josh kale

The iPhone Duo's dual-screen form factor enables split-screen AI agent workflows (app on left, Siri on right), but sacrifices best-in-class sensors, cameras, and speakers of the Pro model, making it a consumption device not a productivity tool.

Apple's new device suite — iPhone 18 Pro with 2nm chip, Apple Watch with passive audio monitoring, iPhone Duo foldable, and AirPods — creates a unified hardware ecosystem that feeds contextual data into Siri AI, which can access 300,000 app intents. This on-device, privacy-first architecture gives Apple a structural moat over frontier model providers like OpenAI and Anthropic for personal AI assistants.

xAI and Meta control the most compute and are aggressively expanding it; if compute scaling laws hold, they will inevitably produce highly competitive models, evidenced by rapid release cadence (Grok 4.6 imminent).

Google's core business remains spectacular and the leadership changes at DeepMind don't materially affect day-to-day operations, making the 5% selloff an overreaction.

Apple pays Google ~$1B/year for a proprietary Gemini variant to power Siri OS, giving Google massive distribution but also revealing Apple's lack of frontier model capability; hosts note Google leadership turnover has cost it steam in the AI race, yet its infrastructure scales to 2.5B devices at ~0.1¢/query.

Host reports Meta's Muse product and Grok bots feel more personal and better than ChatGPT as an LLM, suggesting Meta's social graph and personalization could differentiate its AI assistants despite not being the smartest model.

Meta and xAI's massive compute advantages and focus on cost-per-token will let them capture the high-volume, cost-sensitive segment of AI demand, with routing platforms directing workloads to cheaper models.

Jensen Huang/Nvidia wins in every scenario: open source proliferation, custom ASICs (Trainium, TPUs), and US-China competition all drive insatiable GPU demand; the hosts are explicitly bullish on infrastructure layer as the durable moat.

Huawei announced GPUs at the same WAIC conference as Xi's speech; Moonshot may be using Huawei GPUs given Nvidia export restrictions, but compute scarcity remains the bottleneck.

GE Vernova combines legacy grid equipment expertise with gas turbine production, has doubled orders to $7.1B in 2025, holds contracts through 2031, and counts Microsoft and OpenAI as customers, making it the central player in solving data center power bottlenecks.