CoreWeave just paid $9 billion to buy Core Scientific, eliminating $10 billion in future lease payments. The AI company decided it was cheaper to buy its landlord than keep paying rent. Markets weren't impressed with the math.
Good morning from foggy San Francisco, Teamwork, but for silicon brains. Tokyo startup Sakana AI let ChatGPT, Claude, and Google&
Japanese researchers prove AI models work better as teams than alone, boosting performance 30%. TreeQuest system lets companies mix different AI providers instead of relying on one, potentially cutting costs while improving results.
Good morning from foggy San Francisco,
Teamwork, but for silicon brains. Tokyo startup Sakana AI let ChatGPT, Claude, and Google's Gemini tackle the brutal ARC-AGI-2 test as a relay team. Solo models solved <5% of problems; TreeQuest's tag-team cracked 30%. Translation: shunt the easy stuff to bargain models, reserve premium tokens for the hard bits—good-bye vendor lock-in.
Musk files for political separation. After Trump signed a $3.3 trillion spending spree Musk called "insane," the billionaire yanked his $290 million stake and launched the America Party on X. Tesla slid from $488 to $315 per share overnight. Now he's aiming a $300 billion war chest and 200 million followers at 2026 swing seats. Third parties usually fizzle—none have started this loaded.
Stay curious,
Marcus Schuler
Japanese researchers just proved what office workers know: teams beat solo acts. Sakana AI's new study shows AI models working together outperform individual systems by 30%.
Their TreeQuest framework lets companies mix different AI providers instead of betting everything on one. Think ChatGPT for reasoning, Claude for safety checks, and Google's models for image tasks. Each handles what it does best.
The test results tell the story. On ARC-AGI-2, one of AI's toughest benchmarks, most models solve under 5% of problems. TreeQuest's collaborative approach cracked 30%. One model would start a solution, another would debug it, and a third would polish the result.
The business case writes itself. Companies can route simple tasks to cheap models and complex work to premium ones. No more paying GPT-4 rates for basic questions. No more vendor lock-in nightmares.
Sakana released the whole system as open-source software. Any company can download it, modify it, and use it commercially without fees. The code sits on GitHub with documentation included.
Why this matters:
• AI's future lies in coordination, not just bigger models—think jazz ensemble versus solo artist
• Vendor diversification becomes practical, letting companies escape single-provider dependence while cutting costs
Read on, my dear:
Maria Garcia
Prompt:
Closeup portrait, pastel pink and yellow marble helmet, robotic, brunette girl, baroque, upper body, cinematic, castle background
Elon Musk announced the America Party on Saturday after his $290 million friendship with Trump exploded over spending. The world's richest man had warned he'd start a third party if Congress passed what he called an "insane spending bill." Trump signed it Friday anyway.
The breakup was swift and ugly. Trump called Musk's party idea "ridiculous" and accused him of going "off the rails" on Truth Social. Musk fired back that Republicans had "the nerve to massively increase government size" despite controlling everything.
Tesla stockholders felt the pain. Shares crashed from $488 in December to $315 last week as their bromance soured. Musk went from dancing at Trump rallies to launching a rival political movement in five months.
The America Party targets 2-3 Senate seats and 8-10 House districts in 2026. Musk wants to create a swing bloc rather than attempt a full national campaign. Early polling shows 40% of voters would consider his party.
Musk faces the same math that crushed Ross Perot and every third-party effort since. But he has resources Perot never dreamed of: $300 billion and 200 million X followers.
Why this matters:
• Musk proves that $290 million can't buy lasting political loyalty when egos clash—money talks but spite screams louder
• A billionaire with unlimited cash is finally testing whether third parties fail due to structural barriers or insufficient funding
Read on, my dear:
Marcus Schuler
I'm looking for a practical mindfulness guide that fits into a busy lifestyle. I'm a [your situation - e.g., working parent, student, etc.] with about [time available] minutes per day to dedicate to mindfulness practice. I'm particularly interested in:
Please provide specific examples and a realistic routine I can start with this week. I'd prefer [format preference - e.g., step-by-step instructions, a weekly schedule, etc.]
Windows 11 has overtaken Windows 10 to become the most popular desktop operating system, claiming 50.88% market share compared to Windows 10's 46.2%. The milestone comes just months before Windows 10 loses support in October, finally giving Microsoft's controversial OS the momentum it needed after a sluggish four-year climb.
Huawei's secretive AI research lab broke its usual silence to defend against GitHub accusations that it lifted uncredited code from rivals for its Pangu AI model, insisting it followed proper open-source licensing rules. The rare public rebuttal comes as pressure mounts on Chinese tech firms to prove their AI breakthroughs aren't just well-executed homework assignments.
DragonForce and RansomHub, the ransomware groups behind recent attacks on M&S and other major retailers, have started feuding with each other in a turf war that began when DragonForce took down its rival's website and declared itself a "cartel." The criminal spat could leave companies facing double extortion as both groups target the same victims to prove who runs the underworld's most profitable protection racket.
At least 36 startups have achieved unicorn status in the first half of 2025, driven largely by an AI investment frenzy that has investors throwing billion-dollar valuations around like confetti. While most new unicorns focus on artificial intelligence, the list includes surprises like probiotic soda maker Olipop and prediction market Kalshi, proving that even in an AI-obsessed market, there's still room for fizzy drinks and gambling on election outcomes.
Apple is appealing the European Union's €500 million fine for blocking app developers from steering users to cheaper purchase options outside the App Store, calling the penalty "unprecedented" and claiming regulators are forcing bad business terms. The company already made the required changes in June to avoid daily fines while fighting the decision, proving sometimes it's easier to comply first and complain later.
Capgemini agreed to buy IT outsourcing company WNS for $3.3 billion to expand its "agentic AI" capabilities, which is tech-speak for artificial intelligence that can make decisions and complete tasks without human hand-holding. The French consulting firm is betting that clients will pay premium rates for AI systems that can essentially manage themselves, though the irony of buying a human-heavy outsourcing company to build human-free AI seems lost on everyone involved.
JPMorgan CEO Jamie Dimon says remote work fails to train junior bankers because the traditional apprenticeship model of learning by watching senior colleagues "doesn't work" over Zoom calls. The problem gets worse as AI eliminates the routine grunt work that used to teach foundational skills, leaving firms scrambling to figure out how to train the next generation without cubicle-side mentoring or mind-numbing spreadsheet tasks.
Google spent a year making Moto X smartphones in Texas starting in 2013, complete with patriotic marketing and customizable wooden phone backs, before selling the whole operation to Lenovo and admitting defeat. The failure had nothing to do with American manufacturing costs or capabilities—the phone just wasn't good enough to compete with the iPhone, proving that even the best assembly line can't fix a mediocre product.
Hangzhou has become China's answer to Silicon Valley, spawning AI giants like DeepSeek and Alibaba while attracting young entrepreneurs who call themselves "villagers" and code in coffee shops by day before gaming together at night. The city's combination of government subsidies, low rents, and a backyard startup scene has created six major AI companies that Chinese media dubs the "six tigers of Hangzhou," proving that tech ecosystems can flourish even when the government is picking up the tab.
DeepSeek's ultra-low pricing of $0.55 per million input tokens initially shook markets, but users are now fleeing the company's own service for third-party providers that offer the same models with better performance. The Chinese AI company achieves rock-bottom prices by forcing users to wait many seconds for responses and limiting context windows, prioritizing internal research over customer experience.
Oracle is offering the federal government a 75% discount on its database software and substantial savings on cloud services through November, making it the latest tech company to cut prices for the Trump administration. The deal follows similar discounts from Salesforce, Google, and Adobe as the government pushes to modernize its notoriously outdated IT systems while extracting savings from tech contractors.
US industrial companies like Honeywell and Generac are abandoning their struggling traditional markets to chase the AI boom, pivoting to sell cooling systems and backup generators to data centers instead of factories and homes. While manufacturing contracts and home generator sales tank, these companies discovered that tech giants will pay premium prices for the same pumps and power equipment when it keeps ChatGPT running.
How to Write Product Descriptions Fast
Copysmith creates product descriptions, blog posts, and marketing copy in minutes. It offers templates for different types of content and can write in your brand's voice. The tool also checks for plagiarism to ensure your content is original.
Tutorial:
Three Google defectors bet Tokyo could spawn the next OpenAI. They were right – sort of. 🐠
The Founders
David Ha and Llion Jones (ex-Google Brain legends) plus diplomat-turned-exec Ren Ito launched Sakana AI in July 2023. Just 30 employees. Based in Tokyo because they wanted to prove world-class AI doesn't require Silicon Valley zip codes.
The Product
Forget massive models. Sakana builds swarms of small AI agents that work together like fish schools. Their flagship "AI Scientist" literally writes research papers and runs experiments autonomously. Papers have passed peer review at top conferences. They use evolutionary breeding to merge models – Darwin meets deep learning.
The Competition
Battling OpenAI and Google globally while outpacing Preferred Networks domestically. Every GPT breakthrough raises the bar for their distributed approach. Swimming upstream against industry obsession with bigger-is-better models.
Financing
$30M seed (January 2024) exploded into $210M Series A (September 2024) at $1.5B valuation. Japan Inc. went all-in: three megabanks, Sony, NTT, Fujitsu, plus Silicon Valley's Lux Capital and Nvidia's venture arm. Fastest Japanese unicorn ever.
The Future ⭐⭐⭐⭐
MUFG just handed them $34M to automate banking documents. Japan desperately needs an AI champion, and Sakana fits perfectly. But they must prove collective intelligence beats brute-force scale before the hype cycle turns.
Smart money says these fish will keep swimming upstream. 🚀
Good Morning from San Francisco, Microsoft axed 9,000 workers Wednesday. Record profits didn't save them. The company
Good Morning from San Francisco, Meta wrote checks for $300 million. OpenAI researchers said no. Zuckerberg offered packages that make
Good Morning from San Francisco, Cloudflare just declared war on freeloading AI bots. The company that handles 20% of internet
Good Morning from San Francisco, Meta just poached four more OpenAI researchers. Mark Zuckerberg's $100 million talent raid
Get the 5-minute Silicon Valley AI briefing, every weekday morning — free.
