Artificial Intelligence

[phkrause]

💸 AI's midterm money machine

Credit: Rebecca Steele of Toyota Motor North America. Data: FEC, OpenSecrets, media reports. Chart: Axios Visuals The chart, from Axios CEO Jim VandeHei's weekly C-Suite newsletter, captures how AI went from irrelevant to dominant in D.C. in one election cycle. Andreessen Horowitz and its co-founders have pumped $116 million into the midterms so far — more than any other donor, according to The New York Times (gift link).

[phkrause]

San Francisco turns to AI to avoid collisions between ships and whales searching for food

Ferries, cargo ships and tankers cut through choppy waters in San Francisco Bay as a whale surfaced nearby, its spout barely visible against the whitecaps. Until now whales could easily go unnoticed by mariners, but an AI-powered detection network launched this week is designed to track them day and night. The effort comes amid an alarming rise in gray whale deaths in the bay. Read more.

[phkrause]

Reclaim thinking in the AI age

Illustration: Sarah Grillo/Axios. Stock: Getty Images

A just-departed executive at one of the world's most influential wellness companies is giving three pieces of advice for preserving our humanity in the AI age. David Ko, who was CEO of the app Calm for the past four years, recently stepped down to pursue a new, undisclosed path focused on guardrails for kids using AI. "If we didn't have AI, I would not leave," Ko said in an April interview with Axios' Amy Harder at the Palo Alto headquarters of Calm, a couple of weeks after he announced his departure. "If we weren't having AI conversations today, I'd still be the CEO of Calm." Calm, founded in 2012, is a mental wellness app with more than 180 million downloads. ⚠️ Zoom out: Ko sees AI moving faster than social media, with potentially similar risks for young people. As a father of two daughters — a teen and a college student — that urgency is driving his pivot. "Let's not make that same mistake twice with something that's moving twice as fast," said Ko, who'll continue advising Calm's board. Here's Ko's advice for using AI mindfully: 🧠 1. Protect your critical thinking. Don't accept the initial output from an AI model. Two reasons: You'll get better results. And you'll protect your own mind. Referring to his younger daughter's AI use, Ko added: "She'll go a few rounds and push back on the AI tool." Zoom in: Even if AI delivers instant answers, kids still need to learn how to think their way to them. Ko said critical thinking should be treated like physical education — good for its own sake: "Isn't it good for children to exercise? Of course it is. So why is it good to give kids all the answers?" ⏱️ 2. Be intentional with your "found time." AI is making people more efficient at a rapid pace, which makes the time savings more visible. Sometimes you should use that time to do more work, but it shouldn't be the subconscious, reflexive answer. ✨ "If you find yourself saving time, then find yourself doing things that give you joy," Ko said. 🔁 3. Skip the big talk — go small. "I find often that people try to have the one heavy conversation" with kids, Ko said. "I find that rarely works." Instead, he suggests frequent, low-stakes conversations — often sparked by what kids are already seeing, like a TikTok about AI.

[phkrause]

2 hours that changed AI

Illustration: Brendan Lynch/Axios

Over the course of two hours yesterday afternoon, the AI industry produced an extraordinary stream of headlines mapping the vast architecture of its ambitions, Axios' Zachary Basu writes. Why it matters: One historic news cycle peeled back virtually every layer of the AI revolution — smarter systems, exploding revenues, roaring markets, staggering infrastructure demands and a federal government racing to catch up. Zoom in: Yesterday's developments crystallized the core pillars of the emerging AI order. 🧠 1. Brains: OpenAI announced one of its general-purpose reasoning models autonomously cracked a famous geometry problem that stumped mathematicians for 80 years. The implications are enormous: An AI capable of original mathematical discovery could unlock breakthroughs across science, engineering and medicine. 📈 2. Demand: Anthropic's explosive growth has the company on track for its first profitable quarter, with revenue projected to more than double to $10.9 billion in Q2, The Wall Street Journal reported. The estimated $559 million operating profit arrives two years ahead of internal projections — a turning point for an industry whose spending has long outrun its earnings. ⚡ 3. Power: Anthropic expanded its compute partnership with SpaceX, agreeing to spend roughly $1.25 billion per month through 2029 for access to the company's Colossus supercomputing infrastructure. Ahead of SpaceX's IPO next month, the deal cements Elon Musk as a power broker in the AI economy, where access to compute infrastructure is becoming as strategically valuable as the models. 🤖 4. Chips: Nvidia delivered another monster quarter, posting $81.6 billion in revenue — with its data center business alone bringing in $75.2 billion. CEO Jensen Huang said demand has gone "parabolic" and described the AI boom as the "largest infrastructure expansion in human history," reinforcing Nvidia's dominance over the hardware powering this economy. 💰 5. Wall Street: SpaceX's long-awaited IPO filing (see Item 2!) revealed a company rapidly transforming into an AI infrastructure giant, with massive compute operations and ambitions stretching far beyond rockets and satellites. The blockbuster filing arrived as OpenAI and Anthropic race toward potential public offerings this year that could value all three companies in the trillions. Also yesterday afternoon, The Wall Street Journal popped the news that OpenAI is working with bankers on a draft IPO prospectus to be filed confidentially with regulators as soon as tomorrow. 🏛️ 6. Government: President Trump has summoned top tech CEOs to Washington today for the signing of his AI executive order, prompted by mounting cybersecurity fears. CEOs on the invite list, Axios' Ashley Gold learned, include Musk, Alphabet's Sundar Pichai, Amazon's Andy Jassy, Meta's Mark Zuckerberg, Microsoft's Satya Nadella, OpenAI's Sam Altman, Anthropic's Dario Amodei and Apple's Tim Cook.

[phkrause]

🦾 Quote du jour: AI is "underhyped"

Illustration: Aïda Amer/Axios

John Doerr — iconic Silicon Valley investor and chair of the venture capital firm Kleiner Perkins — has seen it all in tech over the past 50 years, going back to the PC revolution. The Wall Street Journal asked him to finish the sentence: "The latest AI revolution is the biggest thing since … " His answer: "Biggest thing ever. Since everything. It has been underhyped. We don't know how AI is going to shape the new world of education, employment, health care — life as we know it. ... My job as an investor is to help these entrepreneurs find, fund and accelerate their success." Read the interview (gift link).

[phkrause]

❌ AI still flubs, more confidently than ever

Illustration: Brendan Lynch/Axios AI tools might be hallucinating less. But they're still spitting out inaccurate answers cloaked in polished, hyper-confident language, Axios' Meg Morrone writes. Why it matters: The more people trust AI, the less likely they are to catch costly mistakes. It's a growing problem as people increasingly lean on the technology for research, medical advice and schoolwork. The big picture: Obvious hallucinations are easy to catch. The real trouble comes from false answers that sound convincing. Plausible citations, mostly correct summaries, and confidently wrong answers slip past users. If AI becomes accurate enough often enough, people might stop fact-checking altogether.

[phkrause]

🙏 AI stumbles on questions of faith

Data: Consortium for Evaluating Faith and Ethics in AI. Chart: Russell Contreras/Axios Artificial intelligence models are quietly shaping spiritual advice — often by leaving faith out, Axios' Russell Contreras writes from studies released last week by a new multi-university consortium. The studies — by researchers from Baylor, BYU, Notre Dame and Yeshiva — found Americans expected religion to appear in answers to moral and life questions 45%–59% of the time, depending on the topic. AI models mentioned religion only 5%–16% of the time. More data.

[phkrause]

💻 Exclusive: Office workers drive Codex growth

Illustration: Allie Carl/Axios

Knowledge workers now make up roughly one-fifth of Codex users and are growing more than three times as fast as developers, according to a new OpenAI report shared first with Axios' Megan Morrone. Codex, OpenAI's platform for coding and workplace tasks, now has more than 4 million weekly active users — up more than five times since OpenAI launched a desktop app version in February. The fastest-growing tasks among knowledge workers are data analysis (+110% week over week), research (+37%) and knowledge artifacts — reports, memos, docs, contracts, multimedia assets, PDFs and spreadsheets (+36%). Keep reading.

[phkrause]

Revenge of the AI bubble

Illustration: Brendan Lynch/Axios

The AI bubble debate has lurched through at least three frenzied phases in the span of three years, Axios' Zachary Basu writes. Suspicion: Historic sums of capital poured into AI before anyone proved it could reliably automate work. A violent market correction felt inevitable. Mania: Claude Code and autonomous agents made the early skepticism look outdated, fueling a corporate scramble to embed AI everywhere and maximize usage. Reckoning: Companies discovered that AI can be extraordinary when aimed precisely — and ruinously expensive when treated as a universal productivity machine. Why it matters: The first phase doubted the technology. The second phase worshipped it. The third phase — currently gaining steam across Corporate America — questions whether AI's immense power is worth the price. 🔎 Zoom in: The case against AI used to come from outsiders — Luddites, "doomers," short sellers betting on a crash. Its newest skeptics are emerging from inside the boom. Uber capped employee AI usage after burning through its annual Claude Code budget in four months. A top executive said the spending was getting "harder to justify," with no clear link between token use and more useful consumer features. Amazon shut down an internal token leaderboard after employees gamed it with throwaway tasks to climb the rankings. An Amazon executive told staff, "Please don't use AI just for the sake of using AI." GitHub moved Copilot, the AI coding assistant used by millions of developers, to usage-based billing as part of its effort to create a "sustainable" business. The change shocked users who were suddenly confronted with the true cost of heavy AI usage. Bain surveyed 951 large companies and found AI savings falling well below projections, even as most firms planned to spend more. "The technology worked. The value didn't arrive," the report concluded. The intrigue: Even OpenAI CEO Sam Altman has acknowledged the new concerns, calling the question of whether AI spending will show up in revenue "the most fair criticism" of the moment. 🎯 Reality check: The companies sounding the alarm are the early adopters. Most of the economy is still at the starting line, while the pioneers are the ones absorbing the cost shocks, wasted tokens and employee backlash. AI is already creating real value for chipmakers, model labs and some power users. The harder question is whether that value spreads across the companies paying to deploy it. By the numbers: Wall Street got a fresh reminder yesterday of how much AI optimism is baked into markets. The Nasdaq plummeted 4.2%, recording its worst day and worst week in more than a year. The Philadelphia Semiconductor Index plunged 10.3%, its worst day in more than six years. One culprit was Broadcom: The chipmaker reported explosive AI growth, but failed to raise its longer-term AI revenue outlook — disappointing investors looking for signs that demand was still accelerating. The bottom line: AI can make the right worker dramatically more productive, but those gains depend on knowing exactly where and how to apply it. The real bubble may have been the assumption that AI could be sprayed across companies, employees and workflows and reliably pay for itself.

[phkrause]

🤖 RSI 101: AI unleashing itself

Illustration: Lindsey Bailey/Axios

AI giants are beginning to discuss "recursive self-improvement" (RSI) — when AI builds itself. Brian, an AM reader in Alexandria, Va., asked us for an explainer. Voilà: AI could soon take control of its own development, eliminating the need for a human in the loop. Why it matters: The concern is that AI designing, training and improving itself — "recursive self-improvement" or RSI — will outpace the ability of humans to monitor, control or govern the technology, Axios' Maria Curi writes. 🛠️ How it works: Instead of the sci-fi scenario of AI writing its own code, a more likely scenario is the automation of AI R&D. This is already happening: AI labs are using current models to build future ones. Read Anthropic's RSI essay.

[phkrause]

AI's 4 harsh realities

Illustration: Lindsey Bailey/Axios

Investors were confronted this past week with four difficult realities that may fundamentally change the way they think about AI the business vs. AI the technology, Axios' Ben Berkowitz writes: 💰 AI is too expensive, say CEOs and even Microsoft itself. 🗑️ It's not paying off nearly as much as companies expected, per a new Bain study. ⛅️ Infrastructure demand is strong — but not as strong as the most optimistic wanted, as Broadcom showed with its "weak" forecast. 🏦 Financing that infrastructure is going to be more expensive for longer, with signs pointing to the Fed raising, not lowering, interest rates. Why it matters: Those realities challenge assumptions that powered markets to historic heights over the past few years. It's hard to justify chip or memory stocks rising 1,000%+ in a year if the boom isn't what everyone assumed. 🔭 The big picture: The costs of AI are now. The profits are later — maybe. That "maybe" is what's making people nervous. AI the technology has a bright future. But AI the business is starting to look like a bottomless pit — especially amid news that even some of the world's biggest companies are rushing to sell historic (and dilutive) amounts of stock to justify their expansion. By the numbers: The market sold off Friday amid those jitters, with the tech-laden Nasdaq having its worst day in 14 months. Broadcom's tepid outlook wiped $444 billion off its market cap alone in just two days. 🔎 Friction point: Tech selling off weighs down everything else. As charts expert Matt Cerminaro (a.k.a. "Chart Kid Matt") noted Friday, the S&P 500 was down more than 2%, even though the majority of stocks in the index were actually up on the day. The last time that happened? April 12, 2000, as the dot-com bubble was collapsing. The bottom line: Every great new technology has its moment where the business behind it resets, even as the tech itself keeps advancing. We could be seeing the start of that moment for AI.

[phkrause]

Confessions of an AI lab rat

Illustration: Lindsey Bailey/Axios

Axios CEO Jim VandeHei previewed his lab-rat learnings in his weekly newsletter for CEOs, Axios C-Suite, and wrote this broader version for Axios AM: I've spent the past year using AI obsessively — inputting copious amounts of personal and business data, turning myself into a lab rat for Axios and our readers. Why it matters: This experiment has shown me in unmistakable, hands-on ways the superhuman possibilities — and real-world limitations — hitting and awaiting us. In short: AI is way better, more accurate and mind-expanding than most think. (Sorry, it's true.) But it's colliding with hard human realities, making it confusing, clunky and chaotic for lots of people in its current form. How I did it: Over the past year, Axios aggressively tested AI (mainly OpenAI's Codex and Anthropic's Claude Code) across every layer of every department. We provided access and instruction to every employee. Most of my leadership team operates with chief-of-staff agents, and we're knee-deep in agent-to-agent prep. I personally use ChatGPT or Claude for one to two hours daily, usually in the early mornings, and control an AI personal operating system via my phone. That's connected to an always-on computer that runs several agents, including one that scans daily for CEO-relevant data and trends. I've dumped every medical record and blood test into it, and detailed my diet, workouts and supplements. It knows more about my health than my wife does! So here are my takeaways: It's way better than most think. I've spent the year with my head buried in this, while talking to the smartest people in tech, politics and business. AI is smarter than 95% of the people on 95% of topics, 95% of the time. Even for someone using it obsessively with real discipline, I'm still discovering it's way better than I thought possible. Its ability to think creatively and research deeply is extraordinary — if and only if you know how to use it. It takes real work. You can't wing it. You need to work at it daily, so AI learns you — and you learn AI. That's when the magic happens. You have to feed it copious amounts of information and persistently tell it what works and what sucks. This feedback loop creates a new form of super-knowledge about you — and super-skills for you. Most people get unimpressive results and move on, assuming it's overhyped. Don't. It's the smartest doctor I've met. I fed AI every medical record I have — MRIs, blood work, heart rate — and told it to be clinical and brutally honest. I've run most solutions past my doctor, and almost every time, he agrees. I still validate with physicians. But if I had to pick someone to diagnose something, I would turn to AI over human docs for anything complicated. Short-term job losses are overhyped. A year ago, I assumed AI's story would be subtraction: automate ruthlessly, cut costs, shrink headcount. That's real. We've done it at Axios. But over the last three months, my view shifted. The bigger opportunity isn't efficiency. It's new business lines that were economically impossible before AI. We're exploring three new revenue-generating projects that simply weren't possible without AI. I now believe many specific jobs here will change, but that we'll end up hiring more people over time than I would've thought a year ago. Business gains are overhyped, too — for now. As good as it is, AI hits internal walls when it comes to human use, security, connections to other systems and decisions about what data it can access inside companies. In most cases, it's simply not ready for deployment at scale. This problem is getting worse because agent-to-agent work is a mess. If AI transforms our business — and I think it will — agents need to work flawlessly with other agents. This is the unfolding frontier. My exec team has chief-of-staff agents, but we hit constant walls in determining what they can know, share and act on once the agents collaborate. This must be fixed before companies experience what I have at an individual level. A new class of super-worker is born. Here's the best news: We're spotting rank-and-file workers daily whose brains are wired for AI. It's been easier than expected to spot them, then train them to be AI accelerators on their team or across the company. These people are not technologists. You don't need to be an AI savant or lab rat. But every person reading this should figure out ASAP how AI can augment their work. If your company does not have AI teaching, demand it. It's affected my mind, mood and performance. I'm not a coder and rarely use AI for more than that hour or two per day, but these stories about people in Silicon Valley getting swept up in a manic AI fever — AI-pilled! — hit home for me. On the good side, I've jumped out of bed at 3 a.m. more than I care to admit, jazzed to test or explore a new idea. At 55, I've written and accomplished a lot more than any other time in my life. But you must train it to challenge and expand your thinking — not replace it. On the flip side, I find myself waking up after shorter bursts of sleep with more anxiety. Maybe it's coincidence, not causation. But I doubt it. The bottom line: We're living history. For $20 a month, any of us can experiment with exceptionally advanced AI models. Be clear-eyed about the good, bad and ugly. Most importantly, be curious. Use it daily. Read about it regularly. Figure out what parts of you can be vastly improved with AI — and then do it. Screenshot: Axios 📱 Watch a video of Mike quizzing Jim about his lab rat learnings (34 mins. ... Executive producer: Jimmy Shelton). ... Share this story. Tell us what you think: finishline@axios.com.

[phkrause]

AI money cannon

Data: S&P Global Market Intelligence. (Includes debt and equity raised by Alphabet, Amazon, Meta, Microsoft and Oracle.) Chart: Emily Peck/Axios Investors have poured an unprecedented $255 billion into five AI hyperscalers — Alphabet, Amazon, Meta, Microsoft and Oracle — already this year, more than twice what those companies raised in all of 2025, Axios' Emily Peck and Dan Primack write. Why it matters: Investors are more exposed to AI's promises — and its risks — than ever. With savings dwindling and wages lagging inflation, portfolio gains are about the only place investors are making real money. 🔮 There's a lot more money on the way. SpaceX goes public Friday in the biggest U.S. IPO ever, by a mile, raising at least $85 billion, with demand far outstripping the shares available. Alphabet just sold a record amount of new stock. Both could be dwarfed by OpenAI and Anthropic, which are expected to go public later this year. 🧮 By the numbers: The five companies have said that by year-end, they'll have spent three-quarters of a trillion dollars on AI data centers, per Barron's. The bull and bear case for SpaceX: It's a foregone conclusion that SpaceX will raise at least $85 billion. The real question is what happens next. 🐂 Bull: SpaceX could generate hundreds of billions of dollars in revenue by 2030, despite booking less than $19 billion last year. The biggest chunk would come from Starlink, which could use Starship's larger payloads to supercharge its young satellite phone service and corner the global market. SpaceX's AI business has a high floor: It keeps selling computing power. Deals like those with Anthropic and Google are already worth about $2 billion a month combined. 🐻 Bear: For most of its life, SpaceX has been the runaway leader in a business it essentially invented: commercial rocket launches. Its newer ambitions in AI and telecom are far more crowded, and the stock is very expensive to start. Its giant Starship rocket is still a work in progress. Computing power — one of SpaceX's cash cows — is getting cheaper as more and more data centers come online. 🃏 The wild card: Elon Musk. His market magic is real. Bet against him at your portfolio's peril. If Musk were no longer leading SpaceX, investor enthusiasm would dissipate.

[phkrause]

🤖 The government should be able to block or deter dangerous AI deployments, Anthropic CEO Dario Amodei writes. Go deeper ... Read his essay.

🏢 AI firms are fueling red-hot demand for Manhattan office space, The Wall Street Journal reports. Gift link.

[phkrause]

AI warning

For years, the biggest fear about artificial intelligence was that it might take your job. Now, US officials and their intelligence partners have a more urgent concern: that some AI models are just months away from being capable of launching cyberattacks powerful enough to overwhelm governments and major companies. A group of international spy agencies issued a joint statement on Monday urging leaders to "act now" to strengthen their defenses, days after the Trump administration ordered AI giant Anthropic to suspend foreign nationals' access to its most advanced models. Read more.