Jonathan Ross

David Senra × Jonathan Ross

Groq’s rumored $20B NVIDIA partnership moved from first call to money in the bank in about three weeks, after Jensen Huang saw its GPU-LPU integration and proposed making it available to all NVIDIA customers.The architecture treats GPUs and LPUs as complements for different bottlenecks, while faster inference can deepen model search and compound across agentic AI, leaving compute scarcity and whether capital still creates advantage worth monitoring.

Jonathan Ross × John Yetimoglu

Memory’s pricing power may be self-limiting: Deep Seek’s V4 reportedly compressed likely KV cache by 90%, while the proposed response is to build more memory fabs as bottlenecks attract solutions.The unresolved intelligence debate matters for AI economics, with agentic systems and self-training potentially favoring the smartest model despite humanly invisible differences.

Jonathan Ross

Nvidia-Groq’s LPX rack is already in production for Q3 availability, splitting inference between LPU FFN layers and GPU attention layers to improve utilization, latency, and throughput per megawatt.Ross expects fast-tier pricing to remain super-linear because speed increases iteration capacity, while inference revenue scales with users; energy supply, networking, and whether customers can support ultra-tier token spending remain the watchpoints.

Harry Stebbings × Jonathan Ross

Jonathan Ross says OpenAI, Anthropic and every hyperscaler will build chips for control over their own destiny, while Nvidia could be worth $10 trillion in five years.Doubling inference compute could nearly double their revenue within one month as rate limits suppress engagement, making HBM, energy, Groq’s six-month supply chain and the United States’ 2-3 year away-game window key signals.

Harry Stebbings × Jonathan Ross

Groq’s differentiated inference wedge avoids scarce HBM, supporting claimed 5x lower cost and one-third the energy per token while Nvidia concentrates on training.Aramco-backed capex and $1.5B in revenue—not fundraising—support a target of at least half of global inference compute by end-2027, but a power bottleneck in three to four years and Nvidia’s uncertain trajectory remain risks.

Harry Stebbings × Jonathan Ross

DeepSeek’s real breakthrough was fully automated verifiable-reward RL, while its $6M training figure excludes substantial distillation or scraping costs.Ross argues LLMs are now commoditized with no switching cost, pushing OpenAI toward open source while inference—not training—could reach 95% of compute spending.That expands the Nvidia opportunity through Jevons paradox, but CCP data access and cloud-based GPU workarounds leave export controls unresolved.