JUNE 4, 2026 | YOUR DAILY AI INTELLIGENCE BRIEFING
ANTHROPIC HEADS TO WALL STREET: THE CLAUDE-MAKER FILES FOR IPO AT A NEAR-TRILLION DOLLAR VALUATION AND PUTS AI STOCKS ON NOTICE
Anthropic, the company that makes Claude, the AI you are probably using right now to do half your job without telling your boss, just filed for an IPO. Confidentially, of course, because in Silicon Valley even going public has to be a vibe.
Here is what you need to know. Anthropic filed a draft S-1 with the SEC on June 1, 2026. No share price. No number of shares. Just the paperwork that says we are thinking about letting civilians buy in. The company is currently valued at $965 billion, which is not quite a trillion but is close enough that you can smell it from here. They got there by raising $65 billion in a Series H round the prior week, which sounds made up but is not.
The revenue numbers are the kind that make venture capitalists text their wives to say they might be late for dinner. Anthropic annual revenue run rate hit $47 billion in May 2026, up from roughly $10 billion a year ago. That is not a typo. That is a company that grew its revenue by almost five times in twelve months while also, by the way, being founded by a bunch of former OpenAI researchers who quit in a group drama that would make excellent television.
Dario Amodei, the CEO, originally positioned Anthropic as the safety-focused alternative to OpenAI. The company was going to be the responsible one. The grownup in the room. The lab that cared about whether AI killed everyone before figuring out how to monetize the killing. And to their credit, they have kept that positioning even as they sprinted toward a trillion-dollar valuation, which is a neat trick if you can pull it off.
What makes this IPO particularly interesting is the competition lined up behind it. OpenAI is also expected to go public. SpaceX just filed for its own IPO targeting a $2 trillion valuation. So we are about to enter an era where the biggest names in the AI race are simultaneously on public stock exchanges, which means every quarterly earnings call will be a live referendum on whether artificial intelligence is actually going to pay off or whether we are all participating in the most expensive science project in human history.
The investors include Sequoia Capital, Dragoneer, Altimeter Capital, Capital Group and a few dozen other firms that got in early and are now sitting on returns that would make most hedge funds cry into their Bloomberg terminals. There is no pricing date set and no confirmation of when shares will actually trade. But when they do, it will probably be one of the biggest financial events of 2026 in a year that is already stacked. The AI economy is no longer a startup story. It is now a Wall Street story, and Wall Street is paying very close attention.
Source: TechCrunch
MICROSOFT SAYS THANKS BUT NO THANKS: SEVEN NEW IN-HOUSE AI MODELS TAKE DIRECT AIM AT OPENAI AT BUILD 2026
Microsoft spent years being the company that was very publicly in a committed relationship with OpenAI. They invested $13 billion. They integrated GPT-4 into everything from Word to Bing. They basically told the world that their AI strategy was whatever Sam Altman is selling, we are buying all of it.
That chapter appears to be ending.
At Microsoft Build 2026, held in Seattle this week, the company unveiled seven new AI models built entirely in-house under the MAI brand. Not one. Not two. Seven. And these are not hobbyist experiments nobody will use. MAI-Thinking-1 is a 35 billion parameter reasoning model with a 256,000 token context window that Microsoft says outperforms Claude Sonnet 4.6 in blind testing and matches Claude Opus 4.6 on coding benchmarks. That is a direct shot at the company Microsoft has been paying billions of dollars to not have to compete with.
The rest of the family fills in the gaps in ways that matter. MAI-Code-1 is already live inside GitHub Copilot and VS Code, meaning millions of developers are using it right now without necessarily knowing it. MAI-Image-2.5 handles image editing. MAI-Transcribe-1.5 covers speech-to-text across 43 languages. MAI-Voice-2 adds new voices in over 15 languages. Collectively these seven models cover almost every major use case the big AI labs compete on.
The message from Redmond is clear. Microsoft is no longer content to be a distributor of OpenAI work. They want their own models, their own infrastructure, and their own AI identity that does not depend on a partnership with a company now raising money at an $852 billion valuation and looking increasingly like a rival. The Azure business can now run entirely on Microsoft-owned models without touching OpenAI APIs at all. That shifts the financial equation considerably. Every token an enterprise runs through a MAI model instead of a GPT model is a token Microsoft keeps more of.
For developers and regular users, this is genuinely good news even if the corporate chess is the real story. More frontier-level models competing in the same market means better pricing and more options. MAI-Thinking-1 being genuinely competitive with Claude Opus keeps everyone honest, and an industry where Microsoft is building real models from scratch instead of reselling someone else’s is a more interesting industry than the one we had last year. The era of one lab dominates and everyone else distributes is over. The era of everyone building their own everything has started.
Source: Microsoft Build 2026
BOSSES GOT PLAYED BY THEIR OWN AI PITCH: STUDY FINDS COMPANIES THAT FIRED EVERYONE FOR AI BOTS ARE NOT MAKING ANY MORE MONEY
Okay so here is a story that is equal parts schadenfreude and genuinely useful information. Gartner, the research firm that companies pay enormous sums of money to tell them things they should have figured out themselves, just released a study with a conclusion that stings: companies that fired workers to replace them with AI are not making more money. They are just companies with fewer workers and the same problems they started with.
The study surveyed 350 businesses with revenues above $1 billion, all actively deploying AI automation. About 80 percent had reduced their headcount as a result. The return on that move? Elusive. Companies that cut the most workers were just as likely to see negative outcomes or marginal gains as they were to see any meaningful return on investment. The conclusion from the lead analyst: workforce reductions may create budget room, but they do not create return.
Which, if you think about it, is obvious. If your sales are bad because your product is bad and your customer support is a disaster, firing half the support team does not fix any of those problems. It just means your support is now handled by an AI that fails about 70 percent of the time, which is what an earlier Gartner study found about AI agents operating on real office tasks. That number bears repeating: AI agents fail at standard office work roughly 70 percent of the time. And yet companies have been laying off the humans who do those tasks and replacing them with the 70-percent-failure bots.
The 2026 tech layoff count stands at over 142,000 workers cut, with a significant portion attributed explicitly to AI adoption. Meta alone cut 8,000 employees in May. Intuit axed 3,000. Multiple other companies cited AI as the public justification. Sam Altman acknowledged this week there is some AI washing where people are blaming AI for layoffs they would otherwise do anyway, which is a remarkable thing to say out loud about your own industry.
The companies actually seeing strong results from AI, according to the Gartner research, are doing the opposite of cutting headcount. They are investing in workers who use AI as a tool, building new roles, developing new skills, and designing workflows where AI amplifies what people do instead of evicting people from their jobs. The organizations with the best ROI are not the ones who replaced humans with machines. They are the ones who pointed machines at problems and let humans steer.
Gartner longer view is oddly optimistic: autonomous AI will eventually create new job categories that do not exist today, probably starting around 2028 or 2029. So the people who just got laid off have a two-to-three-year gap to fill before the promised new jobs arrive. Sounds very manageable.
Source: The Register
GOOGLE JUST KILLED THE BLUE LINK: AI MODE HITS ONE BILLION USERS AND THE INTERNET AS YOU KNEW IT IS OFFICIALLY OVER
One year ago, Google launched AI Mode in Search and the internet erupted into two camps: one saying it would save Google from irrelevance, the other saying it would destroy every website that depends on search traffic for survival. Both groups were, it turns out, correct at the same time.
The one-year scorecard: AI Mode now has over one billion monthly users. AI Overviews, the broader version that appears across regular search results, is used by 2.5 billion people per month. Queries in AI Mode have more than doubled every single quarter since launch. This is not a niche product anymore. This is the main product.
At Google I/O this week, the company announced what it called the biggest upgrade to Google Search in over 25 years. That phrase should make you sit down for a moment. Google built its entire business on a search box and a list of blue links. That was the whole product for a quarter of a century. And now they are replacing it with something that answers your questions directly, builds custom tools for you on the fly, and creates personal agents that monitor the web around the clock on your behalf.
The new default model inside AI Mode is Gemini 3.5 Flash, which Google describes as frontier-level intelligence at four times the speed of comparable models. You can now create what Google calls information agents that act like personal research assistants living inside your search bar, automatically surfacing updates on topics you track and watching for new developments while you do other things.
The implications for anyone who makes content for the web are severe and already arriving. Search traffic has been the oxygen supply for most of the internet for twenty years. When Google answers the question directly instead of sending you to a website, the website does not get the visit. Publishers and journalists and bloggers and independent creators all built their businesses on the assumption that Google was a traffic source. Google just quietly finished converting itself from a referral partner into a competitor for that audience attention.
DuckDuckGo reported a 30 percent increase in installs in late May from users actively trying to avoid Google AI search results. Which means the product is now large enough that people are fleeing it specifically, which is the kind of success metric that only makes sense in tech.
The question nobody has a clean answer to is what the internet looks like when the system that used to drive people toward content now reads the content for them and summarizes it. We are about to find out, because a billion people already switched and the rest are on their way.
Source: Google Blog
GOOGLE ENGINEER USED COMPANY SECRETS TO WIN $1.2 MILLION ON A BETTING WEBSITE AND IS NOW STARING DOWN FEDERAL CHARGES
This is the kind of story that makes you wonder whether Silicon Valley has collectively lost its mind, and then you remember that this is Silicon Valley and the answer was always going to be yes.
Michele Spagnuolo, a 36-year-old Italian software engineer who worked in security at Google, has been charged with commodities fraud, wire fraud, and money laundering. The allegation: he used his access to confidential Google internal data to place winning bets on Polymarket, a platform where people wager real money on real-world outcomes, and walked away with $1.2 million in profit.
Here is how it worked. Polymarket runs markets where users bet on things like who will be the most searched person on Google in a given year. These markets are actively traded and depend entirely on nobody knowing the answer before Google publishes the results publicly. Spagnuolo allegedly knew the answer before the public did because his job gave him access to the internal data from which the Year in Search report is compiled long before it goes public. He then opened a Polymarket account under the name AlphaRaccoon, a detail prosecutors included in the complaint and that will now live in federal court documents forever.
Between October and December 2025, he wagered approximately $2.75 million across multiple bets linked to this data. He won about $1.2 million. He was caught, presumably because the combination of working at Google in a data-adjacent role, having access to exactly this type of internal information, and then betting correctly and repeatedly on exactly this category of market is not a coincidence pattern that requires a genius detective to spot.
He appeared before a federal magistrate in New York and was released on a $2.25 million bond. He is now the second person this year to face criminal insider trading charges related to Polymarket. The first was a US special forces soldier who allegedly used advance knowledge of a planned military operation to place bets. The emerging category of crime here is knowing something real and betting money on the fictional version of knowing it, which is both a new legal problem and a very old human problem.
The legal hook is commodities fraud, which is interesting because prediction markets occupy a genuinely gray regulatory zone in the United States. The government is treating Polymarket bets like commodity trades, giving federal prosecutors jurisdiction they need. Whether that legal theory holds up is something courts will eventually decide.
The lesson here is not do not use insider information to bet on prediction markets. The lesson is do not use insider information from a company with detailed access logs to bet on a market that any attentive investigator would notice. AlphaRaccoon had no chance.
Source: The Register