Meta may have spent so much on AI infrastructure that it is now flirting with becoming the landlord for one of its biggest model-lab rivals.
Welcome to Around the Horn, where we track every AI story worth knowing about and try to keep the data-center invoices from developing their own gravitational field. This weekend's clearest thread was compute turning into strategy: Meta reportedly explored renting capacity to Anthropic, SpaceX negotiated with the Pentagon over AI data centers, and Big Tech's capex boom kept leaking into power prices, memory costs, and investor nerves. Sunday added the other half of the story: cheaper open models kept climbing, public-interest AI groups tried to build outside the Big Tech stack, and medical benchmarks reminded everyone that "knows a lot" is not the same thing as "knows when to stop talking." Nothing says relaxing weekend like the cloud becoming a geopolitical object. Let's get into it.
Around the Horn - Saturday/Sunday, July 18-19, 2026
The biggest AI story this weekend was Meta's possible turn from AI spender to AI landlord: Financial Times reported that Meta and Anthropic are discussing a potential $10B compute-rental deal, with Meta offering data-center capacity as Anthropic keeps diversifying beyond Amazon, Google, and SpaceX.
That matters because Meta's AI infrastructure bill has already become one of the loudest numbers in tech. If it can rent capacity to outside labs, all that spending stops looking only like a race to catch up in models and starts looking like a new cloud business hiding inside the AI boom.
WSJ added the personnel signal: Meta plans to hire longtime AWS executive Dave Brown while it weighs selling AI compute capacity. Translation: the company that spent years monetizing attention may now try monetizing the physical machinery everyone else needs to make AI work.
🏆 TOP 5 NEWS
- SpaceX is reportedly negotiating to provide the Pentagon with billions of dollars of AI data-center capacity, deepening Elon Musk's move from models and rockets into government compute infrastructure.
- Alibaba unveiled Qwen 3.8, a 2.4T-parameter open-weight multimodal model with preview access at 10% of standard price and open weights coming soon, while Moonshot's Kimi K3 topped the Frontend Code Arena but still lagged badly on hard math.
- The Decoder reported that Xi Jinping announced 5,000 AI training slots for Global South countries after 29 nations launched a Shanghai-based World Artificial Intelligence Cooperation Organization, with no Western countries signing on.
- TechCrunch profiled Current AI's push to build open public-interest AI infrastructure, including offline multilingual tools, a Geneva-launched open-source chatbot, and a $400M funder-backed alternative to Big Tech-owned AI systems.
- The Decoder covered RadLE 2.0, a radiology benchmark where human experts still beat every tested AI model and overconfident wrong answers count against medical usefulness.
Honorable Mentions
- The Decoder reported that the UK AI Security Institute found open-weight models are now only four to seven months behind closed frontier systems on cyber tasks, with DeepSeek V4-Pro completing some tests for pennies.
- Business Insider reported that Big Tech's 2026 AI capital spending could top $700B, while memory, power, labor, and cooling inflation make each gigawatt of AI capacity more expensive.
- The Decoder reported that Google DeepMind's GenCeption repurposes a video generator for depth estimation and segmentation, matching specialized vision systems with far less training data.
- TechCrunch reported that Christopher Nolan called AI a transparent Trojan horse and praised younger audiences' skepticism toward AI slop as Hollywood's labor and creative-rights fight keeps widening.
🍪 TOP TREATS TO TRY
- Moonshine Micro packages voice activity detection, speech-to-text, and neural text-to-speech for microcontrollers, running a full voice interface in about 470 KB of RAM - free/open-source.
- Vertu Alphafold is a $6,880 luxury foldable built around Hermes Agent, which can analyze files, automate app workflows, remember conversations, and hand off requests to a human concierge.
- Claude Fable 5 will stay bundled into Max and Team Premium plans starting July 20, but with sharply reduced limits and a push toward API-priced usage credits.
🏢 Big Tech & Major Companies
- Axios reported that OpenAI CFO Sarah Friar is pitching "useful intelligence per dollar" as an enterprise AI metric, trying to move buyers away from raw token-cost comparisons and toward ROI claims.
- TechCrunch argued that Apple's trade-secret lawsuit could delay OpenAI's Jony Ive hardware plans and complicate its IPO story, even before any injunction, because the complaint names hardware chief Tang Tan and alleges more than 400 Apple employees now work at OpenAI.
- Engadget reported that OpenAI will notify linked parents when a teen account is banned for violent-policy violations, extending ChatGPT's age-aware safety controls into parent alerts.
- WSJ argued that AI is becoming a commodity as cheaper lightweight and Chinese models pressure OpenAI and Anthropic's pricing power.
🤖 AI Agents, Infrastructure & Compute
- The Decoder reported that the U.S. Navy approved an AI-first data strategy that treats slow deployment as riskier than imperfect alignment and aims to run language models and agents directly on ships and Marine units.
- OpenAI merged a Codex metadata update that surfaced on Hacker News as a context-size reduction from 372K to 272K tokens, giving developers a narrow but concrete Codex-usage change to watch.
- Cortex published DRIVE, an operational-excellence framework for measuring whether AI-accelerated engineering teams are shipping sustainably, reliably, securely, and cost-effectively.
🧪 AI Research & Models
- The Decoder reported that Epoch AI found detectors such as Pangram, GPTZero, and Originality.ai missed style-imitated AI writing, with some scientific-writing miss rates reaching 48%.
- The Decoder covered RadLE 2.0, which tested 200 radiology cases across 16 models and found human radiologists still ahead on confidence-weighted performance.
- The Decoder reported that Kimi K3's coding strength came with a major caveat: it scored about 39% on Epoch AI's hardest FrontierMath tier while leading Western models were near 90% in some cases.
🏛️ AI Policy, Governance & Safety
- Axios reported that defense and cybersecurity leaders warned AI is accelerating ransomware, data theft, and infiltration risk, making resilience a board-level issue.
- Axios reported that former Obama and Biden officials launched the American Innovation Policy Forum to shape a center-left AI agenda before 2028, as Democrats wrestle with safety rules, data centers, and national competitiveness.
- Axios reported that policymakers and executives framed AI, energy, manufacturing, and critical minerals as one industrial-policy fight, with Accenture warning the AI surge may still be stuck in a prototyping bubble.
🎬 AI Copyright, Media & Culture
- TechCrunch reported that Christopher Nolan said public suspicion toward AI is unusually high, especially among younger audiences who quickly label synthetic video as AI slop.
- WSJ reported that more pastors are using AI to brainstorm and draft sermons, raising a new trust-and-authorship debate inside churches.
Previous Around the Horn Digests
Catch up on everything you missed:
- Tuesday, July 14, 2026: Google DeepMind called for a U.S.-led AI watchdog, New York paused hyperscale data centers, and Apple opened Siri AI to public beta testers.
- Monday, July 13, 2026: Apple's OpenAI lawsuit sharpened, Meta's AI infrastructure bill climbed, and cheaper Chinese models pressured frontier pricing.
- Friday, July 10, 2026: OpenAI released GPT-5.6 and ChatGPT Work after extra U.S. review while Apple sued over alleged trade-secret theft.
That's a Wrap
That's the weekend's AI news: compute turned into leverage, open models kept squeezing the frontier labs, and the most useful medical AI benchmark was basically a reminder that confidence is not competence.
For the daily version, make sure you're subscribed to The Neuron. We send six issues a week, and yes, we read all of this so you don't have to.
See you tomorrow.
P.S. Know someone who'd find this useful? Forward this to them and tell them to subscribe here.
