The Board Room
Uber's CTO publicly admitted burning through the company's entire 2026 AI budget in
Meanwhile, teams running optimized inference stacks operate at 5-8x lower cost than default deployments, meaning the financial gap between AI leaders and laggards widens with every API call your team makes.
Enterprise AI Budgets Just Broke — Consumption Pricing Hits
Uber exhausted its full-year AI budget in months. Anthropic shifted to consumption pricing. TSMC's 40.6% beat confirms demand is real. But GPT-4-class inference fell 50x to $0.40/M tokens — the 5-8x cost gap between optimized and naive deployments is now the largest hidden P&L variable in enterprise tech.
The Open AI Commons Is Dead — Three Giants Close the Door
Meta's Muse Spark is entirely proprietary (first product from its $14.3B Scale AI acquisition). Alibaba reserves its most capable models for cloud customers only. Anthropic gates Mythos at 13.5 SWE-bench points above Opus 4.7. Independent testing shows a $0.11/M token model found the same bugs Mythos showcased — the moat is scaffold, not model.
AI Backlash Goes Bipartisan — Infrastructure Constraints Follow
AI now polls below ICE with Americans — 77% see it as a risk to humanity, and enthusiasm lags China by 46 points (38% vs 84%). An anti-AI-slop site hit 25M uniques in 30 days. 40% of 2026 data center projects face delay from community opposition. Maine enacted the first data center construction moratorium. This is a compounding constraint, not a comms problem.
Block's 'Dorsey Mode' Sets the Org Restructuring Template
Block cut 40% of headcount and flattened to 2-3 management layers, betting AI automates middle management within 3 years. Mutiny killed 8-figure ARR SaaS to go all-in on AI. But the Jevons Paradox counter-argument — that AI efficiency will massively expand developer demand — is gaining traction among 30K+ engineering leaders. Both theses can't be right, and your org model depends on which is.
Tech at 2018 Multiples with 43% Earnings Growth — M&A Window Opens
Tech trades at a 25% market premium — 2018 levels per Goldman — while earnings growth surged to 43.4%. Insider buying hit a 15-year high. a16z's Casado publicly declares AI models 'not hard to build,' signaling the moat has moved to data and distribution. 37% of enterprises now report quantifiable AI ROI, up 23% QoQ. This is a rare acquisition window that may last 2-3 quarters.
Enterprise AI Costs Just Broke — The Consumption Pricing Inflection
The Open AI Commons Is Dead — Three Giants Close the Door in One Week
Block's 'Dorsey Mode' vs. the Jevons Paradox — The Org Design Bet of the Decade
- OpenAI Codex superapp hits 3M weekly users with 70% MoM growth — consolidating ChatGPT, Atlas, background computer use, and parallel agents into a single workspace that threatens every single-purpose developer tool
- Amazon acquires Globalstar for $11.57B — gains satellite spectrum, government contracts, and Apple's Emergency SOS dependency; connectivity is becoming a proprietary platform layer, not a commodity utility
- OpenAI slashed ChatGPT ad CPMs 58% in 9 weeks and dropped minimums 80% to $50K — building an advertiser base at blitz speed with self-serve tools, Criteo partnership, and CPA/CPC models in development
- Update: Opus 4.7 safety red flag — Anthropic's attempt to differentially reduce cyber capabilities during training failed; model scores higher than 4.6 on exploitation benchmarks despite deliberate constraints
- Uber commits $10B+ to physically owning robotaxi fleets ($7.5B purchases, $2.5B equity in WeRide/Lucid/Nuro/Rivian/Wayve) — repudiating the asset-light platform model as Waymo simultaneously removes US waitlists and begins London testing
- Update: 1,500+ state AI bills now in play — California watermarking mandate Aug 2026, Colorado algorithmic discrimination law July 2026, New York protocols for $500M+ revenue model makers Jan 2027; compliance surface rivals GDPR in complexity
- AI-driven commerce traffic surged 393% in Q1 — Yale/Columbia research shows product naming changes swing AI agent selection by 41-80 percentage points; 'Sponsored' labels actually reduce AI agent selection rates
- Eli Lilly's $2.75B Insilico bet validates AI drug discovery at enterprise scale — compressed discovery from 5-6 years and 200K+ compound screens to 18 months and fewer than 80 compounds
- Compute scarcity reframes AI economics: Ben Thompson argues opportunity cost — not marginal cost — is the binding constraint, calls OpenAI 'serially unfocused' and potentially 'the biggest loser' in a scarce-compute market
Three AI giants — Meta, Alibaba, and Anthropic — simultaneously moved their best models behind paywalls this week while Uber's engineers blew through a full-year AI budget in months under the new consumption pricing regime. The 5-8x cost gap between optimized and naive inference deployments means the financial winners and losers of this era are being decided not by which model you pick, but by how you run it — and independent testing showing a $0.11/M-token model matching Anthropic's $25/M Mythos on flagship tasks confirms that the moat has moved from model capability to system engineering and organizational design.