OpenAI files. Anthropic projects first profit. Both labs walked to the public markets in the same week.
OpenAI is confidentially filing its IPO prospectus with the SEC today at an $852B private valuation. Anthropic told investors Q2 will land at $10.9B revenue with $559M operating profit — first profitable quarter, three years ahead of guidance. Plus Claude Managed Agents got MCP Tunnels + self-hosted sandboxes at Code with Claude London.
Friday morning. OpenAI is confidentially filing its IPO prospectus with the SEC today — WSJ's scoop, with CNBC, Reuters, Axios, and Quartz independently confirming. Goldman Sachs and Morgan Stanley are running the deal. The company's last private valuation was $852B; the public framing of a trillion-dollar listing would make this the largest tech IPO in history. The confidential filing window means a public S-1 sits sixty to ninety days behind today, so the earliest visible prospectus lands in late July or August, and the company is targeting Labor Day to Thanksgiving 2026 for the actual listing window. 123
Forty-eight hours earlier on the other side of the duopoly, Anthropic told investors — via the financing-document WSJ surfaced — that Q2 2026 will land at $10.9B in revenue, a 130% jump over Q1's $4.8B, with $559M of first-ever operating profit. Last summer's guidance said no full-year profitability until 2028 at earliest; that window just compressed by three years. Stock-based comp is excluded; model training cost is in. The growth rate currently outpaces the historical Q-over-Q peaks of Zoom, Google, and Facebook. Public listing open by year-end, October the most-cited window. The duopoly narrative is no longer about who's at the frontier — it's about who lists first. 456
Underneath the capital story, Anthropic shipped two new pieces of agent infrastructure at Code with Claude London (Wed-Thu). Self-hosted sandboxes (public beta): Claude Managed Agents now execute tool calls inside customer-managed environments — Cloudflare, Daytona, Modal, Vercel, or the customer's own infra. MCP Tunnels (research preview): Claude agents reach private MCP servers behind enterprise firewalls without exposing those servers to the public internet — single outbound encrypted connection via a customer-deployed lightweight gateway. No inbound firewall changes required. The agent-orchestration plane stays on Anthropic; the execution plane is the customer's. This is the enterprise-grade architecture pitch — and the structural answer to the question every CIO has been asking about agent deployment governance. 8910
Carry from Thursday catch-up: the $15B/year Anthropic-xAI Colossus deal is now widely understood. Spotify's 4,500-deploys/day Claude story continues to propagate as the reference enterprise customer. @decodeddaily07's 'three labs, three scarcities' frame still reads cleanly — and the IPO twin-filing this week is the most concrete possible expression of Anthropic and OpenAI's competing scarcity bets meeting the public markets at the same moment. 162021
SP-Index +1 to 74. Capital substrate moved on twin IPO trajectory. Frontier release velocity moved on agent infrastructure ship. Everything else holds. The structural read for Q3: public-markets discipline arrives at both labs in the same window. The next twelve months are the lock-up and lookahead window — and the singularity narrative writes itself in S-1 risk-factor sections. 148
Two confidential filings, one duopoly. The singularity now has a ticker.