AI Lab Power Rankings

0:00Today on the AI Daily Brief, find out who both me and AI think are the leading AI labs in our first ever AI lab power rankings. And before that, in the headlines, some big updates to the Microsoft OpenAI Partnership. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. All right, friends, quick announcements before we dive in. First of all, thank you to our sponsors, KPMG, Blitzy, Granola, and Mercury. To get an ad-free version of the show, go to patreon.com slash ai daily brief, or you can subscribe on Apple Podcasts. podcasts. To learn more about sponsoring the show or really find out anything else about the show, head on over to AIDailyBrief.ai.

0:41One of the things you can find access to there is play.aidailybrief.ai, which is where we put all the companion experiences for this show, including the place where you can build your own AI Lab Power Rankings. And you can also find links for things to our extension programs like our free self-paced program, AgentOS, which has just launched and now has multiple thousands of you building agentic operating systems. Again, you can find all of that on AIDailyBrief.ai. On Monday, OpenAI and Microsoft announced that they've signed an amended agreement that unwinds some fairly big parts of their long-term partnership.

1:14Microsoft will continue to be OpenAI's primary cloud partner, but they will no longer be their exclusive cloud partner. That clears the way for OpenAI to serve any of their products on AWS through their new partnership with Amazon. Microsoft will continue to hold a license to OpenAI's IP and models through 2032, but that license is also now non-exclusive. Models must be released first on Azure, but we have no information on how long that exclusivity window is. In exchange for opening the partnership, Microsoft will no longer pay a rev share to OpenAI for serving their models. OpenAI will continue to pay a rev share to Microsoft through 2030 at the same percentage and subject to the same total cap.

1:51This revenue share has been previously reported at 20 % of total OpenAI revenue, and the cap is understood to be some multiple of Microsoft's early$13 billion investment. Microsoft will retain their shares and continued being a 27 % shareholder in the company. Maybe biggest of all, certainly from Microsoft's perspective, is that revenue and IP sharing is no longer conditional on it being pre-AGI. Remember, one of the weirdest clauses of their partnership was that the deal fell apart if OpenAI declared that they had achieved AGI, but there wasn't really a definition of AGI built in, meaning that Microsoft was sort of subject to the whims of OpenAI.

2:22At the beginning, that didn't seem like such a big deal, but I think it became a major liability after the whole dust-up around Sam and the board a couple of years ago. The joint statement insisted the split was amicable, with the companies framing it as a simplification. In terms of winners and losers, most reporting framed this as OpenAI breaking free of a relationship that was holding them back. OpenAI has been pushing for the ability to host their full suite of products on AWS, which involved a few workarounds that resulted in lawsuit threats from Microsoft. This new deal means OpenAI has zero exclusivity and can also pursue a partnership with Google if they choose.

2:53The information, on the other hand, initially reported the deal as a win for Microsoft. They argued that the financial benefits are pretty decent. with Microsoft holding on to a 20 % revenue share and looking towards a huge windfall gain from their 27 % equity stake. One aspect that's a little overlooked was that Microsoft was previously looking at a revenue stake reduction to 8 % by 2030. If OpenAI continues on their trajectory, retaining the enhanced revenue stake could be a massive contributor to Microsoft's bottom line. Now, after a little more sourcing around the negotiation process, the information assessed the deal as a win-win in follow-up reporting on Tuesday.

3:26Or, more accurately, they explained that the deal allowed OpenAI and Microsoft to avoid the lose scenario of a protracted legal battle. They reported that the negotiation essentially came down to OpenAI winning the ability to form other partnerships in exchange for allowing Microsoft to retain the 20 % rev share. Removal of the AGI clause also loomed large. For my eyes, it's very hard to see this as anything but a win-win for both companies, and I actually think that Rezo has the right of it when he writes, While everyone else is obsessing over the revenue share drama, the real story is much simpler.

3:55OpenAI has grown too big for any single cloud to fully serve. I think that's dead on. The deal was a constraint on everyone, it was structured at a different time, it needed an update, and frankly, huge kudos to Sam and Satya for figuring this out over a handful of meetings and just allowing everyone to keep building. Now, not wasting a single day of their newfound freedom, OpenAI models are now available on AWS. AWS CEO Matt Garman announced on Tuesday that GPT-5.4 is now available as a limited preview and 5.5 will be coming within weeks. AWS will also serve Codex through their infrastructure.

4:25The announcement also reintroduced Amazon Bedrock's managed agents platform, now branded as powered by OpenAI. The platform will use OpenAI's harnesses and models, making it seem pretty similar to the managed agents OpenAI introduced with their Frontier platform in February. Importantly, in the context of the value of the deal that we were just discussing, there isn't a single workaround in sight. This is just OpenAI's products running on AWS Bedrock in the same way they've been available on Microsoft Azure. Garmin explained why this is such a big deal for AWS, commenting, this is what our customers have been asking for for a really long time.

4:55Their production applications run in AWS, their data is in AWS, they trust the security of AWS. Amazon is also getting into the agent game with a new desktop computer use assistant. Called Amazon Quick, the agent is in a similar space to some of the things we're seeing around tools like Cloud Cowork. Quick can access local files, create live dashboards, and generate work-related outputs like slideshows. It can connect through the apps that store your context, including calendars and email clients, alongside professional tools like Slack and Jira. Amazon says the agent will automatically learn from every task, building personal context over time.

5:27The information framed this as one of Amazon's big plays for the agentic era, commenting, AWS is still chasing its white whale, creating a hit enterprise application. Now the response of some was summed up by Tmuxvim who writes, okay, how many of these do we need? While others like Brandon Pizzacalla saw the potential value. Brandon wrote, Every AI product we've built, the model took a week to get right. Data wiring took another month. Hooking agents into real context, actual email history, and support tickets, that's where most of these break. If Quik actually solved that piece, it's worth paying attention to.

5:59Look, I do not know how this all shakes out, but it is very clear that this agentic do-everything, work-everywhere desktop app is a major vector of work AI competition, and we're going to see a lot more iterating around it. On the Anthropic front, Claude announced a bunch of small but potentially exciting integrations, New connectors for Claude include Adobe's Creative Cloud apps, Affinity, Blender, Ableton, Autodesk, and more. Now, since the subject of our main episode today is going to be a power ranking taking into consideration all of these new moves among the major labs, I did want to quickly mention this Wall Street Journal story that got everyone on Wall Street all bothered yesterday.

6:31The point that I wanted to make, this was the story, of course, that said that OpenAI had missed key revenue targets, as well as user growth targets, which ripped a bunch of market cap off of OpenAI partners. People following this closely understood that fundamentally, these numbers are just out of date and don't reflect the world that we actually live in anymore. Beth Jezos wrote that the revenue slowdown is literally all clawed eating their lunch. In other words, a lagging indicator. Cracked devs on X are leading indicators and everyone I know switch to codex. Give it a month and it will be apparent in revenue numbers.

6:59Dan McCatier writes, The Wall Street Journal is living in the past with this report. AI agents started to work at the end of 2025. OpenAI has the compute capacity, they will need it. The point that I wanted to make is actually much broader than this one report. I think for the next couple of months, we're going to be in a really weird period where you're going to see a bunch of research and studies that come out that are just unbelievably disconnected from the reality of AI on the ground. The structural shift in the industry from the pre-agent to the agent period means that data from that pre-agent period just will not reflect the reality anymore.

7:31I say this just as a caution because it's going to take a little while for research and data to catch up. Maybe don't freak out and sell off your stocks all at a Wall Street Journal report. For now, though, that is going to do it for today's AI Daily Brief headlines. Next up, the main episode. One of the most important AI questions right now isn't who's using AI, it's who's using it 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.

8:05They 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 slash US slash sophisticated. That's kpmg.com slash US slash sophisticated. If you're looking to adopt an agentic SDLC, Blitzy is the key to unlocking unmatched engineering velocity. Blitzy's differentiation starts with infinite code context. Thousands of specialized agents ingest millions of lines of your code in a single pass, mapping every dependency.

8:42With a complete contextual understanding of your code base, enterprises leverage Blitzy at the beginning of every sprint to deliver over 80 % of the work autonomously. Enterprise-grade, end-to-end tested code that leverages your existing services, components, and standards. This isn't AI autocomplete. This is spec and test-driven development at the speed of compute. Schedule a technical deep dive with our AI experts at Blitzy.com. That's B-L-I-T-Z-Y dot com. Today's episode is brought to you by Granola. Granola is the AI notepad for people in back-to-back meetings. You've probably heard people raving about Granola.

9:13It's just one of those products that people love to talk about. I myself have been using Granola for well over a year now, and honestly, it's one of the tools that changed the way I work. Granola takes meeting notes for you without any intrusive bots joining your calls. During or after the call, you can chat with your notes, ask Granola to pull out action items, help you negotiate, write a follow-up email, or even coach you using recipes, which are pre-made prompts. Once you try it on a first meeting, it's hard to go without. Head to granola.ai slash ai daily and use code ai daily. New users get 100 % off for the first three months.

9:45Again, that's granola.ai slash ai daily. This episode is brought to you by Mercury, radically different banking now available for personal accounts. I already use Mercury for my business. So when they introduced personal accounts, it made immediate sense for me. I try to bring the same level of intention to my personal finances that I bring to building companies, and most traditional banks just do not feel designed for that. With Mercury Personal, you can toggle between business and personal in a click. You can set up subaccounts for specific goals, automate transfers so projects and savings fund themselves, and put idle cash to work with high-yield savings, all without friction.

10:19It's built for people who care about how their money moves and want tools that actually keep up. Visit mercury.com slash personal to learn more. Mercury is a fintech company, not an FDIC-insured bank. Banking services provided through Choice Financial Group and Column N.A. members FDIC.

10:59what some perceive as overdue update from Google around Gemini. Point being, I had a suspicion that this week would be a bit of an in-between week, which relative to AI at least it has been. To the extent that there is a story, it is definitely the repositioning of relationships between all of these different labs. Zooming out and looking more broadly, we're still in the middle of the transition to the agentic era from the pre-agentic era. As we could tell from Wall Street's response to reports that last year OpenAI missed some revenue targets, the broader market and world is still digesting the fact that things have changed.

11:31Yet of course for those of us paying attention, the shape of this new world is starting to become a little bit clearer. It's a world characterized by shortages of tokens and compute, and of the business model shifts that will have to happen to deal with those shortages. It's a world where we're rapidly redesigning not just workflows, but entire conceptions of how work gets done, and updating interfaces for a totally new category of agentic user. In short, there is a lot changing right now. And so as a way to help ground people, we're diving into what is consistently one of the biggest questions in AI, which is lab competition.

12:02For our purposes, I thought it would be fun to break that lab competition into a few broad-based categories, try to give some weighting to those categories, and then see where things shake out in terms of the competition. To me, this is way less about trying to pick a winner and more about the discussion of what is important to labs right now and where different labs have unique strengths or critical weaknesses. For you guys as listeners, this might be useful as you're trying to make decisions about where to invest, especially if you're helping make decisions for an enterprise. And even if that's not the case, with my full scorecard, I've now given you 72 different places to either agree or disagree and argue in the comments.

12:36So what we're going to do is we're going to talk through the different categories and weighting. I'll quickly go through the AI assessments, i.e. how Gemini, Grok, ChatGPT, and Claude all rank things. And then I'll discuss why I put things where I did. So first up, let's talk methodology. I divided the power rankings into nine categories. Compute and infrastructure, enterprise positioning, platform and ecosystem control, consumer positioning, model leverage, momentum, branded narrative, wedge, and x-factor. Many of these are fairly self-explanatory. The two that I'll highlight are wedge, which is some really unique asset or entry point that others can't easily copy.

13:10This can overlap a little bit with platform and ecosystem control. And what's interesting about it is that when you look across all the different labs, the wedges that they have are really different, and you can see it flowing through into their strategies. Now, of course, X-Factor is just my catch-all for anything that doesn't fit neatly into these categories, but that I think should be included. In terms of the weighting, right now I put compute and infrastructure as the biggest category. And when I had the AIs do their assessments, I asked them first to use this rating scale, but then to make an argument for what they would change if they were in charge.

13:41And a number of them suggested that compute and infrastructure should be even more than the 20 points that it accounts for here. A couple of the models also thought that the momentum rating should be higher. It accounts for 10 points out of 100 in this methodology, and I can definitely see it being worth more, although I was concerned with giving it more points than that, A, because the momentum can shift so quickly, and B, because I think momentum feels a little bit more significant to those of us who are paying attention on a daily basis than it does in general out in the rest of the world. Lastly, I don't think this will be that controversial, but right now this methodology has enterprise positioning as worth more than consumer positioning.

14:15That won't necessarily always be the case, but it's pretty clear that locking in with business is one of the two major vectors of competition with the other one being developer devotion right now. So I think it justifies consumer being just a little bit lower. Given that the token shortages are also coming from work-based agents, I think you could make an argument for them being even farther apart than the difference between 15 and 10 points that they are currently. Overall, the aggregation of the AIs has Google at number one, OpenAI at number two, Microsoft at number three, Anthropic at number four, Amazon at number five, Meta at number six, XAI at number seven, and Apple at number eight.

14:50I was back and forth a little bit on whether even to include Apple, but at the end of the day, they have a massive consumer base. They're doing deals with Google Gemini. They may not be choosing to fight the same battles that the other labs are, but they do have a stake in this race. Now, as I mentioned, all of the different models put Google in the top slot, which I think pretty simply comes down to Google's full-spack strengths. It's got a strong ecosystem, it's got strong models, it's got consumer and enterprise adoption, and it's got compute and infrastructure. There was, however, more variety in the number two slot.

15:21Claude had Anthropic at number two, although it was just barely above OpenAI. ChatGPT had OpenAI at number two, despite suggesting that Anthropic had the hottest model and enterprise momentum stories. And interestingly, both Grok and Gemini put Microsoft in the number two slot, putting a lot of emphasis on both their infrastructure as well as their enterprise incumbency. Now you can see that overall, the AI scored things pretty highly. Google's average score was 91.4 out of 100. OpenAI's was 85.4. Microsoft's was 84.9. Anthropics, 83.1. And Amazon at 80.4. The top five were all above 80. Compare that to my scores.

15:57I only had three labs score above a 70. Anthropic at the number three slot at 70. And then OpenAI and Google tied at 74. The rest of the labs I have clustered between Apple on the low end at 58. and Amazon at number four, scoring a 64. As Codex pointed out, and you can probably tell that I built this with Codex given all the rounded edges and boxes, overall I was much harsher than the AI. Now I'm not going to go through every single category and score, but I did want to highlight first some of the key differences among the top three, where I think I might be over or undercounting things, how I look at things a little bit differently than I think the AI did, and where we could see some big changes pretty quickly.

16:35First of all, among the top three, I put Google significantly higher at compute at a 17 as compared to OpenAI's 12 and Anthropics' 10. I think there's an argument that Anthropics should even be lower or at least farther away from OpenAI, and I did think it was important to put some pretty significant space between OpenAI and Google, because although yes, OpenAI has been scurrying around for the last year to get compute deals, being dependent on deals with others that themselves require financing is very different than owning a big chunk of that in-house. When it came to Enterprise, I think this is the one where Gemini partisans are going to be most angry at me.

17:08Out of 15, I gave Anthropic a 14, OpenAI a 10, and Google an 8. And let me explain why. First of all, one of my beliefs overall, with Enterprise, is that incumbency right now in the Enterprise is worth less than I think people think it is. So for example, I gave Anthropic the same score that I gave Microsoft. Both have a 14. One could argue that that's totally insane given how much enterprise dominance Microsoft has. But I think that when it comes to AI, enterprises are treating this as a much bigger transformation than just picking a new software vendor. And I sort of think that they're treating the leading model labs, specifically Anthropic and OpenAI, much differently than they would have treated even successful startups of the past.

17:48Microsoft scores incredibly highly for having distribution, but at the end of the day, it's serving these other companies' models. Some enterprise buyers will be fine with that, and in fact, they'll want more choice than the individual labs can provide alone, but I think a lot of them are showing that they want to go direct to the source. I think OpenAI's 10 is a little aspirational. Enterprise is clearly growing in importance for them right now. It's never not been a thing, but compared to how historically important it's been to Anthropic, that's one where OpenAI is definitely racing to catch up a bit.

18:15Now, like I said, the company that gets punished the most on this is Google. And I think Google's enterprise relationship has always been kind of a weird one, even before AI. On the one hand, they have a ton of incredible tools. Companies that aren't locked into the Microsoft ecosystem often by default find themselves working in the Google workspace. We use Google Drive and Google Sheets and Gmail and all these sorts of things. But I think Google has struggled historically to convert that to the highest levels. Their attention is clearly fragmented across their massive consumer empire as well, and enterprises can tell that.

18:46I think that has followed them into the AI realm, and I think Gemini in the enterprise has been weaker than they'd like. Now, they still have a bunch of structural advantages. This is a number that could change very fast. But for right now, I think it's reasonable to place them behind both OpenAI and Anthropic. On the platform category, what's interesting about this is that they all have such different platforms that they're working with. Microsoft and Amazon's are the most similar, with Bedrock and Azure competing in very similar, enterprise safe, don't-have-to-choose-a-model type of space. But when it comes to Anthropic, OpenAI, and Google, the platforms that they're building on top of are really different.

19:18Google obviously has consumer, but it also has its incredibly vast suite of tools that people already use that they can integrate AI into. OpenAI has their vast consumer base to build on, although interestingly, they're obviously now trying to head into similar space as Anthropic's Cloud Code platform with the increased emphasis on codecs. On models, I had OpenAI and Anthropic tied at a nine. My read right at this moment, at least from my personal behaviors and from what I'm observing, is that people are shifting a lot of behavior to GPT-55, whereas the reception to Opus 4.7 has been much more mixed, and so maybe you could argue that OpenAI should be a point ahead here.

19:52But given that we know Mythos is there somewhere, I thought it was better to keep it a tie, at least for right now. Now, moving on to Momentum, it's the category where Google is hurting most right now. They have just had a really hard time breaking into the conversation in 2026, despite the fact that they came into the year with the best narrative positioning that they've ever had. The problem is just that this year has been absolutely dominated by agentic use cases and use cases that are built on top of the coding capabilities of these models. and basically no one is looking to Gemini for that above GPT or Opus.

20:24I gave them a 3 out of 10 on momentum, but this is certainly the area that could pick up the fastest for them. Again, Google I.O. is in just a couple of weeks, and that is their big momentum moment. I almost gave them a higher X factor because of Google I.O. The only reason I held back was that we just got those reports of the Sergey Brin-led strike team that is now working on coding models, and I don't know if that is going to have time to materialize before I.O., And if it doesn't, almost no matter what else Google puts out, if they're not a real contender on coding-based use cases after I.O., I think they're going to continue to struggle, at least in the immediate term context.

21:00Now, a couple other momentum scores worth noting. I put Anthropic at an 8 and OpenAI at a 10. If you look across the course of all of 2026, Anthropic has certainly had the biggest momentum. I mean, just look at the growth in ARR. This jump ahead of OpenAI reflects a very recent term shift around 5.5 and the shifting over of behavior to codecs, which is coming at a time when it can capitalize on Anthropic struggling to keep up with its own demand. I also gave Amazon a score of 6 on Momentum because we're watching them use both their cache and their compute to really throw around their weight, and I think that they're fairly undercounted right now relative to the others, giving them more space to move.

21:38Now other scores among the other labs that I wanted to point out. Amazon, Microsoft, and XAI all got fives on model. The interesting thing is that Amazon and Microsoft's fives are very different than XAI's five. For Amazon and Microsoft, that five is about having access to all the models, but not owning any of them. Whereas for XAI, it's having very competent, but still behind the state of the art of its own models. What that means, though, is that when it comes to picking up ground, I think XAI has a lot more room to rise than either Amazon or Microsoft do. Now, yes, Microsoft does have efforts going into their own internal models, as does theoretically Amazon, although that's been much more quiet recently.