Stratechery

Ben Thompson and Andrew Sharp discuss Ben's recent article on the transformative value of LLMs that write things down, helping users uncover blindspots. They note the availability of omnicompetent assistants to everyone, and examine the arrival of OpenAl's Astra models, which point toward a future where models use computers. The conversation also covers the rapid pace of AI progress in fall 2026, Meta's Muse as a tool to unlock agent potential for normal people, and a pair of viral p(doom) tweets from Anthropic employees that revive fears about out-of-control AI, with a pandemic-era callback offering an instructive parallel. Finally, an emotional announcement about the iPhone Air rounds out the episode.

Ben and Andrew discuss the release of Fable 5.1 and Anthropic's walkback of its data retention safety initiative. They react to the viral OpenAI-Hugging Face incident and the broader problem of anthropomorphizing AI discourse. Follow-ups cover the tension between libertarian and conservative impulses in Meta's content moderation, why Shopify is a platform and not an aggregator, and a question about 3-D printing. The episode also includes a home networking anecdote, an F1 app PSA, and a listener's girlfriend who hates vibe coding.

Ben and Andrew react to Meta's teen restrictions settlement, Nvidia's reported Hugging Face acquisition and what it says about open source, Netflix's new willingness to funnel users to rivals, ESPN's ticker removal, Druckenmiller's AI-written op-ed, and a grab bag of topics including the mysteries of model performance and the beauty of hurling.

Andrew and Ben react to Apple's latest App Store news, arguing that even significant platform changes feel diminished by the scale of the AI transformation. They discuss OpenAI's pause of internet model training and the misaligned incentives that make cybersecurity so challenging, especially as AI compounds the threat. The conversation covers how black hat and white hat actors are converging on similar AI tools, and what bubble-era benefits might survive the current AI investment frenzy. A Q&A segment addresses career concerns about AI replacing financial planners and the evolving role of AI in the writing process.

Ben and Andrew discuss Nvidia's new funding model for AI infrastructure, including how it compares and contrasts with railroad expansion 150 years ago and why LLMs were both a gift and a curse to Nvidia's business. They examine the growing pressure from Google and Amazon, Google's attrition as a possible positive sign for its frontier efforts, and the future of AI-generated output and Anthropic's watermarking plan. The episode ends with listener questions about Vision Pro, Starlink Mobile, and Ben's enthusiasm for his vibe coded app.

Ben Thompson and Andrew Sharp analyze the latest earnings from Microsoft, Meta, and Google, focusing on Microsoft’s enterprise AI middle-layer strategy and its vulnerability as models commoditize. They contrast Google’s hedging of frontier bets with Meta’s all-in spending and new enterprise ambitions, then examine the lending environment for hyperscalers and whether AI spending is a bubble. The episode also covers a dust-up between Citadel and the Situational Awareness newsletter, a close encounter with Ben’s vibe-coded app, and an emailer’s theory on permanent Daylight Savings Time.

Ben Thompson and Andrew Wilkinson discuss OpenAI's recent sandboxing failure, exploring what it reveals about AI safety and the company's track record. They examine Kimi K3 and the emerging open-weights AI market in the U.S., contrasting the training and inference markets and analyzing why intelligence may become a commodity. The episode concludes with geopolitical tensions over Chinese AI models, regulatory disagreements, and the risks of building on Chinese infrastructure.

Ben Thompson and Andrew Wilkinson discuss OpenAI's first hardware product announcement and the strategic implications of OpenAI's hardware ambitions. They analyze Apple's trade secrets lawsuit against OpenAI, including details about how an OpenAI employee's recklessness triggered the legal action. The conversation spans AI's effect on mainframe spending at IBM, how AI will actually manifest in real-world software, Meta's use of messaging features as a recruiting tool, and a brief take on daylight savings time.

Andrew Doyle and Ben Thompson discuss Meta's leadership challenges and messaging gaps, Microsoft's failed gaming strategy with Xbox and Game Pass, the trajectory of AI token costs, and tangential reflections on youth sports economics, American soccer infrastructure, and the gap between tech practitioners and mainstream adoption.

Ben Thompson and Andrew Sharp tackle a wide-ranging summer mailbag episode. They dive deep into the memory chip market, discussing supply constraints, rising valuations, and why recent Apple price moves signal trouble. The conversation turns to Ben's 10-day experiment with AI-assisted 'vibe coding,' along with thoughts on caffeine habits, startup mythology, and OpenAI's consumer app ambitions. Other rapid-fire topics include an EU push on Apple, ocean data centers, metricmaxxing, and a proposed fix for international soccer hydration breaks.

Ben Thompson and Andrew Thompson cover Anthropic's export control challenges, the Trump administration's AI cybersecurity misconceptions, and regulatory risks of ringfencing frontier AI. They discuss Fox's strategic Roku acquisition, Apple's Siri AI as ChatGPT competition, Noam Shazeer's move to OpenAI, the limits of e-commerce automation, and whether autonomous driving can scale on gasoline vehicles.

Ben Thompson and Andrew Calloway discuss five key questions about Apple's WWDC 2026 announcements and the company's positioning on AI, including the Google partnership, Siri's limitations versus frontier AI, and why memory concerns are overblown. They then pivot to analyzing Fable 5's safety guardrails, Anthropic's incentive alignment between safety concerns and business strategy, and what the company's standoff with the Department of War reveals about AI governance.

Ben and Andrew discuss SpaceX's June IPO and its fundamentally misunderstood business model, Nvidia's earnings and the emerging neocloud market, why Google's search has degraded despite dominance, and the cultural backlash to Jony Ive's Ferrari Luce design. The episode covers investment hype cycles, the durability of competitive advantages, and how luxury brands signal exclusivity through design.

Ben Thompson and Andrew Doyle analyze the shift from training to inference compute as a market inflection point, examining Cerebras's differentiation strategy, why latency may matter less than throughput for agentic workloads, and implications for Nvidia and China's AI ecosystem. They explore xAI's dual-company structure — Colossus as training asset versus commercial inference product — and what the Anthropic-xAI compute partnership signals about market consolidation. The episode covers Elon Musk's OpenAI lawsuit strategy, Apple's gross margin expansion, and the founder-mode moment emerging from Claude access.

Ben Thompson and Andrew react to Amazon's strong AI-driven earnings by retracing AWS's original cloud strategy and showing how the company is now applying that same playbook to custom AI silicon with Trainium. They then break down the latest OpenAI-Microsoft deal, explaining Openai's motivation to run on AWS and why this resolves Microsoft's conflict of interest. The episode closes with thoughts on Meta's forthcoming display glasses, the wider AR device race, a monumental patent lawsuit in the wearable space, and an eye surgery epiphany.

Ben Thompson and Andrew Sharp answer listener questions on AI, including the future of consumer demand, Google’s Gemini stalling, compute constraints forcing price hikes, and the diverging AGI strategies of Anthropic and OpenAI. They explore how to message AI to a skeptical public, drawing on an Einstein-inspired thought experiment. The second half covers Amazon’s acquisition of satellite operator Globalstar and its implications for competition with Elon Musk’s Starlink, plus brief notes on Allbirds’ pivot, the ZIRP hangover for engineering talent, Evan Spiegel’s advice for Meta, and the news business.

Thompson analyzes the emerging alliance between Spotify and Netflix against YouTube. Both companies face the same threat: YouTube's free, ad-supported model undermining their subscription businesses. Their response: exclusive content and superior curation.

Thompson analyzes Google's Gemini models and argues that Google's custom TPU chips provide a structural cost advantage that may matter more than model quality in the long run — AI is becoming an infrastructure competition.

Thompson argues that the companies winning the AI race face a paradox: the more they invest in AI infrastructure, the more they commit to a technology whose value proposition is not yet clear. The winner of the AI race may be cursed by overinvestment.

Thompson examines the trend of nations building their own AI capabilities rather than depending on US companies — the geopolitical implications of treating AI as critical infrastructure, like energy or defense.

Judge Mehta rules that Google is a monopolist in search. Thompson argues this is the most important antitrust ruling in tech since Microsoft in 2001 and analyzes what remedies might follow — from behavioral changes to a forced sale of Chrome.

Thompson analyzes Apple's late entry into AI: its integration advantage (controlling hardware, OS, and distribution) versus its innovation disadvantage (no frontier model, no research culture). The question: can integration compensate for arriving last?

Thompson argues that AI assistants will eventually replace traditional search — and this isn't a crisis but an improvement. Search was always a compromise: you wanted an answer, but the system gave you a list of links. AI gives you the answer directly.

Thompson reflects on 10 years of Stratechery: the predictions that were right (Aggregation Theory, Apple's services revenue), the ones that were wrong (Twitter's potential, Facebook's VR pivot), and what he's learned about the practice of technology analysis.

On Apple's 50th anniversary, Thompson examines why Apple's vertical integration strategy — controlling both hardware and software — has produced the world's most valuable company despite violating every MBA textbook on specialization.

Thompson provides the definitive analysis of the OpenAI board crisis. He argues the real lesson isn't about Sam Altman's leadership — it's about the impossibility of governing a trillion-dollar technology through a nonprofit board structure.

Thompson analyzes Meta's launch of Threads to compete with Twitter/X: why Meta's distribution advantage makes Threads a credible threat, and what the competition reveals about platform economics.

Thompson analyzes why TikTok's algorithm represents a fundamental shift in social media: from social graph (who you know) to interest graph (what you like). This shift explains TikTok's explosive growth and why Facebook, Instagram, and YouTube are scrambling to copy it.

Ben Thompson revisits his most influential framework — Aggregation Theory — in light of Apple's App Store battles, Google's antitrust trial, and the rise of TikTok. He argues the theory's predictions have largely held but need updating for the AI era.