This is All-In's 'Best Ideas Pitch Competition,' modeled on the format of the late investor Ira Sohn's charity conference, in which four fund managers each pitch a non-consensus name in six minutes, with the judges (several of the besties + guest Gavin Baker) commenting and scoring on the spot.

The first, Aaron Cowen (Serreta Capital, who ran Soros's equity business and served as Steve Cohen's CIO), pitches US casino stock MGM Resorts, and ultimately wins the judges' grand prize. His logic isn't about Las Vegas itself, but about two 'hidden assets' the market hasn't seriously priced: one is an Osaka, Japan casino license opening in 2030 (geographically closer to Shanghai/Beijing gamblers than Macau or Singapore, and Japan is a developed country with few competing venues), and the other is 300,000 square feet of space quietly reserved inside a luxury complex in Dubai — usable the moment Dubai legalizes gambling. Add media mogul Barry Diller already holding 26% and formally bidding $48 to acquire it after Cowen finished this report (a financial buyer, in it to make money), and the downside is floored by the buyout price while the upside points to a double or triple — he flatly says 'don't sell your shares to Barry.'

The second, Daniel, pitches power stock Talen Energy, borrowing commodity veteran Sam Zell's framework of 'buying future-scarce hard assets below replacement cost': Talen holds 2GW of nuclear + 6GW of gas baseload power, with a current enterprise value of $25 billion versus a replacement cost as high as $45 billion, so the equity could double just returning to replacement cost. He likens data centers to 'refineries' — input power, output tokens/intelligence, at about $50 billion to build per GW, with power being the refinery's raw material. He stresses one key call: 'You don't need AI demand — the power market is already tight enough for the next twenty years; AI just turbocharges it and creates shortages'; Jensen Huang says 1,000x the current power is needed, and the PJM (Pennsylvania-New Jersey-Maryland) region alone must add 106GW over the next decade, roughly equal to all of Japan's electricity use, while the supply chain (critical minerals, nickel superalloys, silver for solar) is also competing for capacity with rocket launches and orbital data centers, doomed to shortages and delays everywhere. Microsoft persuading Constellation to restart the Three Mile Island nuclear plant (a floor price of $100/MWh, locked for 20 years) is a famous example of the scramble for power. Under three scenarios — 'doubling even doing nothing, $70/share signing more data-center contracts, $100+/share building its own plants' — he urges buying outright.

The third, Oleg Nelman (Eco R1 Capital, a value-oriented biotech fund), pitches radiopharmaceutical company Actis Oncology (AKTS). He uses abundant military metaphors to tell the history of fighting cancer: surgery and radiation are like a medieval siege, chemotherapy like mustard gas killing indiscriminately, targeted drugs like GPS guidance, immunotherapy like recruiting local allies, and radiopharmaceuticals like 'a swarm of micro-drones in the bloodstream using molecular recognition to find their target and detonate a warhead with 100-micron precision' — precisely detonating a single cancer cell. The Actis platform has a market cap of about $1 billion, an enterprise value of $500 million, and enough cash to last three years; Eli Lilly put in $100 million as an endorsement, the IPO was oversubscribed 18x, and the targets chosen are the mature nectin-4 (bladder cancer) and B7H3 (nearly all solid tumors); clinical data starts in Q1 2027. The biggest moat is that the supply chain for the radioisotope (actinium, from waste of America's early nuclear program) keeps Chinese copycats out. If a single program succeeds, he sees a $10 billion market cap, $200/share.

The fourth, Kyle Samani (founder of Multicoin Capital, an early heavy holder of Solana), pitches crypto asset GeoNet (token GEODE): it's the world's largest and fastest-growing RTK high-precision positioning network (centimeter-level, about 100x more precise than GPS), using crypto-token incentives to get individuals/small businesses to set up base stations on their own rooftops; its 22,000-node scale is already double the combined total of the traditional big three (Trimble, Hexagon, Topcon), yet at only 1/4 the cost of building your own. Customers include John Deere (autonomous spraying), DJI, TomTom (autonomous-driving maps), and millions of robotic lawnmowers, riding the network-effect dividend of robotics/drone 'physical AI'; 80% of revenue goes to open-market token buybacks (verifiable on-chain), with annualized revenue just past $1 million, up 3x YoY.

The judges' consensus: MGM and Talen can be sized up large because the downside is protected, while Actis and GeoNet are high-payoff 'lottery tickets' with go-to-zero risk; both Gavin Baker and David Friedberg caution that all AI-related names must face the political and regulatory risk of rising electricity prices, and that ground-based positioning networks like GeoNet may in the future be replaced by SpaceX-style low-orbit satellite constellations. In the end the judges' grand prize went to MGM, and the audience-vote winner was Talen.

(For AI investors: Talen represents a pure play on the 'AI power' theme, and GeoNet represents the positioning infrastructure for robotics/autonomous driving — both are easily overlooked pick-and-shovel links in the supply chain.)