The legend of Analyst #3# grows stronger… “As a final push over the edge for Scorpion Antivenom, Analyst #3# proposed that once we’d gotten the lab data back, he was willing to volunteer to cut to the chase, skip the animal trials and get stung by a scorpion to see if it worked - provided we conducted this exercise outside of a hospital that stocked the FDA-approved antivenom.” @citrini

A non-tech, non-AI idea that is 1) very cheap 2) defensive 3) has idiosyncratic value creation opportunities 4) is already executing on those opportunities 5) good mgmt 6) smart/aligned capital allocation 7) real market cap - can own a lot… $MKL is trading ~5x earnings if they were to pursue the Jana plan to return excess capital and spinoff ventures. And trading <8x my estimate of 2027 earnings power as a consolidated entity. $MKL has endured a well documented stretch of self inflicted issues, lasting nearly a decade. In that difficult stretch it still grew tangible book value >10%, a testament to the balance of its business model, engineered for all-weather compounding. The insurance business has recently turned a corner and is under new management, but the market is sleeping on it. Insurance is a business where the people running it really matter and $MKL Insurance now has a real star in Simon Wilson, who took over as CEO in March 2025. He formerly ran Markel international and put up numbers like $KNSL (rapid growth and 70s-80s CR). In his first year Simon has returned the org to the lean decentralized model it was founded on, modernized archaic tech systems holding them back, deployed AI to automate claims management and other manual processes, exited troublesome and capital intensive markets where MKL had no edge, and has improved the profitability of the business despite a soft market backdrop. While MKL’s Insurance turnaround is evidently working. The market isn’t rewarding it yet as MKL trades close to its lowest P/B and P/E excluding GFC and Covid periods. Thankfully the company recognizes the disconnect and is aggressively buying back stock. ~94% of the market cap is covered by the value of ventures (12x trailing EBIT) + public equities + net cash. The market is putting very little value on insurance earnings. Valuing ventures at 12x trailing EBIT, and assuming stocks were sold, paid >$2B taxes on gains, and excess regulatory capital was returned in a dividend, the remaining insurance business would be trading <1x book and ~5x earnings. This for a business with ~10% organic growth, and an improving combined ratio despite a soft market backdrop (demonstrating successful idiosyncratic value creation). Highlights: - New CEO Simon Wilson is deeply strategic, analytical, and attracts and retains talent by enabling them to succeed. - Jana involved and adding - Record buyback pace - Growing insurance profitability in a hard market. Hard changes are in rear view. - Cheap on earnings power, book value, and has excess capital equal to ~20 of market cap - Exiting reinsurance will free up ~5% of capital with no impact on earnings - Aligned comp model: Gaynor gets paid based on long-term share price CAGR

AWS Bedrock is driving 45% of Anthropic API ARR per @SemiAnalysis_ . The market continues to misunderstand and underestimate the degree of positive sum commercial alignment between $AMZN $MSFT $GOOG and Anthropic/OpenAI, treating them more like competitors. In reality the clouds are critical infrastructure providers w/ their compute fleets (~75% of global compute), and demand creators/enablers with commercial sales and implementation ecosystems to handhold customers as they deploy agentic apps in enterprise environments.

Hyperscale clouds + $META account for ~75% of incremental GW coming online in 2026-2028, and I expect them to have even larger share of projects that are successfully approved, completed, and delivered on time: 1) ~1/3 of incremental GW to be added are expansions of existing campuses, Phase II of projects partially completed, or densification of existing facilities. These projects face fewer permitting and power hurdles than greenfield projects. 2) The clouds dominate the secured power pipeline, and this will continue for the foreseeable. Utilities, IPPs, or BTM fleet operators who are building the power systems for DCs need financing. It is infinitely easier to finance a project with $MSFT $AMZN $GOOG $META as a tenant and a long-term contract than with any other party. 3) Neocloud and spec developer deals face a structural adverse selection problem, since the preferred partner for any project is a hyperscaler, and will on avg face more permitting, power, and supply chain problems. 4) The Hyperscalers have been building large DC projects at massive scale for a long time. They know what they’re doing. Only ORCL, CRWV, and Xai have built >1GW, outside of the big 4. Note: the below is from data that I track

In the past year, the AI inference market has seen two massive accelerations, the 1st was Claude Code in Jan-Mar, and we're seeing the second one right now with Fable/Codex this summer (ironically as the AI trade was imploding in July). Both accelerations are clearly visible in Lab ARR (loosely tracked via press leaks and 3P est.) (slide 1/2): Since AI datacenters have 12-18mo lead times, supply could not keep pace with these sudden leaps in model capability and demand, driving inflation for anything in the AI supply chain YTD (most notably memory). This has created pricing power for clouds controlling scarce compute, which I've written about and the market is beginning to appreciate (but still underestimates), and for AI labs, which have seen rising ARR per GW of inference capacity (slide 2/2 next tweet):