AI Enters Commercial Era, Says Kindred’s Jang

Key Concepts

• Compute Demand: The rapidly increasing need for computational power for both training and inference of AI models.
• Model Competition: The global race to develop specialized AI models in various domains (video, audio, physics, medical science, etc.).
• Commercial Application Phase: The shift from experimental AI testing to real-world, revenue-generating applications.
• Generative Media: AI-powered tools that create content, particularly video, which are now generating significant revenue.
• Economic Productivity: The ability of AI applications to generate revenue and value, justifying investment in infrastructure.
• Middleware: Software infrastructure layer that enables value capture and differentiation in the AI ecosystem.
• Sovereign AI: The concept of individual nations developing and controlling their own AI capabilities.
• Stablecoins: Cryptocurrencies designed to maintain a stable value, potentially transforming global payments infrastructure.

AI Compute Demand and Model Competition

The speaker emphasizes the underestimation of compute needs for both training and inference in AI. This demand is scaling across all stages of companies, from early-stage startups to public companies.

• Global Competition: Every region is now participating in the AI model competition race.
• Specialized Models: There's a proliferation of specialized models in areas like video, audio, speech, physics, and medical science. Examples include DeepMind and startups like Cell phones in medical science.
• Shift to Commercial Application: AI is transitioning from an experimental phase to commercial applications.
• Consumer Adoption: Double-digit adoption of AI tools by creators and users on platforms like TikTok, Instagram, and YouTube.
• Enterprise Adoption: Businesses are using AI for voice-based call centers, report writing, and search/answers.
• Inference Demand: Inference demand has already surpassed projections from a year ago by ten times.

Addressing Bubble Concerns and Ensuring Productivity

The conversation addresses concerns about a potential "house of cards" scenario where investments are concentrated in a few players without corresponding productivity gains.

• Importance of Revenue and Adoption: The speaker stresses the importance of seeing real revenue, adoption, and usage to justify investments.
• Generative Video Example: Generative video is cited as an example of AI generating real revenue through video ads on platforms like YouTube and Comcast/Universal Channel. These ads are generating real CPMs for brands.
• Economic Productivity Justifies Investment: The scaling of investment in GPUs, data centers, and energy is justified by economic productivity on the demand side.

The Role of Bubbles and Capital Concentration

The discussion explores the role of bubbles and capital concentration in driving technological progress.

• Bubbles as Catalysts: Bubbles are seen as necessary for attracting talent, engineering expertise, research, and capital, leading to rapid growth.
• Real Revenue Differentiator: This tech cycle is different because there's real revenue and commercial application happening.
• Beyond GPUs: The AI ecosystem extends beyond GPUs to include energy, data space, and software infrastructure (middleware).
• Middleware Value Capture: Middleware is where value capture and differentiation are now occurring. Companies like Together, Fowl, and Generative Media Cloud Space are mentioned.

Crypto and Stablecoins

The conversation shifts to crypto, specifically stablecoins, and their potential impact on global payments.

• Fintech Transformation: Crypto, especially stablecoins, is slowly transforming fintech and global payments infrastructure.
• Underestimation of Potential: It's easy to underestimate the potential of stablecoins.
• Cross-Border Collaboration: Unlike AI, where there's a concept of sovereign AI, crypto sees more cross-border collaboration.
• Government Interest: The US government and other foreign governments are showing increasing interest in blockchains.
• Underlying Compute: Crypto relies on the same high-performance compute and data centers as AI.

Synthesis/Conclusion

The key takeaways are that the AI revolution is driven by rapidly increasing compute demand and the development of specialized models. The transition to commercial applications is underway, with generative media as a prime example of revenue generation. While concerns about a potential bubble exist, the presence of real revenue and economic productivity differentiates this tech cycle from previous ones. The AI ecosystem extends beyond hardware to include crucial software infrastructure. Finally, crypto, particularly stablecoins, has the potential to transform global payments, with both AI and crypto relying on the same underlying compute infrastructure.