Arvind Srinivas

NVIDIA will continue to do great because server-side compute remains essential for frontier intelligence. The shift to hybrid orchestration (local + cloud) does not eliminate the need for NVIDIA GPUs in data centers. AI inference will still run on servers alongside local compute, and NVIDIA's position is strong.

He argues that data-center buildouts alone cannot support billions of always-on AI agents because power permits and HBM/memory are bottlenecks; scaling inference will require tapping local compute and a lot more memory, tightening the AI semiconductor/memory supply chain.

"Amazon has been trying to block your Comet AI browser. They won in court to effectively prevent your service from what they think is scraping their service." AI aggregators like Perplexity threaten traditional search and discovery platforms by bypassing front pages and directly extracting answers. Amazon's aggressive legal defense to block AI scrapers is crucial to protecting its highly lucrative retail media and advertising business. If AI agents can shop on a user's behalf without viewing Amazon's ads, Amazon's ad revenue takes a direct hit. Monitoring this legal precedent is vital for valuing companies reliant on walled-garden data. WATCH Amazon's legal outcomes regarding AI scraping; a definitive victory solidifies its data moat, while a loss could open the floodgates to margin-eroding AI intermediaries. If higher courts rule that AI scraping falls under fair use or benefits the consumer without violating terms, Amazon could lose control over its top-of-funnel product discovery.

"Anthropic is very good at coding. OpenAI's models are very good at writing. Google's models are very good at multimodal." The market previously feared that foundational AI models would become a commoditized race to the bottom. However, because models are developing distinct, specialized strengths, enterprise customers will need to utilize all of them rather than picking just one. This ensures that the major tech giants (Google, Microsoft via OpenAI, Amazon via Anthropic) will share an expanding oligopoly with strong pricing power. Furthermore, running these multi-model digital workers in cloud sandboxes will drive massive, continuous compute demand for their cloud divisions (GCP, Azure, AWS). LONG the mega-cap AI hyperscalers as they benefit from specialized model monetization and increased cloud infrastructure usage. Open-source models could eventually catch up across all modalities, forcing the commoditization that is currently being avoided.

"Anthropic is very good at coding. OpenAI's models are very good at writing. Google's models are very good at multimodal." The market previously feared that foundational AI models would become a commoditized race to the bottom. However, because models are developing distinct, specialized strengths, enterprise customers will need to utilize all of them rather than picking just one. This ensures that the major tech giants (Google, Microsoft via OpenAI, Amazon via Anthropic) will share an expanding oligopoly with strong pricing power. Furthermore, running these multi-model digital workers in cloud sandboxes will drive massive, continuous compute demand for their cloud divisions (GCP, Azure, AWS). LONG the mega-cap AI hyperscalers as they benefit from specialized model monetization and increased cloud infrastructure usage. Open-source models could eventually catch up across all modalities, forcing the commoditization that is currently being avoided.