Are NVIDIA, Anthropic, and Google Hiding the Real AI?

2026-05-17 · Source: AI Advances - Medium · Field: Technology & Digital — Artificial Intelligence & Machine Learning · Depth: Advanced, quick

Summary

The author argues that major AI companies like NVIDIA, Anthropic, and Google are pursuing a "taller ladder" approach with flat transformers, investing hundreds of billions of dollars annually in brute-force computation, which the underlying architecture cannot solve. This strategy is contrasted with a hidden, more efficient approach: Geometric AI, which the author claims would save trillions by providing a comprehensive solution rather than incremental improvements. The core issue, according to the author, is a fundamental geometrical problem within current AI, which cannot be resolved by simply scaling up GPU farms, even with massive data centers like the one reportedly planned for Patagonia. The author suggests that research teams within these companies are, in fact, exploring this alternative geometric approach.

Key takeaway

For research scientists evaluating future AI architectures, you should critically assess the long-term viability of scaling flat transformers. Consider exploring the foundational principles of Geometric AI, as it may offer a more efficient and scalable path forward, potentially saving significant computational resources compared to current brute-force methods.

Key insights

Current AI's fundamental geometric problem cannot be solved by scaling flat transformers; Geometric AI offers a more efficient solution.

Principles

• Brute-force scaling of flat transformers is unsustainable.
• Geometric AI offers a more efficient architectural paradigm.

Topics

• Geometric AI
• Flat Transformers
• AI Architecture
• Computational Scaling
• AI Investment

Best for: Research Scientist, AI Scientist, AI Architect, Director of AI/ML

Related on AIssential

• NVIDIA and OpenAI Are Scaling an AI That Cannot Subtract — Current AI's mathematical foundations are flawed, leading to basic arithmetic failures despite massive investment.
• The Sequence Knowledge #874: Transformers or Not? — Transformers excel due to scalability and general attention, but their computational cost suggests they are a foundational, not final, AI architecture.
• The Thinking Machine That Google and Anthropic will Ship Together — True AI cognition necessitates a new mathematical blueprint, moving beyond current LLM scaling and flat embeddings.
• Google hints at New Topological Flat Transformer — A new transformer architecture based on continuous flow dynamics promises enhanced reasoning and memory efficiency.
• From bigger models to better intelligence: what NeurIPS 2025 tells us about progress — AI progress is shifting from raw scale to efficiency, real-world evaluation, and interactive world modeling.
• I Patented a Four-Sided Box. It's the Best Mental Model I Have for Building Agents. — Effective AI agents depend more on problem representation than on model size, a lesson from a patented "four-sided box" method.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI Advances - Medium.