Loading summary
Transcript1 lines
- [00:00]
A
The EU has just hit pause on its own AI act and the entire compliance map for 2027 is now up in the air. Before that, Google's Athletica just cracked math problems that no human had solved. OpenAI's CFO is reportedly at war with Sam Altman. Over $660 billion in compute spending. And big tech earnings week kicks off tomorrow. So there's a whole bunch of questions that are going to be hanging around over AI and ROI that we're going to be covering. Like, let's get into it. So first I wanted to Talk about Google DeepMind. They've just dropped Athletica this week. This is an autonomous math agent and it's actually just built on top of Gemini 3 deepthink, and the results that it has been producing are really wild. So on the first proof challenge, which is basically a benchmark of unsolved or really complex kind of novel math problems, Athletica actually produced solutions that according to human expert evaluators were, were graded as, quote, publishable after minor revisions. Meaning, like it basically got it there. There might been a couple tweaks, but I mean it solved these really complex math things to a high level, which was basically ready to get published. And it actually did this on 6 out of 10 questions that it was given. It also scored above 91.9% on the IMO proof benchmark. Anjay Miha at A16Z called it the moment, quote, AI math went from playing the game to writing the rules. I think that's kind of a fair framing because up until this point, with all of the different AI models, proving these kind of novel theories is one of the hardest possible tests for these AI agents. And Athletica just did it. Sony is also coming out swinging. So they have something called Sony AI and they have a project called ace and they just published on this, their project that was just published on the COVID of Nature, one of the top journals for science this month. And, and it is the first known autonomous robot that beats elite and professional level human table tennis players in real matches. So there's a whole bunch of, you know, different domains of these advancements that we're seeing in science and technology. But I think the same point is happening. AI is now hitting professional human performance in tasks that require, you know, sub 200 millisecond planning loops. Ping pong. I mean, I am, I'm no expert, but I do love ping pong. And that is a game that if you see the pros play, the ball is flying so fast, is moving so quickly, and the Precision has to be insanely high. Now I think for a lot of people it's just this like pure muscle memory. We're not thinking, we're not computing the angle of the paddle, the spin, how we hit it, like it's all computing in our brain. But this is happening with muscle memory, with years and years of training and the fact that we can just give a robot the ability, Sony's able to give a robot the ability to, to beat a human at this is incredible. And because obviously like ping pong, whatever, I mean, I don't mean to belittle all the ping pong players of the world. This is not, you know, something that has a big impact on society. Ping pong, the ping pong industry. But man, I can just hear all the ping pong, you know, pros being deeply offended by that statement. But this is really incredible when you think of the advancements in robotics. There's so many industries where these robots are going to be able to play a huge role. And I think one of the biggest ones that ever everyone is thinking about is humanoid robots that are going to be able to do so many different tasks maybe. I know this sounds crazy to some people, but like help your aging parents maybe feel more empowered to live in their home without having to go into like a retirement home. Maybe it's able to reach higher things off the, you know, off the shelf for them and help them get up when they fall down. Now I'm sure this sounds like dystopian and whatever. I mean, you know, putting that all to the side, I think there are real use cases where these can really help people in industries where, you know, just wouldn't. It's hard to get that help otherwise. So all of this to say I'm really impressed with what Sony AI has done there and I think we're going to see a lot more of these really incredible advancements coming quickly. So obviously that's exciting. But like just in the last 24 hours, you can also add to that the fact that Apple has a paper that came out at ICLR in Rio, which just wrapped up yesterday, and they have a new paper on privacy preserving on device learning. This is basically a new approach to time test compute. That's really incredible. You can, you have Google's Turboquant, which they showcased at ICLR which cuts inference memory by 6x. I think these are some of the stories and not a lot of people are covering a lot of these frontier model releases. You know, they get the big headlines like oh my gosh, the new version of ChatGPT or the new version of Claude just came out. But I think the next 12 months of the actual economic impact of AI, they're being decided right now in a lot of these kind of efficiency papers and on a lot of these kind of, like, benchmarks. And I think the labs that are getting this right are going to print so much money. So you can kind of see who's putting out a lot of this research. I mean, people like Apple, Sony, Google are doing some really incredible things. But the thing is, the labs that do not get this right, they're going to be paying $660 billion in compute bills for nothing. Which is, as it turns out, exactly what is tearing OpenAI apart this week. So Fortune just broke a piece this morning, which is kind of building on some reporting from the information in the Wall Street Journal. And the picture inside of OpenAI right now is pretty rough. The CFO, Sarah Fryer, reportedly told colleagues earlier this year that she doesn't believe the company is ready to go public in 2026. She's specifically worried about $660 billion in projected AI compute commitments and whether the revenue ramp can support those contracts. Sam Altman is pushing for a Q4 IPO, and Fryer is publicly opposed. She's the CFO inside of OpenAI. So we have some serious headbutting from the top people inside of OpenAI, and I think it gets worse than a disagreement. So the information actually reported that Fryer stopped reporting directly to Sam Al last August and now reports to Fiji Simo, who runs the applications business. And Sam Altman has reportedly been excluding Fryer from certain financial conversations. I think that's definitely not very normal. CEO, cfo, you know, I mean, there's friction there for sure. Right. But what I do think is interesting is that this is a bit of a structural break. Fryer and Altman put out a joint statement calling the report, quote, ridiculous. Which in, you know, CFO speak means it's basically true. And Brad Gerstner over at Altimeter, who is normally very bullish, posted on X that this week that the biggest risk to OpenAI is, quote, the gap between contracted compute and recognized revenue, not the IPO timing itself. Gerstner essentially is backing Friar's read on the situation of this. OpenAI also missed multiple internal Q1 revenue targets, I think partially because Anthropic took enterprise coding share. So I think a lot of the deeper context is OpenAI, you know, signed up for this $660 billion in computer, they missed their Q1 revenue and their CEO, you know, their CFO doesn't think that they should go public yet. Whatever you think of Sam Altman, this is definitely a hard problem to solve in eight months, which is the timeline for when he would like to IPO the next big story we have is that the big tech earnings are starting tomorrow. And I think this is one of the most consequential earning weeks that we have had in years. Microsoft reports their Q3 fiscal and they're doing that on Wednesday after the close. Alphabet, Meta, Amazon, they're all reporting this week to Apple lands three Thursday together. I think these are basically the four companies that are committing about $700 billion in capex for this year and almost all of it is in AI infrastructure. So the question that I think I'm going to be asking, but I think you could probably say every analyst on Wall street is asking. Dan Herbace, check over at Ramsey Theory Group put out a note today saying that Wall Street's AI boom is heading into a quote, ROI reckoning. And I think he's not the only one that's saying this right Now. And Alphabet's Q1 EPS is projected to drop 6.4% year over year despite revenue growing to 106.9 billion. Because the 175 to 185 billion in AI capex line is grinding their margins. Meta is going to be very defensive. I think the same math and is kind of hitting their company. They spent $72 billion last year and they're committing up to 115 billion this year. I think the market is going to want to see actual revenue attribution to AI workloads. And with Meta it's a little bit trickier because yeah, they can say like I've heard Zuckerberg in the past be like, yeah, well we've been using AI for years, like it helps our algorithm, but they're spending a lot of money on LLMs and just building AI tools, general purpose LLMs, which yeah, they kind of embed into their like messaging apps. Like you can chat with them, but it's just not getting like the insane usage. At least what I've been seeing from users as Even things like ChatGPT. No one really believes that Meta AI is as good. And so we're not really seeing it for kind of like agentic productivity or a lot of the other revenue drivers that get people to pay a lot of money for them and it's all free built in. So anyways, this is something that meta is going to have to defend For Microsoft specifically, I think this is the biggest earnings call of the year. And it lands two days after they got, you know, rewritten out of OpenAI exclusively. And their kind of exclusive clause, or OpenAI, was just on Microsoft Azure. Azure's growth has been guided at 37 to 38% in, you know, for a long time. And if this decelerates, this is going to put a lot of pressure on their stock and on their company and kind of their direction overall. What's interesting is we actually have two different takes coming from Wall street right now. We have Truist, which reiterated a buy. They said it's worth 675. That's kind of their price target this week. Meanwhile, Oppenheimer cut to 515. Both of these companies cannot be right. And so Wednesday is going to tell us which one is accurate. Okay, one more thing that I had to cover before the deep dive. The Washington Post Pranshu Verma broke some really interesting news that said more than 600 Google employees, including senior Google DeepMind researchers, signed a letter to Sundar Pichai asking him to refuse a classified AI deal with the Pentagon. Inside the petition it says there is, quote, no way to ensure that our tools will not cause serious harm or violate individual freedoms. And I mean, this is kind of the same thing that a lot of the people over at Anthropic were, you know, were kind of sane and why they were trying to get Anthropic out of their deal with the Department of War. This is just two months ago. And when that happened, Anthropic was dropped from the Pentagon deal because they basically refused to sign, quote, any lawful use clause. So they wanted, you know, they wanted to be able to dictate what kind of guardrails the Pentagon could use their AI model for. And the Pentagon obviously didn't want to sign that. So Anthropic actually got labeled a supply chain risk for using that. They're now suing the DOD over that designation. Google is now kind of doing the opposite. They signed the contract that Anthropic walked away from, and they essentially allowed Gemini to be used for, quote, any lawful government purpose. Gemini 3.1 is already on the Gen AI MIL platform, which is about 3 million Pentagon staff. So Google's own DeepMind people are basically watching the company take the deal that Anthropic refused. I think the reason why this matters is kind of beyond the obvious internal labor story. And it gets right at a moment where this same week the EU is finalizing its own answer to AI governments It's always interesting for me to see how Brussels treats AI versus Washington. We're going to get into all of that, but before we do, I wanted to mention if you're still using multiple subscriptions for things like ChatGPT, Claude Gemini, Grok 11 Labs for audio or any other AI model, I would love to tell you about AI Box, my own startup. I've personally built this and I'm recommending this to all of my friends who ask me how to use AI without having to spend a ton of money on all of the different subscriptions. You get access to over 80 different AI models in one place. You get basically every single flagship AI model from OpenAI, Anthropic, Google Xai Plus. You get a lot of the open source models uh, that you can't get hosted over there. You can pick from any model and if you ask a question, we'll actually pick the best one to answer your question. The thing that I'm the most excited about is the automation builder that we just shipped. You can basically describe what you want in plain English and it will build a workflow for you. So you don't need code, you don't need, you know, zapier and kind of the headache that that causes to wrestling with it. It's 8,99amonth. So if you want to consolidate all of the subscriptions you have for AI tools into one place and start automating real work, go check out AI Box AI. I'll leave a link of that in the description. Okay. The EU just landed a political agreement on the digital omnibus on AI. So this is something that we've been covering for a long time, the EU and AI and everything going on over there. I think the headline for all of this is that 2 August 2026 enforcement deadline for high risk AI systems under the EU AI Act. It's getting pushed back. So it kind of had this deadline for a while. We're pushing it back alone. What they're saying is that standalone high risk systems classified under Annex 3 now have a 12-02-2027 compliance date. So I mean beyond just getting pushed back from August, it's like December of next year. So you really have, I mean people have got a year and a half to to make sure that they're compliant or to worry about this. And AI embedded in regular products under Annex 1 gets pushed all the way to the 2nd of August. So we are pushing this years out. The industry basically just got a 16 to 24 month deferral on the rules that they were supposed to be ready for in three months. How did this happen? Basically, the European Commission introduced the omnibus on November 19 last year and it was specifically trying to, quote, simplify the AI Act. The Council and Parliament both adopted positions earlier this year. As of today, they have a consolidated text, they have some formal adoption that's expected by July, just kind of ahead of the original August deadline. Iverna McGowan, who's over at the center for Democracy and the Technology Europe, basically summed it up in April by saying, quote, the AI policy debate has reached a critical juncture with key disagreements still unresolved around scope and safeguards. I think the translation on basically what's going on is that the industry, the AI industry won kind of this fight for the timeline and the civil society is going to keep fighting about like, what's actually in the bill. So. So I'm thrilled this got pushed back. If you know me, I'm not a big regulation guy and I love, I love Europe. I've spent many years of my life living in France and yeah, huge fan of Europe. I'm a fan of everything France is doing with AI. They're doing some cool stuff and it kills me to see the overregulation that you see in so many other industries in Europe. It feels like we see that with AI specifically and it holds it back. You're not seeing all the top AI firms coming out of Europe. Mistral is doing well, but beyond that, everything else is coming out of the US And I think Europe could, could do a lot better. So anyways, all of that out of the way, I think a few things are happening. One, the postponing all of this is kind of the headline that everyone's talking about. Mistral, SAP, Siemens, every European AI player has been lobbying for this for like a year. And so they all got it. So Mistral's happy. Right. The second thing is I think there's kind of an open question about whether these, quote, high risk AI systems embedded into products already are regulated under EU's product safety legislation. And if that's the case, should those be exempt from AI, the AI acts scope entirely. Right. I think this happens a lot when, when there's a new technology, we're like, oh my gosh, how are we going to regulate AI? And it's like, well, it's kind of already regulated. Like if you're causing people harm, if you're doing bad things, there's already systems set up for that, in my, in my opinion. Right. I'm sure some people disagree with Me. And so, yeah, I think there's a lot of people making that case that it's kind of. It's already. The regulations already there. We probably don't need to have all this extra, you know, bureaucracy and burden. The European Parliament wants this. Basically, the Council is talking about it, and I think the fight is still live. Also, the Parliament added a targeted ban on AI systems that generated intimate content, NSFW content without consent. Plus they're adding some reformulated rules on processing special categories of personal data for bias detection. That kind of stuff, obviously, I think is great, but I feel like there's probably other laws that already cover it, but if not, you know, awesome. So how's the market actually responding to all of this? Brussels Bureau analysis has been pointing out for weeks that without the omnibus delay, half of European hospital procurement, most fintech, credit decisioning, and basically all hiring software would have been hit with this kind of huge paperwork addition in August that nobody was ready to file for. So I think the practical effects of this decision is that European enterprises, a lot of these AI, you know, procurement, and a lot of these Companies get about 16 minutes of Runway to get some of this stuff figured out. The last story of the day that I wanted to bring up is that China has just blocked Meta's $2 billion Manus deal. And this is kind of crazy. I mean, this has been going on for many months, but, you know, Meta has already went ahead and acquired this company called Manus, which technically, to be fair, was. It was started in China. The founders are Chinese. They realized that China was pretty restricted in this, and so they moved the company to Singapore. Now, as I believe, as I understand, the CEO and probably the CFO of Manus are still inside of China, and there's an extra like. Like, they're banned from leaving China, basically, so they're not going to let them out. But they did move the company to Singapore. When they moved it to Singapore. And by the way, it wasn't. I don't believe it was Manus, but they had, like, a different company name that they like Butterfly or something like that that they originally started this as, and then they kind of pivoted, created Manus, moved it to Singapore, and after they moved it to Singapore, Meta said, hey, we'd love to acquire it, because you can't obviously acquire a Chinese startup. They bought the company, they paid $2 billion, and they have, like, deeply embedded it. Like, I've gotten, you know, for, like, Meta ads. Like, I've personally gotten emails from Meta being like, hey, check out the new Manus thing. And, like, if you go to. To Meta ads or even to Facebook, I believe there's like a little bar on the side that says, you know, try Manus. It's like they're. They. They literally built it in. I was actually surprised by how fast that happens. A lot of times the acquisitions will take a long time. Meta grabbed it and immediately implemented it. And it's kind of like, you know, it's like a Claude coworker. Like, it's basically an agent that can actually do stuff for you. I know a lot of people using Manus, especially for Facebook ads. Like, I have a friend that he, you know, manages millions of dollars in Facebook ad spend for different companies. And yeah, he says he uses Manus all the time. He loves it. It's amazing. So Meta deeply integrated this software into there, and all of a sudden, out of Beijing, we get this. You know, basically Beijing saying, hey, you have to unwind the deal completely. This is what they said. They said, the National Development and Reform Commission has made a decision to prohibit foreign investment in the Manus project in accordance with laws and regulations and has required the parties involved to withdraw their acquisition transaction. So I think it's definitely a wild story, especially because Manus has about 100 employees that already moved to Meta's Singapore offices, right? So they acquired it. They move them into Meta's Singapore offices, and the founders are already taking on executive roles at Meta. So the CEO Xiao Hong now reports directly to Meta's coo, John Viej. Olivian. Someone reached out to Meta about it, and Meta said, quote, the transaction complied fully with applicable law. We anticipate an appropriate resolution to the inquiry. Okay, so sounds like Meta is not backing down. And it's kind of interesting because, you know, some tech companies, you feel like China has a little bit of leverage on, I mean, no shade. Well, maybe some shade. I don't know. Apple definitely, right, has all of their iPhones made in China, and it feels like they will comply to the government of China when they ask for certain things. Meta, I feel like, kind of doesn't really care because, like, every single Meta app is already banned and blocked in China. So, like, what? Like, what do they care if China tells them, hey, you have to, like, unwind this? I mean, I guess they could take the CEO who they have, you know, kind of locked inside the country, and they could torture them and tell Zuckerberg, like, they're not getting released from prison until you do that. So, like, they could pull some weird strings like that. But otherwise I'll be very curious to see what happens here. The one other thing that I did think was funny with this story in particular was that Washington, there's some senators, I think Senator John Coryn in Washington already raised concerns about Benchmark, an American VC company that was investing into this, into Manus. Because there's like, why are you, you know, making some investments, you know, into, into a Chinese linked AI company when you're an American company, whatever. So you kind of had this whole beef about it. So it's funny that from the American side, Washington, like, senators are complaining about it because it's it was Beijing backed. And meanwhile China is like, hey, we don't want our technology to leave. So at the end of the day, this will be a fascinating story to watch. But thank you so much for tuning to the podcast. That's it for the show today. If you enjoyed the episode, it would help the show an absolute tonight if you could leave a review over on Apple Podcasts or over on Spotify. It's the about tab on Spotify. But wherever you get your podcasts, if you could leave a review, it helps more people find it. It would make me super thrilled. And also make sure to go check out AI box AI if you want to get access to all of the models we talk about on the show for 8.99amonth all in one place. Hope it saves you a ton of time and money. All right, catch you guys all in the next episode.