░▒▓██▓▒░ ░▒▓████▓▒░ ░▒▓██▓▒░
T H E U N C E R T A I N F U T U R E S
all the news that fits the context window
▓█▓▒░▒▓█▓▒░▒▓█▓▒░▒▓█▓▒░▒▓█▓▒░▒▓█▓▒░▒▓█
No. 27 · Monday, 20 July 2026Feliz lunes. Spain lifted the World Cup in New Jersey last night; enhorabuena, España, and don't cry for me, Argentina. On this desk's beat, Meta may lease its spare computing power to Anthropic, a company it competes with. Investors spent last week asking whether the AI buildout will actually pay for itself, and for a stretch on Friday the doubt made Apple the world's most valuable company again, before Nvidia edged back in front by the bell. The backlash against data centers went national in the United States, with coordinated protests across 42 states. Let's get into it.
░░▒▒▓▓ TOP 5 ▓▓▒▒░░Meta in talks to rent AI computing to a rival, Anthropic, in a deal worth up to $10 billion
Meta is in talks to lease spare data-center capacity to Anthropic for as much as $10 billion over two years, an arrangement Anthropic proposed in June as it scrambles for scarce compute. It would turn Meta's enormous AI buildout into a cloud business to rival CoreWeave and, eventually, AWS, whose longtime computing boss, Dave Brown, Meta is now hiring to run it. The oddity: Meta trains its own Llama models and competes with Claude head-on, yet would become its rival's landlord. Reported by The Next Web and Yahoo Finance (The Next Web · Yahoo Finance).
Apple briefly overtakes Nvidia as the AI trade wobbles, before Nvidia reclaims the crown at the close
Apple briefly overtook Nvidia in intraday trading on Friday, touching about $4.88 trillion to be the world's most valuable company for the first time in more than a year, before slipping back and letting Nvidia hold the title into the close. The near-miss says less about Apple than about nerves over AI: chip and infrastructure stocks slid through the week on doubts about whether the spending will pay off, a jitter that China's cheap new Kimi K3 model only sharpened (more on Kimi in AI, below). Apple, long mocked as the laggard that never built its own frontier model, is suddenly the safe place to hide. Covered by 9to5Mac and Forbes (9to5Mac · Forbes).
The revolt against data centers goes national in the US, with 142 protests across 42 states
Opponents of the AI data-center boom staged 142 demonstrations across 42 US states on Saturday, the first coordinated national protest in the United States against a buildout that has lifted local power bills and roiled town politics. The effort was organized by a new group, HumansFirst, co-founded by a former Tea Party organizer, and lands with opinion already turning: a June Reuters/Ipsos poll found only 14% of Americans would welcome a data center in their own community. The physical cost of AI has become a live political liability. Reported by Reuters via U.S. News (U.S. News).
Netflix says generative AI touched about 300 titles as its growth cools and the stock slides
In its Thursday earnings letter, Netflix said roughly 300 of its titles used generative AI this year, mostly in post-production, with co-chief executive Ted Sarandos pointing to 17 minutes of "AI-enhanced footage" in one show made at half the usual cost. For scale: What's on Netflix, an independent tracker, counted 597 new original titles released in all of 2025; this year's AI count reached 300 with five and a half months still to run. Investors fixed on the outlook: revenue growth slowed to about 13%, roughly in line with expectations, and soft third-quarter guidance sent the shares down around 9% after hours to a 52-week low. AI has arrived in Hollywood less as a synthetic blockbuster than as a line in the budget spreadsheet. Covered by Variety and The Verge (Variety · The Verge · What's on Netflix).
Databricks lines up a $3 billion round at a $188 billion valuation
Databricks has signed a term sheet for a strategic round led by existing investor Coatue that values the data-and-AI software company at $188 billion, about 40% above the $134 billion it fetched in February. The raise, roughly $3 billion, extends an extraordinary run for a firm that reinvented itself from data plumbing into AI infrastructure and is widely expected to go public. In a week when the public markets fretted over AI's payoff, private investors kept writing bigger checks. Covered by TechCrunch and Databricks (TechCrunch · Databricks).
░░▒▒▓▓ STATS OF THE DAY ▓▓▒▒░░░░▒▒▓▓ AI ▓▓▒▒░░Kimi K3 is so popular that Moonshot has stopped selling it. Moonshot AI, the Beijing lab whose Kimi K3 launch we covered Friday, suspended new consumer subscriptions on Sunday, three days after release: "Over the past 48 hours, demand has pushed close to the limits of our current capacity," the company said, promising to reopen spots "in batches" while existing subscribers keep full access. The squeeze is one Chinese labs cannot simply spend their way out of, since US export controls cap the GPUs available to serve a hit product. For the moment, the cheapest frontier-class model on the market is the one you cannot subscribe to (SCMP · TechNode).
Google DeepMind's own safety researcher quits over its Pentagon deal. Alex Turner, an AI-safety researcher at Google DeepMind, resigned in June and said in a July 15 public post that he had failed to stop Google from signing a contract that lets the US military deploy its models for any lawful purpose, with no restriction against autonomous weapons or mass surveillance. Roughly 600 employees had signed an internal protest; Turner noted that Google's 2018 pledge never to build such systems is not binding. A pledge-signer, he wrote, has three honest options: "explain publicly how staying is consistent with the pledge, say plainly that you no longer hold it and why, or quit." As he put it: "Wearing the pledge while saying nothing isn't one of them." (Business Insider · Storyboard18).
Microsoft preps a cheaper answer to Anthropic's Mythos. The company is readying Project Perception, a security product that routes between Anthropic, OpenAI and Microsoft's own models to find software bugs and patch them automatically, pitched as a lower-cost alternative to Anthropic's expensive Mythos, The Information reported. It is the second time in a week Microsoft has been caught building against a partner whose models it also resells (The Information).
░░▒▒▓▓ TECH ▓▓▒▒░░Apple and the Justice Department open talks to settle the iPhone antitrust case. The two sides are in early discussions to end the 2024 suit, brought by the DOJ and 15 states, that accuses Apple of monopolizing the smartphone market through its grip on messaging, wallets, watches and cloud gaming. Apple has floated several settlement offers this year, plainly keen to buy its way out of a case that has yet to earn a trial date (9to5Mac).
San Francisco tells Apple and Google to pull 13 "nudify" apps. City Attorney David Chiu sent cease-and-desist letters ordering both companies to remove 13 face-swap apps used overwhelmingly to make non-consensual sexual deepfakes of women and girls, accusing the stores of long profiting from tools they knew violated state law (WIRED · TechCrunch).
░░▒▒▓▓ POLICY ▓▓▒▒░░Xi headlines China's AI conference in person, selling open source. President Xi Jinping made his first personal appearance at Shanghai's World AI Conference on Friday, urging countries to embrace open-source AI and the China-based cooperation body that 29 nations signed onto last week, while warning against letting national security override shared control. AI, he said, should not be "a solo performance by a single country." (SCMP).
A bipartisan bill would keep a human in the loop before AI kills. Representatives Don Beyer, Tom Barrett and Sara Jacobs introduced the Human Authority over Autonomous Weapons Act, which would write into law that any intentionally lethal use of an AI-enabled weapon must carry human oversight, and, for five years, would force commanders to confirm AI-generated targets against a non-AI source. "No machine should ever be given the power to decide to kill a human being on its own," Beyer said (Rep. Beyer).
░░▒▒▓▓ INFRA ▓▓▒▒░░SpaceX is in talks to sell the Pentagon billions in AI computing. Fresh off the largest IPO in history, the Musk company is negotiating to provide the Defense Department with billions of dollars of AI data-center capacity, deepening its move from rockets into government AI infrastructure and dropping it into the same business Meta and Anthropic are now circling (The Wall Street Journal).
Huawei shows off a rack-scale answer to Nvidia. At the Shanghai AI conference Huawei displayed a prototype of the Atlas 950 SuperPoD, a system that lashes together up to 8,192 of its Ascend accelerator chips and which it claims delivers 6.7 times the compute of Nvidia's next-generation rack, its clearest bid yet to beat export controls with scale rather than per-chip speed (Tom's Hardware · Huawei Central).
░░▒▒▓▓ BUSINESS ▓▓▒▒░░Fireworks raises $1.5 billion at a $17.5 billion valuation. The startup that helps companies run and tune open-weight AI models on their own data raised a $1.5 billion round co-led by Atreides Management, Index Ventures and TCV, after crossing $1 billion in annualized revenue, a pointed bet that the future is enterprises stitching together cheaper open models rather than renting frontier ones (CNBC).
░░▒▒▓▓ GEOPOLITICS ▓▓▒▒░░The war between the United States and Iran escalated over the weekend (update): two American service members were killed and one went missing after an Iranian missile-and-drone attack on a base in Jordan on Friday, the first US combat deaths from Iranian fire since March, and CENTCOM said US forces then struck Iran for an eighth consecutive night. The war's American death toll rose to seventeen over the weekend, the US military said (Al Jazeera · The Washington Times).
Russia fired more than 40 missiles and about 125 drones at Kyiv overnight into Sunday, the largest ballistic-missile barrage on the capital since the 2022 invasion, killing at least one person and wounding 16 as fires broke out across several districts (Al Jazeera).
░░▒▒▓▓ ALPS ▓▓▒▒░░Munich's Microagi raises $55 million in what it calls Germany's largest-ever seed round. The 10-month-old startup, founded by former Formula 1 engineer Bercan Kilic, records factory workers with cameras and sensor gloves and uses the footage to fine-tune existing robots for a specific plant's tasks, closing the gap between a slick demo and reliable work on the line. The superlative is the company's own, though no bigger German seed round is on record: the largest of 2025 was audio-AI startup Mirelo's $41 million, per Tech.eu's league table, and seed checks this size have so far been a French specialty (Mistral's 2023 seed was $113 million). The round, led by Hummingbird with Swiss fund Redalpine among the backers, will also fund a new research headquarters in Zurich (Sifted · SiliconANGLE · Tech.eu).
Germany's leading tech investor says Europe cannot buy its way to AI independence. In an NZZ interview, venture capitalist Judith Dada argued Europe is doubly dependent on the United States, for its AI models and for the computing power to run them, holding roughly 5% of global compute against America's three-quarters. Even a homegrown Claude, she warned, would have nowhere to run without a vast data-center build (NZZ).
░░▒▒▓▓ RESEARCH ▓▓▒▒░░The FDA clears the first cholesterol pill in the class that once meant injections. Merck won approval on July 16 for Lipfendra, the first oral PCSK9 inhibitor, a once-daily tablet that lowered LDL cholesterol by 56% against placebo in one trial and 59% in patients with an inherited form. Delivered as a "macrocyclic peptide" that survives the gut, it brings antibody-grade potency to a roughly $315-a-month pill instead of a needle (CNN · Merck).
An implant plus AI gave a paralyzed man back movement and touch. Researchers reported in Nature Medicine that a "double neural bypass," pairing a brain-computer interface with AI and electrical stimulation of the spine, restored hand movement and a sense of touch to a man paralyzed from the chest down, with gains still present more than two years on (The Next Web).
░░▒▒▓▓ ROBOTICS ▓▓▒▒░░A robotaxi drove into a fire, and now 105 are being recalled. Amazon's Zoox is recalling software in 105 driverless vehicles after one, empty, failed to recognize heavy smoke and rolled into an active emergency scene in Las Vegas last month; the over-the-air fix teaches the cars to detect and back away from smoke. US regulators say they have found a pattern of driverless cars getting in the way of first responders (TechCrunch · CNBC).
A London humanoid maker becomes a unicorn on the strength of German factories. Humanoid, legally SKL Robotics and founded two years ago by Russian entrepreneur Artem Sokolov, raised a $150 million first tranche that values it at $1.2 billion, The Information reported. Its order book leans on European industry: a deal to supply parts-picking robots to Schaeffler, the German industrial supplier, with Bosch set to help build them and Siemens running trials (The Information).
░░▒▒▓▓ REPORTS ▓▓▒▒░░Stanford HAI — "The Commercial Landscape of AI Sovereignty Offerings." Surveying the flagship sovereignty packages Nvidia, Microsoft, Google, AWS and OpenAI now sell to governments, the 20-page brief concludes that absolute AI sovereignty is a practical impossibility today, because every "sovereign" stack still runs on a small set of mostly American chips, clouds and models. The business built on that impossibility is thriving: sovereign AI reportedly accounts for roughly $30 billion, or 14%, of Nvidia's revenue and is projected to reach $200 billion in the coming years, with Nvidia-backed national "AI factories" in 25 countries, from Ecuador to Armenia to Vietnam, and 18 "OpenAI for Countries" initiatives under way. The catch, in the authors' words: the packages "promise that countries will own their AI stack, but at the same time they deepen dependencies on U.S. Big Tech." Even the partner-run European clouds sold as legal shields against Washington do not quite close the door, an exposure Microsoft and AWS have themselves all but conceded, and the brief passes along the label critics use for the sales pitch: "sovereignty washing." Its advice to policymakers is to stop chasing self-sufficiency and treat sovereignty as the capacity to shape and negotiate dependencies rather than escape them, with open-weight models flagged as one meaningful way to cut dependency at the model layer. Disclosure, dutifully noted on page 20: Stanford HAI's funders include Nvidia, Google, Microsoft and AWS (Stanford HAI).
░░▒▒▓▓ COMMS DESK ▓▓▒▒░░As public anger at AI curdles, the industry's executives are spending to protect themselves: Anthropic is running round-the-clock security, and Palantir, Oracle and Salesforce have sharply raised executive-protection budgets, with some leaders now traveling with armed guards and firms quietly telling staff to skip the company logo, The Wall Street Journal reports. Takeaway: reputation and physical-safety risk are converging, and how visible a company's leadership should be is becoming a communications question as much as a security one (The Wall Street Journal).
░░▒▒▓▓ ONE MORE THING ▓▓▒▒░░For years, prediction markets have let ordinary people bet on what President Trump might say. One person had an unusual edge: Gabriel Perez, his teleprompter operator since 2016. Perez allegedly won more than $100,000 on Kalshi wagering on which words and phrases Trump would use in big speeches, the very words he was loading onto the screen, until the platform's surveillance flagged the trades and referred him to federal regulators. He is now on unpaid leave with his winnings frozen, which is one way to learn that the house always watches (NBC News).
░░▒▒▓▓ TRACK OF THE DAY ▓▓▒▒░░Abpfiff — the Uncertain Futures crew