Dylan Patel: $300B OpenAI deal, Codex vs Claude Code, Gemini 3.0

The AI industry is marked by intense competition between OpenAI and Anthropic, significant infrastructure deals like OpenAI's with Oracle, and a future focused on enhancing AI capabilities through better tool use, extended context, and multimodal data.

• OpenAI's massive Oracle compute deal signals a strategic move for independence and scale.

• Anthropic's focused approach to AI for coding and productivity is driving its rapid growth.

• Future AI breakthroughs hinge on improved tool use, expanded context, and leveraging multimodal data, especially on mobile.

The AI landscape is fiercely competitive, with Anthropic's rapid revenue growth potentially surpassing OpenAI's by 2027, driven by its focus on AI for coding and productive tasks. OpenAI, despite its early lead, faces challenges from its less unified vision and Microsoft's reluctance to fund its massive compute needs, which led to a monumental $300 billion deal with Oracle. Future AI gains are expected less from core model improvements alone and more from advanced tool use, expanded context, and leveraging multimodal data, areas where Google with YouTube's data holds a significant advantage.

OpenAI's Oracle Deal

• 00:01:18 OpenAI's $300 billion deal with Oracle is unprecedented, representing the largest contract in stock market history and securing immense compute power for OpenAI outside of competing hyperscalers. This agreement enabled Oracle to issue a four-year revenue guidance, a highly unusual move, causing its stock to surge by 40%. The deal positions OpenAI with significant computational resources and independence from rivals like Microsoft, which hesitated to fully commit due to the financial scale required.

Anthropic's Growth & Focus

• 00:05:13 Anthropic's revenue growth rate indicates it could surpass OpenAI by 2027, despite OpenAI's head start, primarily because its models excel in critical areas like AI for coding, computer use, and productive tasks. The company's success is attributed to its highly unified mission, shared by its technical co-founders who left OpenAI for a clearer vision focused on solving software engineering and computer use to accelerate AGI development. Unlike OpenAI, which has broader goals, Anthropic is singularly concentrated on making AI capable of self-improvement through coding.

AI Coding Wars

• 00:08:55 OpenAI is actively striving to compete in the lucrative software engineering market, with GBD5 focusing heavily on improving coding capabilities and promoting its Codex CLI and extension. While some believe Codex has significantly improved, becoming more powerful for complex debugging than Claude Code, the latter is still preferred by many for its ease of use and ability to explain bugs simply. The ideal approach for engineers might be to use both, leveraging Claude Code for quick UI changes and Codex with GBD5 high for deeper, more convoluted issues, recognizing their distinct strengths in reasoning time and clarity.

Future AI Gains

• 00:18:15 Significant advancements in AI productivity will come from enhanced tool use, extended context windows, and improved computer interaction capabilities, rather than solely from core model improvements. Models need to be able to actively search for and integrate context from various sources like emails, DMs, and documentation, as users often provide insufficient prompts. This enables AI to understand problems deeper and perform tasks more autonomously, effectively bridging the gap between current limited user input and the full potential of AI.

Mobile AI & Multimodality

• 00:31:32 While advanced computer use for AI is becoming more feasible on desktops, implementing similar comprehensive context-gathering and app interaction on mobile devices presents a significant challenge due to walled garden ecosystems (Apple) and the lack of incentive for companies to share proprietary data. Google, with its vast multimodal data from YouTube and its customizable Android platform, holds a unique advantage in developing advanced multimodal AI capable of processing various data types beyond just text. The ability to properly utilize and filter massive amounts of video data is crucial for future AI gains.