Daily AI Briefing — May 30, 2026
The AI industry reached an inflection point this week. Anthropic is on track for its first-ever quarterly operating profit, OpenAI filed its S-1 for what could be the largest tech IPO in history, and Google shipped Gemini 3.5 Flash at Google I/O — prioritizing distribution over benchmark scores. Meanwhile, enterprise adoption accelerated: Salesforce used Claude Code to cut a 231-day migration to 13 days, and DeepSeek V4 Flash cut prices another 75%. Here's what you need to know.
Anthropic Projects $10.9B Q2 Revenue — First Operating Profit
Anthropic informed investors it will more than double revenue to approximately $10.9 billion in Q2 2026 and deliver an operating profit of roughly $559 million for the first time, per CNBC. This follows Q1 revenue of $4.8B and represents 130% QoQ growth.
The milestone is significant because it proves frontier AI labs can be profitable while still training next-generation models. Anthropic's annualized revenue run rate now exceeds $50B — a trajectory that could put it past $100B by year-end. The company's lifetime capital raised stands at over $72B, including a $1.25B/month GPU compute contract with SpaceX (revealed in SpaceX's IPO prospectus).
This fundamentally changes the IPO narrative for both major AI labs. If Anthropic can show sustainable profitability at scale, it validates the thesis that enterprise AI — specifically Claude's safety-first positioning — commands premium pricing and long-term contracts.
OpenAI Files Confidential S-1 — $1 Trillion IPO in Sight
OpenAI filed a confidential S-1 with the SEC on May 22, 2026, targeting a public market debut in September 2026 at a valuation between $852 billion and $1 trillion, per CNBC. Goldman Sachs and Morgan Stanley are leading the offering.
OpenAI's current annualized revenue exceeds $25B, driven by ChatGPT subscriptions, API usage, and its new DeployCo consulting arm. The company's GPT-5.5 Instant model (quietly made the default in ChatGPT on May 5) reports 52.5% fewer hallucinations on high-stakes prompts. OpenAI has also expanded its computer-use capabilities — Codex can now control macOS and Windows PCs autonomously, even while locked.
The dual IPO path (OpenAI in September, Anthropic likely following in 2027) creates a once-in-a-generation opportunity for public market investors to buy into frontier AI. The key differentiator: Anthropic is already projecting profitability; OpenAI's filing will need to demonstrate a credible path to the same.
Google I/O 2026: Gemini 3.5 Flash GA & Distribution Over Benchmarks
Google I/O 2026 shipped major product updates. Gemini 3.5 Flash went GA at $1.50/$9 per 1M tokens with a 1M-token context window, scoring 76.2% on Terminal-Bench 2.1 and beating Gemini 3.1 Pro on coding and agent tasks. Google claims 4× speed over comparable models.
Sundar Pichai summarized the strategy to Axios: "Stay at the frontier, but prioritize models cheap and fast enough to deploy across products used by billions." Google also launched Chrome DevTools for agents (compatible with Antigravity and 20+ coding agents), a $100/month AI Ultra plan for developers, and Modern Web Guidance for AI coding tools.
The takeaway: Google can lose the benchmark war and still win via distribution — YouTube, Search, Gmail, Android — with 3B+ daily users. For developers, this means cheaper, faster models with deep platform integration, even if they're not chart-topping on every leaderboard.
Salesforce Cuts 231-Day Migration to 13 Days with Claude Code
Salesforce moved its entire development organization to Claude Code with no token limits, achieving dramatic productivity gains in April 2026:
- 79% more pull requests per developer
- Five-fold reduction in migration incidents
- A 231-day migration completed in just 13 days
This is one of the strongest enterprise productivity case studies for AI coding agents to date. When a company of Salesforce's scale — with thousands of developers and complex enterprise codebases — reports these numbers, it changes the ROI calculation for every CIO evaluating AI coding tools. For comparison, Cursor, GitHub Copilot, and Codex CLI are the other major options in this space, but Claude Code's unlimited token context and agentic workflow appear to be the decisive advantage for large-scale migrations.
DeepSeek Makes 75% V4-Pro Price Cut Permanent
DeepSeek has made its 75% price cut on V4-Pro permanent, intensifying the already brutal inference pricing war, per AI Weekly. DeepSeek V4 Flash continues to undercut every frontier model on price while maintaining competitive coding and reasoning benchmarks.
The broader trend: Chinese models — DeepSeek V4, Kimi K2.6, GLM-5.1, Qwen 3 — now account for 60% of usage on OpenRouter. The open-weights tier is now Chinese-led, with Meta's Avocado model remaining notably quiet. For cost-conscious developers, the calculus is straightforward: Asian open-weight models deliver 80-90% of frontier capability at 10% of the cost.
Meta Raises Capex to $145B, Debuts Muse Spark
Meta raised its 2026 capital expenditure guidance to $125–145 billion (up from $115–135B), rattling public market investors, per Fortune. The spending funds Meta's Superintelligence Labs under Chief AI Officer Alexandr Wang — hired for $14B via the Scale AI acquisition.
Meta's first model under the new labs is Muse Spark (released April 8), which delivers competitive performance at lower compute cost than Llama 4. The company is also developing four new MTIA chip generations (300–500) to reduce Nvidia dependence. Meta is betting that massive infrastructure investment will unlock AI features for its 3B+ social media users — but the timeline for commercial payoff remains unclear.
Pentagon Tests OpenAI and Google AI to Replace Claude
The Pentagon is testing AI models from OpenAI and Google on classified workflows previously served exclusively by Anthropic's Claude, per recent reports. Anthropic turned down a DoD contract on ethical grounds in early 2026, citing concerns about autonomous weapons. The Pentagon is now hedging — no single vendor becomes the sole national security AI provider.
This creates an interesting strategic tension: Anthropic's safety-first positioning wins enterprise trust (Deloitte, PwC, KPMG) but loses government defense contracts. OpenAI's willingness to work with the Pentagon generated a #QuitGPT backlash (2.5M supporters) but opens a massive government revenue stream. Both strategies are internally consistent, but they point to a future where "safety" means different things in enterprise vs. government contexts.
Andrej Karpathy Joins Anthropic
Andrej Karpathy — OpenAI co-founder, former Tesla Autopilot lead, and arguably the most beloved AI educator — has joined Anthropic to rebuild its pretraining research team. This is the highest-profile AI talent move of 2026 and sends a strong signal to senior researchers about where the most interesting pretraining work is happening.
Karpathy joins a growing roster of talent at Anthropic, following Christopher Olah (interpretability lead, co-presented Pope Leo XIV's AI encyclical) and Dario Amodei (CEO). The concentration of research talent at Anthropic is becoming a competitive moat that's hard to replicate.
Quick Hits
Genesis AI releases Genesis World 1.0: A four-component simulation platform for robotics, achieving 0.8996 Pearson correlation for scalable evaluation. Released May 27.
DeepSeek researchers face travel restrictions: Chinese AI researchers at DeepSeek, Alibaba, and other firms now need government approval to travel abroad, framed as protecting proprietary knowledge. Likely long-term effect: talent flight as restrictions tighten.
Armada raises $230M for deployable AI data centers: Modular units that deploy in days and run on local energy — already on U.S. Navy ships and oil rigs. Edge AI moves from pilot to procurement.
AI companies engage FERC on data center power: FERC preparing a June proposal to speed up data center connections to regional grids. Power availability, not model capability, is becoming the primary constraint on AI growth.
Canada rules ChatGPT violated privacy law: Canada's Privacy Commissioner found OpenAI violated PIPEDA — the first national ruling that training data collection itself constitutes a violation. UK ICO, German DPA, and French CNIL conducting similar investigations.
Tomorrow's Watch
The key story to track over the weekend: whether OpenAI's S-1 filing is made public (confidential filings typically become public 15-21 days before the roadshow) and what it reveals about the company's financials — particularly inference costs, revenue concentration, and the profitability timeline. Also watch for any Anthropic response to the Pentagon's multi-vendor testing — a formal enterprise AI defense offering could change the government contracting landscape.
← Back to all posts
