Leveraging AI

295 | “A Profound Moment for Humanity" (Demis Hassabis, CEO Google Deep Mind). Connecting the dots of this weeks news to understand why. May 22, 2026

• Isar Meitis • Season 1 • Episode 295

Share episode

Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.

0:00 | 44:00

This week wasn’t just another wave of AI announcements.

It may have been the week the industry quietly crossed into a different phase entirely.

In this episode, Isar connects the dots behind one of the biggest weeks in AI so far—from Anthropic’s explosive growth, to Google I/O, OpenAI’s legal win, NVIDIA’s record earnings, and Andrej Karpathy joining Anthropic to work on recursive self-improvement.

Individually, each story matters.

Together, they point to something bigger: accelerating AI capability, accelerating infrastructure buildout, and growing signals from the people closest to the frontier that we may be entering a very different era.

The quote that framed the episode came from Demis Hassabis:
 “We were standing at the foothills of the singularity. It will be a profound moment for humanity.”

This episode breaks down what that actually means—and why the implications go far beyond new models and product launches.

In this session, you’ll discover:
 - Why Anthropic’s projected $44B annualized revenue shocked the industry
 - How Anthropic became more profitable per user than OpenAI, Google, and Microsoft
 - Why Andrej Karpathy joining Anthropic may be one of the year’s biggest AI stories
 - What recursive self-improvement (RSI) means—and why labs are racing toward it
 - How OpenAI’s legal win against Elon Musk clears the runway for a potential IPO
 - Why Google’s AI strategy suddenly looks both confusing and incredibly ambitious
 - What Google’s shift from “search” to autonomous AI agents means for websites and SEO
 - Why AI solving an 80-year-old math problem matters more than most people realize
 - How NVIDIA, SpaceX, and compute infrastructure are becoming central to the AI race
 - Why electricity—not chips—may become the biggest bottleneck in AI expansion
 - What Demis Hassabis means when he says we’re at the “foothills of the singularity”

About Leveraging AI

If you’ve enjoyed or benefited from some of the insights of this episode, leave us a five-star review on your favorite podcast platform, and let us know what you learned, found helpful, or liked most about this show!

GMT20260523-154549_Recording_avo_1280x720

Hello and welcome to a weekend news episode of the Leveraging AI podcast. The podcast that shares practical, ethical ways to leverage AI to improve efficiency, grow your business, and advance your career. This is Isar Meitis, your host, and today is probably one of the most impactful news episodes I ever recorded. I actually thought that this episode is going to focus mostly on Google I/O because that was supposed to be the biggest event of the week, and it was the biggest event of the week, but it's not the most important story of the week. And actually, to be fair, none of the stories of the week is the most important story of the week. It's the combination when you connect the dots between all these stories tells maybe the story of a complete change in the speed and acceleration, and as crazy as it sounds, I think we're at the verge of an inflection point in the speed in which we're going to see AI moving forward, which again, sounds completely crazy when we know how fast it's moving right now. So what I'm going to do today is I'm gonna share with you a lot of stories, but mostly I'm gonna share with you the story behind these stories. If you want the narrative that hides in plain sight, if you try to connect the dots of everything that happened this week. So huge week from an implication perspective or where AI is going. I'm going to basically tell you one long story that is going to touch a lot of touch points and things that happened this week, and I suggest you free the time to listen all the way to the end because literally every one of these points that I'm going to mention is critical. We're not going to do any rapid fire. We're not going to do anything else. It's just gonna focus on that main story of the inflection point I think we're at the verge of. With that, let's get started The first thing I want to start with is that Anthropic's numbers just got released. And when I say numbers is their projections for Q2, and their projections for Q2 are nothing short of insane, mind-boggling, mind-exploding, whatever you wanna call it. So first of all, they're anticipating their revenue in Q2 to be ten point nine billion. That's an annualized rate of about forty-four billion. Just to put things in perspective, last quarter, which was an explosive quarter that everybody was puzzled by, they did four point eight billion. So they more than doubled that between Q1 and Q2 to go to an annual rate of forty-four billion dollars in revenue. If that by itself is not insane enough, and it is insane, then they are anticipating to be profitable in Q2. Based on their current projections, they're going to make a net profit of five hundred and fifty-nine million dollars of operating profit by the end of the quarter. This is two and a half, three years earlier than they expected to be profitable based on the projections they shared at the end of twenty twenty-five. So less than six months ago, they thought they will not turn a profit before twenty twenty-nine, and now we're in Q2 of twenty twenty-six, and they're going to turn a profit. Now, there are a lot of ifs and question marks around this, and there's people that are saying they're calculating their top-line revenue different than OpenAI, which is true, but it doesn't change the fact that they more than doubled their revenue for two quarters in a row in billions or tens of billions of dollars, which is absolutely insane. It has never been done in history before at that scale, at that pace. It is hard to double the revenue of a company that makes two hundred thousand dollar in a quarter. going from four point eight billion to ten point nine billion from one quarter to the next is absolutely insane But what I found is some additional very, very interesting information. a research company called Counterpoint Research shared a document they call Global LLM Adoption and Revenue Snapshot Q1 2026. And as the name suggests, They were comparing the revenue success of the different labs. And what they find is absolutely amazing. In Q1 of 2026, Anthropic led the global LLM with 31.4% share of the global revenue of all the AI labs and tools out there. OpenAI came second at 29%. So Anthropic is ahead in revenue. But the really crazy thing about this is Anthropic did this with 134 million monthly users versus OpenAI's more than 900 million monthly users. That's one-seventh of the customer base, and they're generating more revenue than OpenAI from a revenue market share perspective Now, if you do the math, and this is where it gets really crazy, Anthropic average revenue per monthly active users is $16.20. OpenAI is $2.20. We're talking about almost 8X the revenue per user, which means OpenAI has significantly more expenses because they're serving a much larger customer base and generating significantly less money because of that. By the way, on the same scale, Microsoft is at $5 and Google is at $1.10. So if you wanna get the full picture, this is where we are. So Anthropic is at $16.20 per average user, and everybody else is at five or less. With Microsoft, which I wouldn't really consider them a lab, they're just a distributor more than anything else, is at five, and then OpenAI and Anthropic are at 2.2 and 1.1. Now, two really interesting quotes that relate to this topic and summarize it before we switch gears to tell you what else Anthropic has been up to this month, because this is just the beginning. The first one comes from Derek Thompson, who is a journalist who said the following: "Anthropic just had a profitable quarter at forty-four billion annual run rate with a fairly enormous compute shortage that forced them to ration service and push some customers, perhaps just in the short term, into the arms of competitors. I don't think it's crazy to think their annual revenue would be one hundred billion or more with sufficient compute for inference." And I agree. The second quote, which is sarcastic, but I absolutely love it because of that comes from a Twitter user called Mr. Rateable, who hit almost 26 million views on that tweet, and he wrote: "How is it possible for Anthropic to be profitable despite only having more revenue than Workday, ServiceNow, Palantir, and Snowflake combined?" So that's obviously a fun way to look at this, but, you know, we all assumed that this is a losing proposition in the short term and that they're all working towards a long-term revenue. And here we have a company that's generating more revenue than some of the most respectable, successful tech companies in the world, and it's doing it by doubling and more than doubling its revenue every quarter since Q4 of last year. Now we know that Anthropic has a serious compute shortage, and we've been as a serious Claude user myself, I can tell you that I've been hit with that several times from different angles. And I've shared with you just last week that SpaceX is going to give Anthropic access to Colossus-1, which will allow them to more or less double their capacity almost overnight. Well, that's not the end of it Tom Brown, who is the chief compute officer at Anthropic, announced this week that Anthropic is scaling up a 200 gigabytes capacity in Colossus-2 through the month of June. So in addition to having access to all of Colossus-1, they're going to get a nice share out of Colossus-2 to run more Anthropic tokens through one of the most advanced data centers in the world today Now, why is that important? First of all, because Anthropic are short in compute, and that's gonna allow them to do a lot more things that they couldn't do before. The other reason it is interesting is obviously the SpaceX side of all of this. SpaceX is the owner of Colossus-1 and Colossus-2. I shared with you in the episode that launched on May 9th, so just two weeks ago, that I believe Elon's direction is to become a data center, if you want, compute provider versus a model provider. And I think he understands he cannot compete on the model side, but I think he understands that he can make significantly more money by selling compute, especially later on when he can start launching compute to space, which more or less nobody else can do. And so that move definitely shows that this is the direction that SpaceX is moving. The other interesting part of this point is that as we know, SpaceX is on the verge of the first gigantic IPO of this year. There's at least two in the pipeline, and they have agreed. And what we've learned is in the current deal, Anthropic is going to pay SpaceX forty-five billion dollars over the next three years Or if you translate that, that's about $1.25 billion a month this makes SpaceX significantly more profitable than it was before without doing anything other than selling the compute they already have access to. So it definitely serves SpaceX very, very well when it's going towards the compute. And that obviously serves SpaceX very well as they're marching towards their IPO. Now, to put things in perspective SpaceX biggest revenue generator right now is Starlink that generated $11 billion in 2025, and Anthropic's payment is gonna be more than that, more or less immediately. So that's a huge deal for SpaceX as well, and again, pushing Elon and his global plans for everything, uh, further ahead in his quest to be maybe the most important player from a compute perspective, if he can launch into space what he's saying he will. And again, if he gets it off by 50%, he's still gonna be one of the major players when it comes to selling compute to this era. Combine that with the fact that he's one of the very few companies that actually makes their own chips, so for Tesla cars and the Tesla robots, then you understand that he's gonna be a major, major player in the infrastructure of AI in the next few years but these two things weren't even the biggest news coming from Anthropic this week. So the third news from Anthropic this week is that Andrej Karpathy posted the following on X this week. "Personal update. I've joined Anthropic. I think the next few years at the frontier of LLMs will be especially formative. I'm very excited to join the team here and get back to R&D. I remain deeply passionate about education and plan to resume my work on it in time." So who the hell is Andrej Karpathy, and why do I think this is potentially the biggest news of the week and maybe the biggest news of the year so far as far as the impact on where AI is going? So Karpathy was one of the founding members of OpenAI, like when they just got started. He was recruited by Elon Musk to run Tesla Vision, which was a big part of the Tesla AI group that actually allows their car to self-drive. He was the head of Tesla autonomous vision work for five years. He then returned to OpenAI in 2022, stayed there for a while. He was one of the people that led to the crazy jump between GPT 3.5 to GPT-4, so he was one of the leading researchers over there. And then he left in mid-2024 in order to start his own company that he called Eureka Labs that was supposed to be a education company. Now, while we heard very, very little about Eureka Labs and the education efforts they're doing, we still kept hearing a lot from Andre. I shared a lot of stuff with you. First of all, he's the guy that coined the phrase vibe coding right back in 2025 When he wrote, there is a new kind of coding I called vibe coding, where you fully give into the vibes, immerse exponentials and forget that the code even exists." So that's when people are now talking about vibe coding, which is talking in English and getting the AI to generate the code. He's the one that coined that phrase. He's also the one that has released a lot of really cool things in the past few months. The two most interesting things that he shared that we talked about in this podcast is he ran an experiment that allowed AI to train AI. So he was running it as a self-improving loop that allowed a small language models to self-evolve without human interference. And this is exactly why he's joining Anthropic. So here's what Anthropic had to say about this move. This is from Nicholas Joseph, who's one of the executives of Anthropic, and he said, "Excited to welcome Andre to the pre-training team. He will be building a team to focus on using Claude to accelerate pre-training research itself. I can't think of anyone better suited to do it." Now, why do I think this is such a big deal? We've talked in previous episodes on RSI, which is recursive self-improvement. Basically, an AI that improves AI, which then accelerates AI research faster and faster and faster, which brings us to the singularity very quickly. So if you look back to an interview with Dario Amodei on the Dwarkesh Patel podcast just in February of this year, Dwarkesh challenged Dario and said, basically asking him if recursive improvement is real, because that was the topic of the conversation, why does the labs keep on changing positions between them? Meaning they, it seems that they all be dependent on the researchers' capability to come up with new breakthroughs, because if the AI would do it, one lab will break away very, very quickly as soon as they figure out that it's possible in an effective way. Amodei's answer to that was that until very recently, the compounding advantage of AI-assisted AI research was too small to matter, but it is changing. This is a quote. So if... What does that tell us? It tells us that inside of Anthropic, they understood that they are close, that they are missing a few cents for the dollar to figure out how to create RSI effectively that will then leave everybody in the dust behind them because the AI will improve the process of its own next generation, which again leads to singularity very quickly. Andrej Karpathy is one of the OGs of the AI modern space. He, again, one of the co-founders of OpenAI, one of the most sought-after engineers and researchers on the planet. He is not doing this for the money. Again, If you just think about the amount of stock or options he has both in OpenAI and in probably SpaceX from his years in Tesla, With the two upcoming IPOs, money is not the reason he's doing this. He's doing this because he's chasing the Holy Grail, and he is doing this to be the person who creates RSI and is doing this in the lab that right now is most advanced and closest to that as possible. So what does this first chapter of today's episode tell us when we focus more on Anthropic? It tells us that the company that is growing at a pace that we have never seen before is now getting more compute, significantly more compute, and one of the top researchers on the planet that is going to focus on one thing only, which is to create RSI, which will bring us to the singularity. So this is just chapter one of this episode, and you're gonna see a continuing theme that is going in the same direction The second topic that we're going to talk about that aligns perfectly with how things are going to accelerate is the fact that OpenAI won the Musk lawsuit So a very, very quick summary. Elon Musk issued a lawsuit against OpenAI, Sam Altman, Greg Brockman, filed on February 2024, and they basically claiming that they stole a charity that he helped starting, and he sought after $134 billion in damages and the removal of Sam Altman from the board. And on May 18th, a nine-member advisory jury dismissed all of Musk's claims, and Judge Yvonne Gonzalez Rogers threw the case out. Basically, the bottom line is OpenAI won. Why does that matter for acceleration? Because if they would have lost, that would have put some very serious roadblocks on their way to their IPO, which would have put them in a very, very serious crunch from a cash flow perspective, which would have put an entire industry and supply chain in the same cash flow crunch, which would have had a crazy ripple effect through the AI space. Probably would have helped Anthropic a lot. I have a feeling I know what the outcome they wanted to see, But the outcome is that Musk lost and OpenAI won, and they are moving forward with their IPO. So the Wall Street Journal reported this week that OpenAI has engaged investment bankers and they expect to file their IPR paperwork as soon as this week That puts them on a timeline and a plan for to be IPO ready by September, which based on the current timelines we know from Anthropic, they were planning to go public in October. So that puts OpenAI ahead of Anthropic IPO when they're going public Now, putting things in perspective, what we are expecting in the next six months is three IPOs north of a trillion dollars, and some people say north of $2 trillion. We're gonna get to that in a minute. So we are talking about three gigantic companies, all three accelerating like crazy, all three riding their recent success in the AI space, and obviously SpaceX running their launch capabilities as well. But what we are looking at is three very different company from a financial perspective. SpaceX is obviously a launch company more than they are an AI company, even though they are seeing huge revenue right now from selling compute. But we will see if they can continue scaling that, but the fact that they're planning to do that is definitely gonna help their IPO. So they are probably the weirdest one because they're not a lab developing a model, even though they are with Grok, but I think that's not gonna be their focus in the next few years. Between Anthropic and OpenAI, we see two very, very different companies. We do not know the recent numbers from OpenAI because they kept it very close to their chest. But the last thing we know is that they were looking at a $29.4 billion projected annual income. I'm sure that has changed since, up most likely, but we don't know any numbers since then. that is compared to Anthropic's current $44 billion in revenue. But the more interesting thing is that OpenAI are expecting to lose $14 billion this year. They're expecting to spend over $100 billion each year in the next few years on compute, and in some cases they're expected to spend over $100 billion in compute every year in the next few years versus tens of billion by Anthropic. and they're not expected to be profitable before 2030 when Anthropic is profitable this month. I don't think it's gonna continue to being profitable, but I don't know. But the fact that they were able to have one profitable quarter so early puts them in a very, very different ballpark Or if to quote Connor Sen from Bloomberg, he said, and I'm quoting, "The Anthropic IPO won't be for less than two trillion." So it is very clear why OpenAI wants to be ahead when it's going to the markets. Investors will have to figure out how they put their chips correctly in the most effective way. I'm sure all the big investors will play a role in all the three big IPOs. The question is, in what percentage to each one? And have a feeling that OpenAI are terrified that if Anthropic goes out of the gate first, there's gonna be significantly less money left on the table for them to actually rack up from the chips that's gonna be pushed their way. But that's not the story. The story we're talking about is story of acceleration, and the fact that we're going to have three companies that are in the lead in the AI space doing different things that are going to have additional trillions of dollars in cash available to them to invest in new things is telling you where this is going. This is going to be the largest IPOs in history, all three happening in the same year. The revenues from most of it is going to go to AI acceleration Now, another point coming from OpenAI that was kind of like hiding in the side of this news this week, but is again showing the direction where we're going. On May 20th, OpenAI announced that one of its internal models disproved an 80-year-old conjecture in math in discrete geometry. So it's called the planar unit distance problem, which was posted by a mathematician called Paul Erdős in 1946 Now, we've heard before of AI models solving different problems. The biggest news was seven months ago when OpenAI claimed that GPT-5 solved 10 of Erdos problems, but it turned out that it didn't actually solve them. It just surfaced solutions that already existed in different places in the internet. This one is different, and it's different in a huge way from two different reasons. One, it is a general purpose reasoning model. It is not a system that was trained specifically to solve these kind of problems. It didn't have any special scaffolding or proof strategies to go and tackle this. The second thing is, which is even more interesting, is that it reached into algebraic number theory using infinite class field towers and a 1960 theorem that's called Golod-Shafarevich to crack the problem in a different field of math. So it went from one field to another field, combining two different things together in order to solve a math problem. This is very, very unique, and what it is showing us, it is showing us that AI is now, not just in the future, capable of achieving new scientific discoveries by combining things that are very difficult to do for the human mind. This is a model right now that is a general model that wasn't built specifically for that, which again, showing you the state of current AI and how good it is, and combine that with where we're going, you understand the trajectory and what might be possible. We're gonna talk more about what might be possible when we talk about Google, and more specifically, Demis Hassabis Which leads us to Google, which is a great segue, so let's talk about Google. Google had their largest conference of the year this week. If you remember last week, I told you that they already made a very big announcement which told us we're gonna get a lot of announcements, or a one huge announcement in I/O. So there weren't one huge announcement, but there definitely were many, many, many small ones to the point it was really confusing and not necessarily easy to follow. But the really important thing, going back to the acceleration numbers, is their usage numbers. Again, these are self-reported by Google, but they've always been self-reported, so let's take it through that. So Google is now processing over 3.2 quadrillion tokens per month. That is a trillion plus a zero for those of you who don't know what that means That is a 7X year-over-year growth in a very, very large volume of token usage. Now, to put things in a bigger perspective, two years ago, it was 9.7 trillion per month. Last year, 480 trillion, and now 3.2 quadrillion. This is absolutely insane. Just their API alone is currently processing nineteen billion tokens per minute The other interesting parameter is how fast the Gemini app has been growing, and it's now has over 900 million monthly active users, more than doubled since just a year ago, and the daily request is up 7x from a year ago They also shared that 8.5 million developers work with Google models monthly, and that they have currently in the enterprise and company space more than 375 customers that is processing more than one trillion tokens in just this past year. So that tells you again that the demand for AI is just accelerating more and more and more across all these different platforms. It doesn't matter where they're coming from, and Google is definitely one of the leading players Now, what did they announce in Google I/O? Well, they announced Gemini 3.5 Flash, which is their latest model, but it is a Flash model, right? It's the smaller brother. It's not the Pro Gemini 3.5 that will probably come in the next few weeks. It usually works the other way around. Usually, the companies release the big model, and then they release a distilled smaller model. But this time Google did the other way around. So Gemini 3.5 Flash is currently available. It is close to the frontier models from the other labs while being significantly faster, but as we've learned very quickly after the release, not necessarily much cheaper. So if you compare it to the previous Flash models that we had, this one is faster and not cheaper. The previous ones were faster and cheaper. Beyond the fact that it is not that cheap per token, it is also very verbose, which means using it is gonna use a lot more tokens to get to the same outcome. So it turns out to be a solid model that runs very, very fast, but not necessarily very efficient from a cost perspective and from a token generation perspective. They also generated-- Google also released Gemini Omni, which is a multimodal generation model that can get basically any input and generate any output in a very, very effective way. When you look at what they demoed, it looks very much like a video generation tool. But the reality is, it is a lot more than that. it is the mother of all multi-model models that we have right now that can generate incredible videos, that can edit videos, which by itself opens a huge range of use cases. You can take a previous existing video, whether a real video or a AI-generated video, and edit it with simple prompts while doing incredible things in a very accurate way. This builds on the deep and interesting work that Google has been doing in world models. More on that in a minute and why I think this is important. Now, is that competing with Veo? Is it competing with, uh, Flow? Like, where does that sit exactly in the overall offering of Google is a little confusing and a little unclear, and that is going to be the pattern with the next thing that I'm going to talk about. They also introduced Gemini Spark, which is a 24/7 personal AI agent that can help individuals and potentially afterwards business people in the day-to-day with everything that you do. Where exactly that sits in the overall offering is not very clear. They also introduced AI mode, which is a complete new redesign on the concepts of how search happens through the internet combined with information agents in search to generate basically what you want. So what they are saying is that search in the immediate future, I guess, is gonna be ruled by agents versus the old Google search. You can now set your own agents to search for you, but not just as a one-time thing. You can have them search for you continuously and find the relevant information, sort the relevant information, analyze the relevant information, and provide you relevant answers based on your internal company information as well as well as external world information, all by using these agents. This could be for shopping. You are looking for something specific. The agents will look for that for a specific price or with specific sets of features, and will be able to show that. They'll be able to show it continuously. One of the examples they gave had to do with searching for an apartment. So when you do a search today, you search once. You search for a two-bedroom, one bathroom in that kind of price, in this kind of area, and if you don't find it, you will try again tomorrow or within a week. With the agents, they will continuously look for it, and when they find something that's relevant, they will let you know and provide you a brief together with other briefs that you can ask for that they will continuously harvest data for and will provide you that information. They also announced that Gemini 3.5 Pro, the big brother of the Gemini 3.5 Flash, is in internal testing and is going to be released soon. And they announced that they are getting into the intelligent eyewear game. You all know that Meta have these glasses. You all know that there are several different good contenders. We all know that Apple is developing something. We all know that OpenAI is developing something, not sure if it's glasses or something else. And now Google will be back in the game. If you remember, they were the first one who did anything like this, with the Google Glass that was worn by several different super geeks in Silicon Valley and about that's it. But I think that's gonna be very, very different. This is going to be in collaboration with Qualcomm for the chips and Samsung for some of the hardware, some combined with top leading makers of glasses to make it look cool and trendy, and that is coming next year. And from my perspective, the most expected announcement that I thought is gonna be the biggest announcement is that they're also releasing Anti-Gravity 2.0, which is their vibe coding platform, which I thought is gonna be the main event, but it wasn't because they released all these other things. Why did I think that was the main event? Because this is what took Anthropic from where they were at the end of last year to where they are right now. This is what have pushed OpenAI to kill Sora and to focus back on enterprise instead of end consumers. And I had a feeling Google is gonna go in the same direction and reduce the amount of noise and focus on something that is very specific, and that is not the case. But they did release that. There is a new capable model, again, Gemini 3.5 Flash, that is good at writing code. Not as good as GPT 5.5 and not as good as Claude 3.7, but it is coming pretty close, and it works significantly faster together with a new harness called Anti-Gravity 2.0, which allows people to develop in the Google ecosystem while spending less money than with the leading contenders By speaking of writing code efficiently, Cursor released Composer 2.5, which is their latest coding model that is significantly more efficient than the other labs while generating equal or better results. To put things in numbers, Artificial Analysis in their initial testing of this has rated it third on their coding agent index, just behind Opus 4.7 Max and GPT 5.5 Extra High, but ahead of both Opus 4.7 and GPT 5.5 on their medium settings, and doing this between 10X and 60X lower cost So again, this is a way to write new code that can accelerate research, that can accelerate AI capabilities, that can be done at a significantly lower cost than doing this with the models that we're using right now. But going back to Google, and where does that put us in the whole thing? The first thing is it looks very, very confusing. It looks like Google lost their way again. So if you remember, Google started the AI race completely behind it. Because if you remember how the AI race started, it was Google starting it all, right? So the research paper about attention is all you need with a transformer architecture is something that Google invented. They were very, very far ahead of everybody else when it comes to research. And then OpenAI came out with an app called ChatGPT, and that changed everything. And Google were caught off guard by that. And if you remember in the beginning, they released Bard. It wasn't even called Gemini in the beginning, and it was a complete catastrophe. And then Google finally started to get their act together and released Gemini, which was supposed to take the world by a storm. And then in the beginning it shared advice such as, if you want the cheese not to fall off your pizza, combine it with glue and it will stick better," and stuff like that. So this was the beginning of Gemini. But in 2025, it seemed very, very clear that Google found their mojo and they got their act together and they were running very well and very focused and were releasing the right products and were driving amazing growth. So I assumed we're gonna see more of that. And instead we got this crazy wide range of tools that is very, very unclear where you're supposed to go to do different things because now they have, I don't know, 25 different places and tools and concepts and ideas that you can engage with in their ecosystem. Now, there's several different ways to look at this. Option number one is to say that Google do not know what they're doing. I seriously doubt that. Option number two is to say that Google is just aiming for a different market than OpenAI and Claude right now, and they're seeing significant growth with what they're already doing, and they just wanna provide more tools to more people, and they don't really care that it's a little confusing because different people use different things, and each person will find what they need in a different aspect of this. But I think there's a third explanation to what's going on at Google,

And what I would like to do, I would like to share with you a quote from Demis Hassabis, the CEO of Google DeepMind from the Google I/O event this week. When we look back at this time, I think we all realize that we were standing at the foothills of the singularity. It will be a profound moment for humanity."

GMT20260523-154549_Recording_avo_1280x720

So this is basically one of the smartest people on the planet today when it comes to AI, the person that dedicated his life to AI, the researcher I personally trust the most when I follow because there's never any fluff. The person that says that he wishes that the AI was not released to the public so they can focus on building things that can change the universe and humanity for the better, solving cancer, solving other diseases, solving world hunger, stopping global warming. These are the things that Demis cares about. When he's saying he believes we're in the singularity, I pay attention. And why do I think that might connect to what is happening in Google and all the different things they're releasing? I think Demis doesn't really care much about how many users use Gemini's tools. I think what Demis cares about, and I don't think he's saying that out loud, is helping humanity, saving the planet, and things like that. And I think he's developing things in the back end that he thinks will get him and us as society and humanity closer to that goal. The outcome of that are different capabilities. So he's developing world models, so he can potentially understand exactly how cells work and how really big things work around the world. And if he can do that, he can potentially solve, again, any disease and potentially global warming. The outcome of that is a model that understands the world better that can now generate videos and edit videos. So my feeling is that Demis is just doing behind the scenes on the research side, whatever he thinks will get us closer to helping humanity. And the outputs that we're getting are different variety of tools that can use the capabilities that they are researching and developing behind the scenes. And hence, it looks like there's no real clear strategy on them when the real strategy has nothing to do with us and how we engage with AI, but what Demis thinks AI needs to be used for. Now, am I right? I don't know. Sadly, I've never met Demis. I never had a chance to interview him. I would love to do that if anybody knows a way in, and I would love to ask him exactly that question on what's going on in Google right now compared to what he believes, AI should be used for. But again, the bottom line is we're seeing huge amount of change, including how we are doing search. Think about how we're using the internet. The core thing that we take as second nature in the last twenty years, this is changing, and it's changing right now. And the company that is synonym with search, Google, is now changing how we're going to search. They're moving from human searching to agents searching, and that is happening right now, and that has profound implications on everything, including how people visit websites. Are people visiting websites? Does user interface matter anymore? And all these kind of questions. All I can say is something I've said multiple times on this podcast in the last year and a half, and maybe longer, is that if a majority of your traffic depends on users coming to your website, if your conversion on what you're selling comes from building the best converting landing pages on the planet, you need to start thinking very, very hard on what is your next play because that is going away. It may not be going away this year, it may not be going away next year, but sometime in the next three to five years, that will go away. And if your company depends on that, you better find a solution and find it fast Now, the one company we didn't talk about yet from the really, really big heavy hitters of the AI space is NVIDIA. And NVIDIA just shared their earning for the quarter, and as every previous quarter, they have completely crushed the estimates by the major investors and analysts. So their revenue hit eighty-one point six billion for the quarter. The estimates were seventy-eight point nine billion. Same thing in earning per share, one point eighty-seven dollars versus one point sixty-seven dollars. The data center revenue grew at ninety-two percent pace, up twenty-one percent from just the previous quarter Now, this was the first quarter NVIDIA ever reported the split between the revenue coming from hyperscalers versus everybody else, and hyperscalers are accounting for only, and that actually surprised me, 46% of the total sales of NVIDIA But NVIDIA did share that they are gaining percentage as we are moving forward Now, NVIDIA is doing this while currently selling zero chips to China because of the limitations from the current administration. We're going to talk about the administration in a minute and how that ties to everything So you have the company that sells the majority of the infrastructure for the AI era beating expectations every time, time and time again, even though the expectations are stupidly high and they're anticipating crazy growth that doesn't make any sense. They manage to beat that, which again is showing you the amount of demand there is right now across the board from everything AI, from the end users, the number of growth in the users, the number of growth in tokens usage, the number of growth in capacity that is being generated, and so on and so forth. So where does that put us? It puts us in a situation where the smartest, more capable researchers in the world are telling us we are at the verge of the singularity, which basically means the point where AI keeps accelerating upwards very, very fast without the need for us to help it accelerate, And it's just gonna go straight vertical up in its capabilities, and it will be able to do good and bad. It will just be up to us or it, I'm not sure, how to actually use these capabilities. There are going to be very few things that are going to slow it down because once you figure out RSI, then you just need to give it enough compute. And if you give it enough compute, which will be the priority for the labs who will figure it out, it will then accelerate in a speed that will allow to do things that we can't even imagine. Again, both good or bad. Now, what things could slow this down? The first thing is government and regulation, which is one of the stories that I wanna end up on. So I shared with you in the past few weeks that there's potentially a new AI executive order coming from the Trump administration, and the goal of that was to set up a framework, most likely a voluntary framework, in which AI companies share their models in advance with the government before a public release. There was a lot of back and forth. The government said 90 days. The lab said 14 days And it was on the table, and then it was off the table again, and then it was back on the table And then just hours before the expected signing of this executive order, it was pulled back by President Trump, who said, and I'm quoting, I postponed it. I think it gets in the way. You know, we're leading China, we're leading everybody, and I didn't want to do anything in the way of that lead." There's a lot of rumors on exactly what happened behind the scenes, whether it was the push from the labs themselves, whether it was the fact that they couldn't get the leading heads of the leading labs to participate in the signing event of this new executive order, or that David Sacks, that is Trump's AI advisor, is the one that sacked it and basically told him not to do it. The bottom line is, right now it is not happening. I said that multiple times, that I think the one government is not the solution. This has to be an international, well-coordinated campaign that will include governments from all around the world, starting with obviously US and China, combined with people from the industry and combining from people from academia to work together to figure out how to deploy AI successfully for the betterment of humanity and the planet versus the other way around. So if the government, for now, is not going to slow it down, what are the things that might? I think there are two things that are going to slow it down. One is just friction. It is not easy to deploy AI across everything we know. It takes time. I just came back from speaking at a conference called TechCon that was put together by an amazing company called, CET up in Minneapolis, and everybody in that room came to learn how to deploy AI effectively. And it is very clear that there's a lot of roadblocks. Many companies don't have real access to the data. Many companies have access to the data in ways that are not easy for AI to consume. There are huge issues of data security and governance, and so on. So this is going to slow down the effectiveness of AI and its ability to take over more and more components of at least the business world. So that is a little bit of good news, I think, because it's going to slow down somewhat the ability to use the capabilities of AI, regardless of how good and fast it is moving. The other aspect is electricity. So an interesting piece of news that I found that was hiding in the sidelines, but I think is very important for this conversation, is that Kenya just stopped Microsoft $1 billion data center plan because they cannot afford to give it the electricity that it needs Now, yes, we are talking about Kenya, maybe not the most developed country in the world, but I wanna put things in perspective. The proposed one gigawatt Microsoft data center would consume approximately one-third of Kenya's total national installed electricity capacity Now, how does that relate to the US? it relates to the US because, yes, the US has significantly better electrical capacity than Kenya. That's very obvious. But at the end of the day, you're not installing a data center in the US. You're installing the data center somewhere in a specific city, in a specific county, in a specific town. And when you do that, you are tapping into the electrical grid of just that area. And just that area will not have the same capacity, in most cases, as Kenya, which means there is a very significant shortage of electricity compared to the amount of compute that is going to come online this year. I shared with you before that Elon Musk is expecting that by the end of this year, we will have more and more GPUs that cannot run because there's not enough electricity to power them. So while, yes, GPUs and compute are a bottleneck right now, the bigger bottleneck and the one that is harder to actually resolve is the electrical aspect of this, because it is a lot harder to scale the generation of electricity than to scale the generation of compute, which is already very, very hard to do. So where does that put us for this episode is that things are going to get even crazier than we're seeing right now. The acceleration rate is gonna go even faster than it is right now. It is going to be coming from all sides, and there's very few things, but they still exist, that are going to slow it down. There are more news this week that you can find in our newsletter, and I just want to remind you that we are currently selling the last few seats for the August cohort of our multi-agent orchestration course. We sold out May and June and July, and now we're selling August, and it's already selling fast. If you want to understand how to implement AI effectively while using agents in your business to automate more or less everything in the digital aspect of your business, you cannot stay behind. This is very, very clear, again, from the conference I was just in, that the gap from the people who know how to do this effectively and those who don't is growing, and it's growing fast. The train left the station. You can maybe jump in the last cart and get on board or stay behind, and you do not want to stay behind. So come join us. It's a really amazing course. It is extremely effective at getting you from wherever you are in your journey to the point that you understand how to implement AI skills, AI agents, and combine them together with your current tech stack to automate everything in your business. That is it for today. We'll be back with another detailed how to do something in AI on Tuesday. And until then, have an amazing rest of your weekend.

Head Shot

Isar Meitis

Host