that companies lobbied against mandatory disclosure. Transcript: Jaeden Schafer Okay. Trump has delayed the new AI security executive order. He said that it is all due to US competitiveness concerns that he has right now. Originally, what this was going to do was going to ask AI companies to share advanced models with the government 14 to 90 days before they actually publicly launched them. Right now, they you know, and I think that all came from when Anthropic was like, hey, look, we have this mythos model that can hack everything in the entire world. And so they were like, gee, we're kind of concerned about that. But now they're citing that they're kind of concerned about the competition with China and delaying something for 90 days, for example, so that the US government can look over it and Make sure it's safe. I mean, to be honest, does the US government really know when something is safe? Who knows? I'm sure they have some great technologists there. But these things are really coming out of the top labs. And those top labs are probably the ones that know best what's going on. But that's my opinion. To be fair, the Office of the National Cyber Directory spent the last 18 months developing an AI security framework, but has not published any sort of formal pre release assessment Guidelines. So yeah, they've been working on this for a long time, they don't even have something out. And I think the order that was about to be signed was prompted partially by Anthropics Mythos. And also opening I was like, Hey, look, we have something super dangerous to with chat to 5.5 cyber. And you know, I think they might have been trying to just jump on there. What I will say is that a lot of AI companies have privately lobbied against this kind of mandatory pre-release disclosure, since a lot of those types of provisions have appeared in Different earlier legislation proposals. And I think the public reaction to the delay has basically like no one's really too up in arms about it. So I don't think this is something that's going to get pushed too heavily. ([Time 0:07:42](https://share.snipd.com/snip/0005de9e-5c1f-4085-9486-a5196e1ba405)) - **Host Recommends AIbox To Avoid Model Subscription Overload** - Jaeden Schafer plugs his startup AIbox.ai as a solution to subscription sprawl for accessing many models. - He offers access to 80+ models for $8.99/month and describes mixing Claude, OpenAI, 11 Labs and Google in one chat thread. Transcript: Jaeden Schafer If you are sick of using something like Claude, which doesn't have image generation audio generation or video generation, and there's 100 different models out there to switch between, And you don't want to pay for subscriptions to all the different platforms like Google's new $100 platform, I'd love for you to check out my startup, which is AIbox.ai. I have access to over 80 different AI models, everything from Anthropic, OpenAI, Google, and tons of others for image, audio, text, video generation. It's all on AIbox.ai. It's only $8.99 a month and you get access to over 80 different AI models. You can do it all in the same chat thread. So you can use Claude for helping you with reasoning and then you can pull in OpenAI to generate images in that same thread. ([Time 0:09:20](https://share.snipd.com/snip/e1c22ae7-4e6e-40a3-a277-05d931c74294)) - **Hark Raises Big Bet On Multimodal Assistant Hardware** - Hark raised $700M at a $6B valuation to build a multimodal AI assistant and hardware stack led by Brett Adcock. - Jaeden Schafer highlights Adcock's Figure AI background, a $100M personal investment, NVIDIA B200 GPUs, and plans to ship a multimodal model this summer. Transcript: Jaeden Schafer What are they doing? They're doing an AI personal assistant. That sounds very, I don't know, like everything else. So how are they able to get this much money? What's going on? Well, the reason is because this is coming from Brett Adcock. Now, for those that don't know, Brett Adcock is the CEO and founder of Figure AI, the humanoid robot company. So obviously, he already has a massive company and a lot of, you know, a lot of clout to back this up. What they're betting on specifically with this is that they're building a stack that's designed top to bottom for AI models, software and devices, and they're building all of this Together. Obviously has experience with figures robots. And they're believing that they can succeed in kind of this niche area where Meta's Ray-Ban glasses and a lot of other AI wearables have flopped. This is the this kind of the niche they're going to attack here. Brett actually personally invested about 100 million dollars to launch Hark in kind of the end of last year. The company now has 70 employees. They're running NVIDIA's B200 GPUs in a private data center, which is interesting, right? They're building their own data centers, kind of like what XAI did, and then SpaceX acquired them. So right now, Hark is planning to ship their very first multimodal model this summer. So they have something coming up. I think this is why they were able to go raise a ton of money, right? I mean, he already put $100 million of his own money into this. And if you think about it, there's some really big synergies. If he's doing humanoid robots, a lot of times these humanoid robot companies were making deals with Chad Chibiti or Claude to put the AI models into the robots to make them smarter. If he's going to go and build that himself and make it multimodal so it understands what's around it, basically it's like him spinning out another company that is the brain inside of The robot. So I think this is a really smart move, especially if we think that these humanoid robots are going to scale the way people are predicting right now. For Hark in particular, they're just saying, look, we're making this multimodal model, but they haven't said exactly what form factor of hardware they're going to come out with or What device this is going to be attached to. The funding round is coming from Parkway Venture Capital, Ark Invest, Qualcomm Ventures, Salesforce Ventures. There's about nine other people. One in particular that is talked about a lot is Adcock's former Apple design lead is now going to be serving as the director of design. So they have a really stacked team, a very talented group of people. I'm excited to see what they are able to pull out. ([Time 0:10:30](https://share.snipd.com/snip/8ba671e1-8854-48cc-a11e-20b0b67c9600)) - **Anthropic Faces Profit Now But Compute Risks Ahead** - Anthropic expects a profitable quarter at $10.9B Q2 revenue but warns compute costs may erase profits later this year. - Jaeden Schafer notes Anthropic's $1.25B/month compute commitment to XAI through May 2029 and enterprise traction with Claude. Transcript: Jaeden Schafer Okay, let's talk about Anthropic because they are expecting their very first profitable quarter with $10.9 billion in Q2 revenue. This is insane, especially when you look at OpenAI, which feels like they are miles away from a profitable quarter. For Anthropic, this is more than double the prior quarter. But don't get too excited because Anthropic is actually warning all of their investors that later this year, their ballooning compute costs might actually erase all of the profits By the end of the year. But Anthropic has committed $1.25 billion per month to XAI for compute through May of 2029. This is a deal that is going to increase costs once it ramps in some of the oncoming quarters. Claude has gained, of course, a massive share among enterprise buyers and professionals. And because of this, Anthropic recently launched products targeting small businesses and law firms are trying to diversify a little bit beyond just, you know, some of the early adopters, Some of the early technical users. OpenAI has not yet reported a profitable quarter, and they're also preparing for a September IPO. Anthropic is remaining private with no near-term listing plans. We don't know exactly when Anthropic would IPO, but it seems like SpaceX and OpenAI are really racing for that. ([Time 0:12:47](https://share.snipd.com/snip/cf088087-c60a-4864-9097-78ffb1c69719)) - **Half Of Developers Admit To Not Reviewing AI Code** - Many developers admit to shipping AI-written pull requests without reviewing the code. - Jaeden Schafer references an Anthropic developers conference where about 50% raised hands that they pushed code without reviewing it. Transcript: Jaeden Schafer Number one, there was a big conference that kind of went viral. Anthropic developers were all asked while people using like cloud code for example a bunch of developers at a conference were all asked if they had ever shipped ai written pull requests Without reading the code basically have you pushed code without actually reviewing it coming out of cloud oh 50 of people raised their hand i don't know how to review code and i'm you Know not a developer in any way shape or form i've shipped a whole bunch of websites and apps and tools. I'll usually tell it to go do a security audit, tell it to do an SEO audit, but I don't review it. I just see if it works and I push it. Is that good or bad? You can make your judgment call on that. I'm sure plenty of developers think that sounds terrifying, but I think this is a huge chunk of people right now that are vibe coding. I mean, if you don't really know what code looks like, I don't know how you would review it. So it is interesting that even for, you know, the pros, about 50% are just pushing stuff without reviewing it. ([Time 0:14:00](https://share.snipd.com/snip/aaa86c5f-9ba1-406f-b313-a714165f850a)) - **AI Growth Comes With Massive Burn And Restructuring** - XAI reported burning $6.4B in 2025 while others restructure to fund AI, illustrating heavy capital needs. - Jaeden Schafer pairs that with Meta cutting 8,000 jobs and AMD's $10B commitment to Taiwan AI chip ecosystem. Transcript: Jaeden Schafer Okay, XAI right now is burning and actually burned $6.4 billion in 2025, which is a pretty massive loss. They're a big growing AI company. This is all according to the SpaceX IPO filings that we see. Meta is laying off 8,000 employees to offset their AI investments. We've seen this kind of happening. I even reported on this yesterday, but more people are talking about it today. AMD has committed $10 billion to Taiwan's AI chip ecosystem. Their share has doubled. That to me is particularly interesting because we see a lot of investments coming out of Taiwan. TSMC, you know, building fabs in Arizona is they're kind of worried about the geopolitics of China. But it seems like AMD is doubling down, building and investing more there. ([Time 0:14:53](https://share.snipd.com/snip/b4e00aa5-c7d7-4abe-b59b-84c47b17499c))