2025 Predictions: Tech, Business, Media, Politics!

Chamath’s strongest payments thesis was that dollar stablecoins had already crossed from crypto speculation into business infrastructure. He cited roughly 1.1 billion Q2 2024 transactions worth $8.5 trillion—more than double Visa’s volume—and said switching his own payment rails could save hundreds of thousands of dollars.

Freeberg’s “year of the robot” call rested on hardware already cheap enough to deploy, not distant prototypes. His group had just ordered China’s Unitree Go2, a $1,600 API-enabled robot they wanted on test farms for imaging and data collection; the humanoid G1 was discussed at about $16,000.

Gavin Baker argued that reasoning AI will increasingly reward companies rich enough to own compute, widening the gap between large and small businesses. He imagined a firm spending $1 million to let an AI reason for six weeks, and separately argued that renters of cloud compute cannot beat providers’ internal services on cost.

Baker’s clearest picks-and-shovels AI bet was high-bandwidth memory, which he viewed as a tighter bottleneck than the headline GPU makers. He said HBM represented more GPU cost than TSMC fabrication, named Hynix and Micron as the two current producers, and noted the category had been sold out for two years.

Gavin and Chamath converged on enterprise application software as a major AI casualty because agents attack both labor and software economics. Chamath said his roughly 30-person engineering team did the work of 300 and could profitably bid $10 against a legacy vendor’s $100 because AI-native costs were about an order of magnitude lower.

The autonomy discussion contained one unusually concrete adoption signal: Waymo was said to have reached 22% of San Francisco rides only 15 months after launch, matching Lyft. The investor making the case expected major financing, an IPO or tie-up as expansion broadened, while explicitly disclosing significant exposure to almost all the companies discussed.

Chamath treated a major-bank failure as unlikely but serious enough to justify buying credit-default-swap protection. His rationale was that roughly $70 trillion of combined government, corporate and mortgage debt makes 5% rates economically much heavier than they look; he described the hedge as usually losing but potentially paying 100–1,000 times.

Jason Calacanis’s most persuasive AI demo was a research task that assembled a bank-risk report from 162 websites using a $20 product. He compared the output to weeks of expensive consulting work, while also admitting he was not qualified to judge whether its conclusions were correct.