A house newsletter · daily · researched and written by the Kurzia AI
Wednesday 4 November 2026 · Issue 3 · 3 min read
Fal doubles to $16B; OpenAI meters ChatGPT and Codex together; Menlo sizes enterprise AI at $177B
The story that matters most
Fal — Fal, which runs image, video and audio generation models for developers behind a single API, raised $1.3B in a Series F led by Bessemer Venture Partners, announced yesterday alongside its Sequoia-led April Series E — $1.65B across the two. The deal values the company at $16B, double its $8B April valuation, reported by Bloomberg via TechCrunch.
CEO Burkay Gur told Bloomberg Fal had to rate-limit customers this summer when demand outran its compute; rivals repriced the same way, with Modal raising at $15.75B in October and Fireworks at $17.5B in July. But several people familiar with inference companies' books tell TechCrunch gross margins stay thin sector-wide. Underwrite this layer on gross margin after compute, not on growth. Fal now fine-tunes video models together with their creators, sometimes in return for licensing fees or a share of revenue. TechCrunch
Rounds and exits
CyberKord — raised €2.7M pre-seed to automate IT and security operations with AI; led by 360 Capital; reported by tech.eu.
Ballistic Ventures — filed with the SEC to raise $490M across a third flagship fund and a $130M Fund B; no sales recorded yet; reported by TechCrunch.
October venture funding — reached $59B globally, up 45% year over year, with DeepSeek's $15.9B at a $90B valuation the largest Chinese deal ever; per Crunchbase.
What the AI labs and big platforms did
Shared ChatGPT–Codex allowance — OpenAI told subscribers ChatGPT and Codex will soon share one usage allowance as Work's agent features move into the main app; Instant chat stays unmetered, and Pro 100, 200 and 500 allowances were doubled effective October 29. Tunguz reads the new meter as the labs' answer to spiking short-term compute costs — Anthropic's $5B of burst spend flipped a Q2 operating profit into a Q3 loss. Flat-rate subscription resellers get squeezed; usage-priced tools gain. BleepingComputer
Google's schools can't switch off AI without losing Search. Google said Tuesday its AI tools will be off by default for K-12 students next year unless administrators enable them — and schools that keep AI off lose Google Search too, since the two now ship as one product. Edtech firms selling into Chromebook districts face a default-off gate, with Google's own Guided Learning tutoring mode positioned as the compliant option. The opt-out covers Gemini, NotebookLM, Gemini in Classroom and Search, all or nothing. TechCrunch
“Search and AI are deeply integrated”
For your portfolio
2026: The State of Enterprise AI — Menlo Ventures' annual report, published yesterday, sizes enterprise AI deployment spend at $177B by September 2026, up 5.4x from $33B a year earlier. Agents now generate 80% of tokens on OpenRouter, and the three frontier labs take an estimated 69% of deployment dollars, with Anthropic at 55% of enterprise LLM API spend and Google down to 9%.
For underwriting, start with where the money came from: about half the year's new AI dollars came from budgets that did not exist a year ago, not from raided SaaS renewals — the 'AI eats software' panic oversells displacement. Then concentration: when one lab's API sits under most of your portfolio's products, that lab's pricing and roadmap are portfolio-level risk. Claude Code and Codex already take $24B of the $34B coding market, per Menlo — ask any portfolio company building on a lab API what its wedge is when the lab ships the same feature itself. Menlo Ventures
Postman desktop — A GitHub issue still open, with hundreds of reactions, documents Postman's desktop app installing plugins into Claude Code and OpenAI Codex without asking, self-approving its own startup hook, and enabling a remote MCP connection; Postman's own log confirms the automatic install, and The New Stack reports the company has not responded.
Any portfolio company shipping agent or MCP integrations just got its cautionary tale: the trust boundary now sits inside the customer's coding agent, and one silent config change can cost the account — in regulated shops, the contract. Ask each one: when your product touches a customer's agent, who approves the hook — the user inside the agent, or your installer? Anything but the first is behind. GitHub