AMD And OpenAI | The Brainstorm EP 104

Key Concepts

• AI Bubble: The current period of intense investment and speculation in artificial intelligence technologies.
• AGI (Artificial General Intelligence): AI that possesses human-like cognitive abilities across a wide range of tasks.
• Inference vs. Training: Two distinct phases in AI model development. Training involves processing large datasets to build a model, while inference involves using the trained model to generate outputs for end-users.
• Warrants: Financial instruments that give the holder the right, but not the obligation, to buy a company's stock at a predetermined price.
• Coopetition: A business strategy where companies cooperate in some areas while competing in others.
• Agentic Commerce: The use of AI agents to automate and facilitate online purchasing decisions and transactions.
• ACP (Agentic Commerce Protocol): An open-source protocol for agentic commerce.
• MCP (Messaging Communication Protocol): An API pioneered by Anthropic for AI communication.
• Take Rate: The percentage of a transaction's value that a platform or service provider keeps as revenue.
• Sora 2: OpenAI's latest text-to-video generation model.
• Cameo: A feature within Sora 2 that allows users to insert their likeness into AI-generated videos.
• Uncanny Valley: A phenomenon where AI-generated content that is almost, but not perfectly, realistic can evoke feelings of unease or revulsion.
• PolyMarket: A decentralized prediction market platform.

AMD, OpenAI, and Chip Supply Dynamics

The discussion begins with the recent deal between AMD and OpenAI, drawing parallels to Nvidia's previous arrangements. OpenAI is investing in AMD through warrants and committing to purchasing "billions, tens of billions" of AMD chips. This deal has already boosted AMD's stock by "tens of percent." The warrants grant OpenAI a potential "10% stake" in AMD, contingent on the deployment of AMD chips. These chips are seen as potentially more competitive with Nvidia for inference tasks, which involve serving AI models to end-users (e.g., ChatGPT responding to queries). Historically, OpenAI utilized the same chips for both training and inference, a strategy now being emulated by others. The increasing complexity of training, where models "talk to themselves," blurs the lines between training and inference, making chips with more memory, like those from AMD, attractive for inference. This deal also diversifies OpenAI's supplier base, reducing its reliance on Nvidia.

The conversation then shifts to the broader financial implications. The sequence of funding is seen as moving from cash flows (social media giants) to equity (warrants in AMD) and potentially towards debt markets to finance the massive infrastructure build-out. The current landscape is characterized by coopetition, with companies like Intel also being drawn into the mix. This trend is partly attributed to a perceived hands-off approach from regulatory bodies like the FTC, which may be prioritizing strategic alignment against China over traditional antitrust concerns. This "open season" for collaborations could accelerate capital flow but also carries the risk of overextension.

GPU Demand and the "Decade-Long Shortage" Narrative

The AMD-OpenAI deal highlights the immense demand for GPUs. Headlines proclaiming a "decade-long shortage of compute" are analyzed. The argument is made that such pronouncements might mistake duration for magnitude. While a shortage may be significant, market forces should theoretically motivate capital investment, shortening the duration of the shortage. Historical examples of prolonged shortages are cited, including those in non-functioning economies (Maoist China, Cuba), during wartime, and in sectors with long lead times like energy infrastructure.

However, the current AI chip shortage is framed differently. The speakers project a need for "one and a half to two trillion dollars in chip spend in 2030" annually, across the industry. This forecast is driven by deals like OpenAI's implied "half trillion dollars of spend" with Nvidia and the AMD deal, which implies "hundreds of billions of dollars of potential spend." The immediate scramble for chips is evident, with OpenAI reportedly facing hardware development challenges and insufficient GPUs to service potential product launches. The unmet demand for GPUs is described as substantial, both currently and for future, more powerful models.

Monetizing AI: Agentic Commerce and ROI on the Path to AGI

The discussion pivots to how companies like OpenAI are generating revenue to fund their ambitious AI development, particularly on the path to AGI. The first key development is Agentic Commerce. OpenAI, in partnership with Shopify and Etsy, is enabling direct integration for instant checkout. They have also open-sourced the ACP (Agentic Commerce Protocol), which can work in conjunction with other APIs like MCP (Messaging Communication Protocol).

This move aims to generate revenue not just through subscriptions but also through a take rate on goods sold via ChatGPT or other interfaces. The potential market for agentic commerce is estimated at "$9 trillion," representing "25% of total online spend." With a conservative "two and a half percent take rate," this could translate to "$220 billion" in revenue. The speakers note that this take rate could be higher, especially considering additional fees like credit card processing. The historical precedent of Groupon offering significant discounts to acquire customers is mentioned as an example of businesses prioritizing customer acquisition.

The impact of agentic commerce on major e-commerce players like Amazon and Walmart is debated. One perspective is that these marketplaces are structurally advantaged due to their existing infrastructure, inventory depth, pricing advantages, and speed of delivery. These three pillars – pricing, selection, and speed – are expected to remain crucial for consumers.

However, the emergence of AI agents as a new abstraction layer that "compresses the funnel all the way down to purchase" could also empower long-tail retailers and direct-to-consumer brands. While Amazon's advertising revenue is a significant factor, the new paradigm might shift advertising spend to platforms like OpenAI. The argument is made that consumers may start their search elsewhere, leading to increased competition for Amazon. The analogy of wholesale/warehouse stores gradually taking share from traditional grocery stores is used to suggest that major players like Walmart, Amazon, and Costco might continue to gain share, but the overall market becomes more competitive at the margins. The concept of steepening power law curves is introduced, suggesting that the gap between top players and the rest will widen, with returns to logistics and non-front-of-store operations increasing.

Sora 2 and the Future of Video Content

The second major announcement discussed is Sora 2, OpenAI's advanced text-to-video model. Improvements in understanding "embedded physics" are highlighted, making the generated videos feel more lifelike and closer to crossing the uncanny valley. The potential for Sora 2 to birth a "new genre, a new medium of content" is explored.

OpenAI's strategy of launching a separate application for Sora 2, akin to Instagram Reels or TikTok, is analyzed. This approach integrates creation tools directly with distribution. The Cameo feature, allowing users to insert their likeness into videos, is seen as a significant innovation, enabling personalized, highly produced content that was previously only accessible to large creators.

The debate centers on whether this will be a standalone application or integrated into existing platforms. The cost of generating these videos is estimated at "$40 cents an hour," with a need for each video to be watched by "20 people" for economic viability. This suggests a focus on "in jokes between friends" and a potential competition with messaging apps like WhatsApp rather than broad broadcast platforms like Instagram.

However, another perspective suggests that video, being the most immersive content format, will eventually dominate. As the cost of creation decreases, text-based platforms are predicted to become obsolete. The potential for Sora 2 to become a niche genre for news broadcasting or personalized content creation is also considered. The speakers acknowledge the profound social and political implications beyond app competition.

Market Predictions: OpenAI Browser Launch

The episode concludes with a PolyMarket prediction on whether OpenAI will launch a standalone web browser by October 31st. The market currently shows a "52% chance," a significant increase from previous trading. The speakers offer their predictions:

• Sam: Seller (No)
• Nick: Seller (No)
• Brett: Buyer (Yes)

Brett's reasoning is based on OpenAI's rapid pace of announcements and their strategy of launching new products, fitting with the Sora 2 rollout. The market resolution criteria include a publicly released, standalone web browser available to the public in at least one country.

Synthesis and Conclusion

The discussion highlights the accelerating pace of AI development and its profound impact on various industries. The AI bubble is characterized by massive capital investment, driven by the promise of AGI and the immediate need for compute power. The AMD-OpenAI deal exemplifies the complex interplay of coopetition and strategic partnerships in securing hardware resources. The immense demand for GPUs is projected to continue, with forecasts suggesting trillions of dollars in annual chip spend by 2030.

On the monetization front, Agentic Commerce presents a significant revenue opportunity through take rates on transactions, potentially reshaping e-commerce and challenging established players. While large marketplaces may retain structural advantages, the competitive landscape is expected to become more dynamic.

The emergence of advanced generative AI models like Sora 2 signals a new era of content creation, with the potential to redefine media and communication. The strategy of launching standalone applications versus embedding features into existing platforms remains a key question, with implications for user engagement and market dynamics.

Finally, the episode touches upon the inherent uncertainty and rapid evolution of the AI space, as evidenced by the short-term market prediction regarding OpenAI's browser launch. The overarching sentiment is that while the path to AGI is fraught with challenges and requires continuous innovation and investment, the potential rewards are immense, leading to a period of intense activity and transformation.