A house newsletter · daily · researched and written by the Kurzia AI
Saturday 14 November 2026 · Issue 13 · 3 min read
Nvidia's chequebook goes quiet; LangChain at $3B; Anthropic prices Nov 18
The story that matters most
TechCrunch · yesterday
Huang's exit leaves AI funding without its backstop
Saturday, so a short one, and the week's biggest change for cheque-writing isn't a round but Nvidia: with Jensen Huang gone, TechCrunch reports the startup-investment machine he personally drove is expected to retrench under interim CEO Colette Kress, whom one investor calls much more conservative.
Investors told TechCrunch that with Nvidia off the cap table, pre-revenue infrastructure companies lose the reassurance that let them raise at $1 billion-plus valuations. Crunchbase flags Nvidia's Nov 17 earnings as the first read on how much of Huang's playbook survives under the new leadership.
“Nvidia is carrying almost every AI-native startup from application to neocloud”
Rounds and exits
LangChain — agent development and observability platform behind LangSmith, raised a $300M Series C led by CapitalG at a $3B valuation, with GIC and DFJ Growth joining.
Databento — market-data APIs, raised a $100M Series C led by Fifth Down Capital at a $1.1B valuation; five of six frontier AI labs are customers.
Anthropic — prices its Nasdaq debut at $70–78 a share, with final pricing expected Nov 18 and trading Nov 19 under ticker ANT; Q3 revenue was $17.0B.
What the AI labs and big platforms did
ChatGPT Wallet — lets OpenAI's dots agents finish purchases for US ChatGPT Pro subscribers, each checkout approved by fingerprint or Face ID and paid with a card-network token instead of the real card number. It flattens standalone agent-checkout and personal-shopper startups and opens a merchant-enablement category — OpenAI published a Commerce Partner Playbook and signed fraud and payments partners including Stripe and Adyen. Visa ran 22 models through shopping tests this month; 42.8% of the attempts went wrong, The Next Web reports. TechCrunch
For your portfolio
Clouded Judgement · yesterday
Fable and Mythos 5.2
Takeaway: $8/$40, a 20% cut
Jamin Ball's Friday note argues frontier-model pricing has lost its stability: two months after GPT 6 Astra and Fable 5.1 held at $10/$50 per million tokens, Anthropic's Fable and Mythos 5.2 shipped at $8/$40, a 20% cut.
The new models run about 30% faster, which Ball works out to roughly 23% lower cost per token, so Anthropic is mostly passing efficiency through rather than discounting from weakness — and he expects the pattern to repeat the way Databricks and Snowflake cut prices as serving costs fell. Every 20% off frontier inference widens the gross margin of model-dependent app startups; any moat built on cheap tokens alone narrows with it. Question for any founder with model-heavy COGS: when the frontier price drops another 20%, does your own price follow it down, or does the saving stick to gross margin — and if it sticks, what stops competition from handing it back?