A

Today on the AI Daily Brief, Apple sues OpenAI and tensions are ratcheting up as the AI competition shifts. We're going to discuss what happened and how it potentially benefits you. Before that in the headlines, reports suggest the Trump administration could be thinking about a new AI Executive order focused on open source. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. Alright friends, quick announcements before we dive in. First of all, thank you to today's sponsors, KPMG Blitzy, Retool and Airtable. To get an ad free version of the show, go to patreon.com aidaily brief or you can subscribe on Apple Podcasts. And to learn more about sponsoring the show, send us a Note@ SponsorsIDailyBrief AI finally, for those of you who have been jonesing for some summer learning activities, keep an ear out. We are going to be announcing some fun things soon. Your AI adventure awaits. For now though, let's talk about this new potential Executive order. Now the theme of our main episode, as you will see, is the increasing intensity of AI competition galvanized by this liminal in between period where there are major questions about which types of models are going to have value accrued at them, whether alternative architectures will change the entire shape of the AI business, and surrounding all of that is a geopolitical dimension, of course, which has made everything much more acute. Now last week we explored some reports that China was potentially considering ways to restrict or limit Western access to leading open source models. Now, however, there are rumblings that the Trump administration is itself considering another executive order to deal with what they perceive as the threat of Chinese open source AI models. In a Politico newsletter, journalists claim the White House is working on another EO to tackle the issue, writes Politico. Nine people familiar with the subject said the administration officials appear to be holding at least early stage discussions of how to deal with open source AI, a technology that poses potential security risks that existing US Policies are ill equipped to handle. Now, officials deny that this executive order is in the works, but reading the tea leaves, it would not be at all surprising to see the admin heading in this direction. During the short period last month where Mythos and Fable were offline, we had tons of headlines about GLM 5.2 and China creeping back up on frontier level performance. Now, of course, for those who are deep in the weeds of AI, we can understand how GLM 5.2 could be extremely valuable without it actually being at Mythos level. But it's fairly unlikely that senior White House officials at this point are running their own benchmarks to test model quality. Or much to their detriment, listening to the AI Daily Brief rather than simply reading headlines like this one from Fortune which read buckle up, the bad guys now have an AI Model as powerful as Mythos. As an aside, for what it's worth, even the ZAI CEO Zhi Tang said that they are not there yet, although they do expect to have an open source model at Mythos level by the end of the year. Additionally, there's been some chatter that the administration's AI policy isn't going as well as they'd hoped. Last week, Politico reported on an AI infrastructure export program that received a pretty tepid response embarrassing commerce officials. The report noted that Chinese firms are successful in quote, churning out low cost open source models that developing countries are increasingly adopting, and there are some indications that this administration is interested in pushing US models and infrastructure out into the world, viewing Chinese models as threatening that strategy. This is the flip side of the export control debate that we've been having forever in the US which is do you one restrict access to the most frontier models and the infrastructure to build them? Or two do you try to get models everywhere and use models themselves as the sphere of influence? And although most of our US policy has so far focused on the restriction side of things, it seems that some are interested in open source diplomacy now. Ultimately, at this stage all of this is just rumors. Andrew Curran commented, It's not clear if the Executive Order would target Chinese open source models specifically or apply more broadly to open source AI in general, but something is probably under discussion and indeed that plausibility seems to be the general tone. Adam Tirer from the R Street Institute said, I would not be surprised if this admin eventually sounds off about governmental use of open source AI and even potentially limits agency use of various open source tools. Some open source advocates are certainly wary, with Interconnect's Nathan Lambert writing a piece called Six Months to Live for Open Models, with the subheader saying it all staring down the barrel of policy action that could make open models a permanent second class citizen. This is certainly something we will keep an eye on, but in terms of policy that is here right now, the Trump administration has eased export controls for the uae, paving the way for AI mega clusters in the Middle East. In a rule change posted on Friday, the Commerce Department wrote that the UAE government and approved companies will now be able to access advanced AI chips without a license. The notice, which also enabled military exports cited new technology protection measures under the export deal signed in May of last year. That deal also contained domestic investment pledges that would see Emirati groups accelerate foreign direct investments in the U.S. the Commerce Department also cited cooperation from the Emirati government in the conflict with Iran. The Commerce Department said that they plan to favorably review applications from Emirati investment firms G42 and MGX, allowing them to shortly begin chip exports. It's unclear exactly how large the data center's plan for the UAE will end up being. Last year's deal mentioned 500,000 chips, but once MGX and G42 are approved, there's effectively no limit. Now, this is a fairly big development in terms of geopolitical precedent. The ability to import controlled goods without a license has previously been restricted to countries that are signed on to multilateral export control regimes. Generally, NATO countries are members of other formal alliances. By way of example, other friendly nations in the region, including Saudi Arabia and Israel, do not enjoy this benefit and need to go through the formal licensing process before getting access to US Tech. Even the Commerce Department acknowledges that this is a significant step, writing that it would, quote, significantly upgrade the status of the UAE in recognition of their status as a US Major defense partner and its support in advancing US national security interests. Unsurprisingly, the move is very controversial. Senator Elizabeth Warren blasted the deal as, in her words, corrupt due to President Trump's outside dealings with MGX through his family's crypto business. Warren claimed the president received a $263 million windfall from the deal. Meanwhile, Chris McGuire, a noted China hawk from the Council on Foreign Relations and former Commerce Department official, also had deep concerns over this policy change, he said. Now it is likely that the world's largest data centers will be in the UAE instead of the United States and will be operated by firms that will provide backdoor access to China. There is only one explanation for why Commerce made this change the UAE paid for it. Ryan Fedisiak from the American Enterprise Institute differed, however, commenting, I'm not sure this was the right decision, but with respect, I think some of these concerns are either overblown or fail to account for where the world is heading. He acknowledged concerns of trans shipment of chips to China, remote access to training clusters, and overall technology leakage, but continued, it is clear we are moving closer toward a world where the Gulf is an essential node in a globally distributed network of hubs running USAI inference workloads. I don't see an alternative to that future. And so if there is some upside to this decision it is that exporting large quantities of chips with less supervision will bring forward this reality sooner than later. We are living through an essential window where China's advanced chipmaking capabilities are still in their infancy to drive down AI inference costs, win new customers and remain the preferred partner for global software developers. The United States needs to be using this time to install chips and sockets as quickly as possible. I do agree with Ryan that whatever else is going on, there has clearly and rightly or wrongly been a shift towards viewing the Gulf as a key ally in the AI geopolitical era, and so I would expect more, not less, of this type of policy Moving over to Markets SK Hynix stock surged after completing the largest ever US IPO for a foreign company. Bucking the broader downtrend in semiconductors, the South Korean memory maker raised 26.5 billion on Friday in their Nasdaq debut. The company has already been public in South Korea since 1996, but this listing makes the stock far easier for US investors to access. The offering was slightly larger than Alibaba's 2014 IPO, which raised 25 million and not quite as large as the $29.4 billion raised by Saudi Aramco in their 2019 IPO. But that wasn't a US listing. The Nasdaq version of SK Hynix jumped by 13% on Friday, driven entirely by a massive IPO pop to begin the day. Now, this was an especially strong result given that semiconductors are in the midst of a pretty significant correction, with the sector's main index down 9% so far this month. Now, in many ways, this IPO is where some of those geopolitical discussions we were just having started to meet the markets coming into the ipo. Commerce Secretary Howard Lutnick urged SK Hynix and their South Korean competitor Samsung to build more capacity in the U.S. the two chip makers recently announced a $550 billion long term construction plan to expand capacity, but at this stage the buildout will be entirely within South Korea. Lutnick's comments were actually made during an event hosted by Micron, the largest US Based memory supplier. Referring to the Micron, CEO Letnik said, you know he won't like it, but I want to bring his competitors Samsung and SK Hynix here to America to build. Meanwhile, SK Hynix Chairman Che Tae Wan was in town for the ipo, and although he didn't commit to building capacity in the U.S. he did push back on the idea that new facilities would lead to oversupply. SK Hynix has committed to doubling capacity over the next five years. But Taewon commented, all my customers said that, well, that's not enough man, and well, we need more. Referring to a possible collapse in AI focused high bandwidth memory, the chairman doesn't think it's likely. The demand, he said, is enormous exponential, so I don't really see it. The AI agent, physical AI robot, all that needs a lot of memory chips. Taewon also reiterated his previous view that supply issues will continue to get worse, commenting, we expect 2027 to be the worst year in terms of memory supply shortage. Lastly, one little feature update from a story from last week. Meta has rolled back a controversial feature on their new image model after just a few days. Last week, you might remember, Meta released their impressive new image model directly into Instagram and WhatsApp. As part of the rollout, they introduced a feature that allowed users to include another person in their image generation by simply tagging their Instagram account in the prompt. The model would then use public Instagram images as part of the context, with users able to disable the feature. Still, despite the ability to disable the feature from the get go, controversy built over the following days, with the Screen Actors Guild warning their members that they should disable the feature immediately, among much other similar discourse. Ultimately, the decision to turn off the controversial feature isn't all that surprising, but it does provide one more indicator of where and where not the public is when it comes to AI imagery. Still, of the rest of the controversies and contentions we are going to discuss today, that is about the least of them. So with that, we will close the headlines and move on over into the main episode.

B

One of the most important AI questions

A

right now isn't who's using AI? It's who's using it?

B

Well, KPMG and the University of Texas at Austin just analyzed 1.4 million real workplace AI interactions and found something surprising the highest impact Users aren't better prompt engineers. They treat AI like a reasoning partner. They frame problems, guide thinking, iterate, and push for better answers. And the good news? These behaviors are teachable at scale. If you're trying to move from AI access to real capability, KPMG's research on sophisticated AI collaboration is worth your time. Learn more at kpmg.com us sophisticated that's kpmg.com us sophisticated with the emergence of AI code generation in 2022, Nvidia master inventor and Harvard engineer Sid Pareshi took a contrarian st Inference, time compute and agent orchestration, not pre training would be the key to unlocking high quality AI driven software development in the enterprise. He believed the real breakthrough wasn't in how fast AI could generate code, but in how deeply it could reason to build enterprise grade applications. While the rest of the world focused on co pilots, he architected something fundamentally different. Blitzi the first autonomous software development platform leveraging thousands of agents that is purpose built for enterprise scale code bases. Fortune 500 leaders are unlocking 5x engineering velocity and delivering months of engineering work

A

in a matter of days with Blitzi.

B

Transform the way you develop software.

A

Discover how@blitzi.com that's blitzy.com this episode is brought to you by Retool. Generating a working app now takes about five minutes thanks to AI getting it safely into production with auth permissions, audit logs, security reviews. That part still takes time. That's the gap Retool closes. Build apps however you want natively in Retool with Claude code, codecs or any coding agent or by importing React you've already built it all. Deploys into Retool and picks up governance automatically. Security lives in the platform, not in whatever the AI wrote, so your team ships at AI speed without the shadow

B

it and the endless reviews that stall

A

out most vibe coded projects. It's why teams at Amazon, Stripe and Brex build on Retool and new enterprise customers who sign by September 30th get up to $10,000 in AI credits per year. Start building@retool.com aidaily this episode of the

B

AI Daily Brief is brought to you by Hyperagent, where you run fleets of agents your team can manage together. New users get $1,000 in inference. Forget local agents and chat workflows waiting on your laptop to be prompted. Hyperagent deploys always on agents in the cloud, doing real work across the tools your team already uses. Marketing's agent turns competitor, moves into landing pages. Sales agent enriches leads, drafts emails and updates. The CRM Ops agent chases the paperwork and tracks the budget. Every agent has access to shared context and follows your rules about scope and approvals. It's time you add agents that feel like teammates. Hire yours at HyperAgent, built by the team at Airtable. Claim your $1,000 in inference at hyperagent.com aidaily Brief.

A

Welcome back to the AI Daily Brief. Today we have a number of stories that, while seemingly disconnected at first, I actually think are part of a larger theme that's happening right now. AI competition has been fierce for some time, and despite the fact that the industry has at least thus far been pretty much a rising tide lifts all boats kind of environmental and will I think in many ways continue to be so. The competitive dynamics of the field are shifting right now in ways that feel more intense than just the past questions of who has the best frontier model. Part of that is that model is not the only vector of competition anymore, and in fact the core architectures of particularly how businesses use AI is itself in a potentially transitional period. It is in that particular cauldron, that in between moment that all of our stories operate today, the first of which is that Apple has sued OpenAI for stealing trade secrets in a lawsuit with fairly significant implications. On Friday, Apple filed a blockbuster lawsuit that alleges OpenAI had stolen hardware designs and assorted other IP. Now, despite the two companies nominally being partners, there's been rumblings about bad blood between the two for some time now, dating particularly back to OpenAI partnering with legendary Apple designer Jony I've in May of last year, many took that partnership itself as a clear indication that Sam Altman wanted OpenAI to become the next Apple and to produce a device just as revolutionary as the iPhone. What followed was a huge poaching campaign with OpenAI and Jony I've's design firm hiring more than two dozen hardware and AI specialists away from Apple. They also partnered with several Chinese companies within the current or former iPhone supply chain. In one specific example, Apple claims that OpenAI told a supplier that Apple had consented to them using the same metal finish as an iPhone, which wasn't the case. Now none of that is illegal, especially in Silicon Valley, where non competes are both illegal and culturally frowned upon. And in fact I think there are many people who would say that when it comes to their talent leaving Apple has kind of made their own bed through the actions or lack of actions they've taken around this dynamic new field. Still, what Apple is alleging goes far beyond the realm of strong competition. The lawsuit centers on an iPhone engineer named Chang Liu, who left Apple to join OpenAI early in 2025. Lou left Apple without returning his company MacBook, and also kept in touch with former colleagues that Apple claims were feeding him information. The most damning allegation is that Lou had knowledge of a software bug that allowed him to gain access to Apple's servers. In their lawsuit, Apple included a text message from Lou to a then Apple employee named Alyssa Peng. Lol, wrote Lou, I found out I can access the network storage. So funny. Apple claims that Lou used this access to download presentations, hardware designs, manufacturing details and testing procedures. While he was working at OpenAI a few months after that incident in April of last year, Peng also left Apple to join OpenAI. Apple claims that over 400 former Apple employees have joined OpenAI to date, largely to work in their burgeoning hardware division. Apple point to former iPhone design lead Tang Tan as the root of the problems. Tan had risen through the ranks to become one of Apple's top executives by the time he was thinking of moving on to join Jony I've as a co founder in his new design firm. In late 2023, Apple allowed him to stay on until early the following year, but they now allege he was already secretly working on plans with Ivan Altman. Apple alleges that Tan instigated the recruitment drive that followed. Now, at this point, you might be saying to yourself, well, yeah, it stinks for Apple that a bunch of their employees left for OpenAI, but that's just kind of what happens. What's more, you might rightly point out that with hiring 400 former Apple employees, you are going to inherently get lots of IP that's locked not in people's hard drives, but in people's minds from their experience at that company. And yet the most important claim in the lawsuit is that Apple is claiming that OpenAI encouraged, actively encouraged employees to steal IP on their way out the door, urging new hires to study confidential material before interviews or even bring hardware components and prototypes to show and tell sessions at OpenAI headquarters, writes Apple in their lawsuit. OpenAI's nascent hardware business now rests on the shakiest of foundations, rotten to its core by its illegal reliance on misappropriated trade secrets. The Wall Street Journal called this Apple taking the thermonuclear option. In their lawsuit, Apple claims that they tried to deal with this issue peacefully for months, asking OpenAI to investigate and rectify the situation. They claim they never got a response, so had little option. But the lawsuit the Journal reflected on Steve Jobs, notoriously declaring thermonuclear war on Google in 2010, calling the Android operating system a stolen product. They believe that Tim Cook, as one of his final acts of CEO, is taking a similar approach to stifling OpenAI's device. Experts have only had a few days to mull over the pleadings, and the consensus is still pretty mixed. Gene Gann, the director of legal at Saville's Singapore group, noted that California courts have rejected non competes, adding, so every allegation rests on conduct, retained devices, unauthorized access, misused documents, coached evasion in a jurisdiction where talent moves freely by design. Trade secrets law is the only legal perimeter left around institutional knowledge, and Apple has pleaded squarely inside it. Paul Simanza, the chair of the Engineering Management School at Santa Clara University, commented, getting an existing Apple employee to take the risk of bringing parts to an interview session seems more like a test of how desperate they are to work at OpenAI than anything else. Targeting Apple's supply chain is a declaration of war, and given that Apple fought Samsung for years over rounded corners, it is hardly surprising to see Apple listing metal finishes as an example of IP theft. The question here is how this gets settled, given that unlike with Samsung, Apple is unlikely to be interested in cross licensing anything from OpenAI. Now at this stage we have very little from OpenAI. Just a brief statement to the press which reads, we have no interest in other companies trade secrets. We remain focused on building innovative technology that empowers people everywhere. Now this is one where I very much want to wait to see what OpenAI files as a response before I really make up my mind on it. Googly Arose wrote, I am sometimes really surprised by how dumb very highly paid people in tech can be. 1. Leave Apple 2. Start building hardware at OpenAI 3. Access confidential Apple files on hardware from an unreturned Apple laptop 4. Expect what to get away with it I agree that this seems strange. It also feels very weird to me that OpenAI leadership would actively encourage such easy to document types of IP theft. It seems strange and desperate in a way that doesn't really comport with where OpenAI has seen themselves, but who knows, Smarter people have done dumber things. I think Ricky Ho is right when he writes that Apple's lawsuit against OpenAI is more than a dispute over trade secrets, but a signal that the AI race is entering a new phase where hardware, not just models, has become a new strategic battleground. That is really the theme right now, that it is no longer just about models, it's about the entire ecosystem around them now. One person who was not shy about commenting on this, as you might guess, was Elon Musk, who retweeted a post about it and said they sure put a lot of effort into this crime. In fact, Elon decided to use this weekend to resume his petty public feud with Sam Altman, retweeting a repost of his own post where he had said scam. Altman is super good at scamming. Elon added, he takes scamming to a whole new level. Sam Altman deciding to willfully ignore Michelle Obama's recommendation to go high when they go Low retweeted Elon Musk's post and wrote, homeboy, you're the one selling public market investors on short term space data centers. To which Elon responded, we start flying them next year. Maybe you can come see them if your parole officer approves. After stealing an open source AI charity, then you stole all of Apple's phone technology. Wow. What do you plan for an encore that's tough to beat? In a separate post, which, if we are declaring a winner in this unseemly tit for tat, has to be the victor, Altman tweeted, There are a lot of benchmarks that suggest 5, 6 SOL is the best model in the world right now, but the most reliable way to tell is that Elon is obsessed with me. Again, perhaps getting a preview of what OpenAI's tone is going to be with regard to this lawsuit when an I like Tesla's fan account tweeted, sam Altman wasn't afraid of Elon, but he is terrified of Apple. You can tell by all his posting today. Altman actually responded, I am not afraid of Apple, but I have tremendous respect for them. S tier company now, while all that was going on, the competition that is much more relevant for most of us right now is in fact the new competition between Fable 5 and GPT5.6 Seoul. Indeed, that competition seems to be potentially giving us a little bit of a reprieve at the end of the subsidy era. Over the first weekend with GPT 5.6, people were having a lot of fun discovering what the powerful new model could do. But there was one very common complaint. People were burning through all their tokens at an unbelievable pace. The problem was so severe that AI power users on X started to share tips on how to use 5.6Sol without immediately hitting the usage limit. Now, part of the problem was that this is one of the first models where you probably need to control reasoning effort rather than dialing the settings to the max. But there also seemed to be some configuration issues on OpenAI's end that contributed to excessive token burnt. Usage limits were reset several times over the weekend to compensate users, but that didn't do much to fix the underlying issues. On Saturday, Chubby commented, seriously, this has to stop. I've now set GPT5.6 from high to medium, not fast mode of course, and I'm still burning through my rates at an insane rate. My five hours are almost gone again, I've already used up all three resets. OpenAI needs to work on its efficiency. This is the biggest bottleneck on Sunday Morning, Tebow from OpenAI checked in with the Fix posting the last 48 hours of Codex and ChatGPT work have been intense. Three important updates 1 temporarily removing the 5 hour usage limit restriction for all plus business and pro plans 2 rolling out changes that will make GPT 5.6 SOL more efficient across the board and that will be reflected in less usage being used so that it can take you further exact impact to be quantified and shared. 3 We hit 6 million active users and are landing a usage reset in the next hour. Go do things According to early reports, some of the changes do seem to be helping and OpenAI said they'll continue to work on token efficiency tweaks. Now heading into the release of GPT5.6, many thought OpenAI would use the model released to poach users from Anthropic. The model was expected to land just as the trial period for Fable expired and Anthropic subscribers would be forced to pay full price. Of course, that's not how it played out with Anthropic extending the trial period until Sunday just as GPT5.6 was released. Then on Sunday they extended it again, including Fable in the subscription for another week and keeping Claude code limits 50% higher. Developer ECAS wrote, Capacity wars between labs are one of the best things that can happen to us who build with this. And while indeed this is an unambiguous short term win for the power users and really, what are you even doing listening to me right now? Go use this subscription subsidy while it lasts, because you do have to wonder how sustainable this price war will be. Semi Analysis recently updated their understanding of the size of token subsidies and and found the subscriptions are still a ludicrously good deal. The $20 a month tier still allow $400 of usage for Anthropic or 700 from OpenAI, while the $200 a month tier are now running at 8,000 in max tokens from Anthropic or a staggering 14,000 in tokens from OpenAI. Now obviously the average user isn't actually getting this much value from their subscription, but on a weekend with multiple usage resets from OpenAI and anthropic extending their Fable subsidies, there is very clearly going to be a price war that, at least for a while, we can all take great advantage of. And yet we know the frontier isn't going to be subsidized forever, even if we are getting a temporary reprieve. I do believe that it is temporary and I think that the race to find alternative architectures is going to do nothing but continue as we heard in the headlines today and in a main episode last week, the one answer of simply turning to cheaper Chinese models looks more in jeopardy than it has in the past. Interestingly, the founder of Chinese AI lab Zhipu, also known as Z AI, recently wrote a note imploring why frontier AI should stay open to all. He wrote. Recently we released GLM 5.2, our most capable open source model to date. It supports a genuinely practical context window of 1 million tokens, continues to lead in long horizon tasks, and is available to all users. It will also be officially open sourced under the highly permissive MIT license. Anyone will be able to download it, deploy it, and use it commercially with no restrictions based on the type of user or organization. This is the company's firm position expressed through the form of its product we choose to believe in a different path Frontier intelligence should not belong only to a select few, nor should access to it be withdrawn at any moment by a small group of rulemakers. It should be open, usable and buildable, and it should serve every developer. With one hand we reach upward to challenge the limits of intelligence. With the other, we build roads downward, making the most advanced capabilities as open and broadly accessible as possible. The heights we reach belong to all humanity, and the roads we build belong to everyone. And of course, it's not just the Chinese labs that are trying to own a different narrative around AI. Microsoft CEO Satya Nadella once again took the X to write a blog post further articulating the new vision for AI that they're starting to promote. In it, he argues that consumers and businesses right now, quote, essentially pay for intelligence twice, once with money and again with something even more valuable. The proprietary knowledge you must reveal to make that intelligence useful. The better you want the model to perform, the more of that knowledge you have to feed it. He basically talks about how the frontier model providers get richer from usage models. He writes, learn from exhaust the prompts people write, the tools agents use, and especially the corrections people make when the model is wrong. Every correction is distilled into institutional know how. It's the kind of knowledge a competitor could never buy, and the kind that leaks almost imperceptibly, trace by trace, correction by correction, eval by eval. In consuming intelligence, you are creating intelligence, and what you create should belong to you. And here's his subtle declaration of war. While the great innovation that comes from model providers having fair use rights to train models on public data is needed, I find it ironic that the status quo is to then turn around and impose restrictive terms on distillation and to reserve the right to learn from customer usage and interaction data. If learning flows in only one direction, economic value converges towards the owner of the learning infrastructure rather than the creators of the knowledge itself. Therefore, it's imperative that we distribute the learning infrastructure to every firm so that they can control their own learning loop. He then refers to a recent interview with Palantir CEO Alex Karp, who said, what the technical customers want is control over their compute, their models, their data stack and their alpha. They want to know they own the means of production and it's not being transferred to someone else. The current regime, Nadella says, does precisely the transfer Karp and companies fear. So what should enterprises do? Well, Vercel CEO Guillermo Rush sums it up by saying, make the model a cog in a machine you own. Startups and enterprises must own their data evals Model choices Software layer don't outsource your brain CNBC's Deirdre Boso summed up some of the changes going on right now in a piece called the AI race is shifting from bigger models to cheaper, smarter systems, which is of course exactly the shift that we've been charting on this show. It's also one that markets are paying attention to. Big Shorter Michael Burry retweeted the CNBC piece, affirming that this is what he's been hearing from his contacts in Silicon Valley around where the AI race is shifting from bigger models to cheaper and smarter systems. Now, for some like Burry, this is how an AI bubble comes crashing. In short, if people don't want to buy anthropic and OpenAI tokens anymore, all the infrastructure deals around them crumble and everyone goes home a loser. Gavin Baker, on the other hand, thinks that this represents a big opportunity. Retweeting Michael Burry, he wrote the mega bull case for AI infrastructure would be if market share shifted away from certain frontier labs with 90% plus inference margins towards cheaper models, whether open source or closed. It would increase the ROI on AI spend for end customers by increasing intelligence per dollar, which would drive incremental token demand. Margin dollars would effectively get redistributed from the frontier labs to AI infrastructure providers. The infra winners would be those with the lowest per token cost, and the winners at the model layer would be those with the highest token efficiency. There are many reasons Jensen is so focused on open source, but this is likely the most important one. Lower margin percentage at the model layer equals more margin dollars at the infra layer. All else being equal now, Baker points out, this is not happening yet. Cheap, mostly open source tokens are likely the majority of volume today, but the majority of economic value is still accruing to the most intelligent models. Might change though. We will see. It's beyond the scope of this show, but one thing that I will explore in future episodes is what I find as a faulty assumption that even if this is the shift in buying behavior that happens, that somehow the leading Frontier Labs just aren't going to participate, all the evidence suggests that that's not going to be the case. And if you need evidence of that, just look at the announcement for GPT 5. 6. The cheaper versions of that model, I. E. Terra and Luna, were basically better than GLM performance for lower than GLM costs. If you think that OpenAI and Anthropic are just going to roll over and let the market shift away from them, you're nuts. I also think that the speed at which these sort of changes are going to happen are wildly overstated. Most firms are still trying to get their employees to use their CLAUDE subscriptions, not thinking about advanced model architectures even if they should be. And then of course I haven't even mentioned Google, who are off cooking something and have already released a lot of products that show that they're paying attention to the efficiency and cost side of the market. This might be an opportunity for them to be a leader in a really important market segment, even if their leading Gemini model at the moment can't match up with Fable and GPT 5.6. And still, for all that being said, there is no doubt that the tectonic plates of AI competition are shifting. I think a big reason for the intensity that you saw at the beginning of this episode, the growing public spats and acrimony, is that no one has a handle on exactly how things are going to shift next. Those unstable foundations create lots of anxiety and stress and tension even for companies that are doing really, really well now. What we see from these subsidy extensions is that at least in the short term, this sort of competition is really good for all of us. And so if you take away nothing else for this in between period, at least take advantage of the competitive upheaval because if you're smart, you can probably as an individual benefit from it pretty mightily. Anyways guys, for now, that's going to do it for today's AI Daily Brief. Appreciate you listening or watching as always. And until next time, peace.