Investors should monitor the approaching performance ceiling of traditional Transformer-based AI architectures, as pure scaling is hitting real-world adaptability walls. Watch for hardware infrastructure leaders like Nvidia (NVDA) that benefit directly from ongoing demand for specialized GPU kernels and custom chip efficiency. Keep an eye on private funding rounds for stealth AI startups like Core Automation, founded by former OpenAI and Google Brain executives, for future market-shaping opportunities. Investors should prioritize hardware and infrastructure plays that solve computational depth bottlenecks rather than standard software applications. Expect major architectural shifts toward continuous learning and reinforcement learning integration over the next 12 to 24 months.

Core Automation (Private Company)

• Founded by Jerry Tworek (former VP of OpenAI, leader of Strawberry and Reasoning teams) and Rohan Anil (former pre-training lead at Gemini, Google Brain, and Anthropic).
• Focuses on developing a transformer replacement architecture capable of true test-time learning and continuous adaptation.
• Aims to build the most automated AI research lab in the world to rapidly iterate on deep learning architectures, optimizers, and custom GPU kernels (such as Blackwell chips).
• Addresses current industry bottlenecks by combining pre-training and reinforcement learning (RL) end-to-end to drastically improve computational depth and efficiency.
• Currently operates as a private early-stage startup backed by Sequoia Capital and is not publicly traded.

Takeaways

• Investors tracking the AI sector should watch for the limitations of current Transformer-based architectures, as leading researchers believe scaling them alone hits a wall for real-world adaptability and continuous learning.
• Infrastructure and hardware optimization remain major bottlenecks in AI, highlighting the ongoing importance of custom hardware efficiency, tensor processing, and specialized kernel generation.