> ## Content Index
> Fetch the complete content index at: https://wire.fourthweb.ai/llms.txt
> Use this file to discover other available public pages before exploring further.

# US Tried to Crush China's AI Dreams and Failed
- URL: https://wire.fourthweb.ai/us-tried-to-crush-chinas-ai-dreams-and-failed/
- Published: 2026-10-15T08:41:19.000Z
- Updated: 2026-10-15T10:25:54.000Z
- Description: The containment strategy won't work, and the acceleration strategy can't win. NYU professor Gary Marcus says the US-China AI race will end in a tie, not a decisive victory for either superpower
- Author: Travis Wright
- Tags: AI Agent Economy, AI Infrastructure, Compute Wars, DeFi, OpenAI, Anthropic, Google AI, Nvidia, Funding Rounds

**The containment strategy won't work, and the acceleration strategy can't win.**

### The Summary

- [NYU professor Gary Marcus says the US-China AI race will end in a tie](https://www.bloomberg.com/news/articles/2026-10-15/us-china-locked-in-unwinnable-war-on-ai-nyu-s-marcus-says?ref=wire.fourthweb.ai), not a decisive victory for either superpower
- Marcus is focusing on [building trustworthy AI](https://www.bloomberg.com/news/videos/2026-10-15/nyu-s-gary-marcus-on-building-ai-we-can-trust-video?ref=wire.fourthweb.ai) rather than just fastest AI, suggesting the current race metrics miss the point
- Implication: pouring billions into an arms race neither side can win means misallocated capital and delayed progress on AI that actually works

### The Signal

Gary Marcus, the NYU emeritus professor who's been the industry's most persistent skeptic of large language model hype, just told the Bloomberg New Economy Forum in New Delhi what no one in Washington or Beijing wants to hear: [this AI race is unwinnable](https://www.bloomberg.com/news/articles/2026-10-15/us-china-locked-in-unwinnable-war-on-ai-nyu-s-marcus-says?ref=wire.fourthweb.ai). Not because one side will fall behind, but because both sides are racing toward a finish line that doesn't exist.

The US and China have collectively committed hundreds of billions to AI development, with export controls, chip restrictions, and talent poaching defining the competition. But Marcus argues the framing is wrong from the start. You can't "win" a technology race when the technology itself hasn't solved its core reliability problems.

> "The question isn't who builds AI first. It's who builds AI we can actually trust."

Marcus has spent the past few years arguing that current AI systems are sophisticated pattern matchers, not reasoning engines. They hallucinate. They can't reliably fact-check themselves. They break in novel situations. [His focus on trustworthy AI](https://www.bloomberg.com/news/videos/2026-10-15/nyu-s-gary-marcus-on-building-ai-we-can-trust-video?ref=wire.fourthweb.ai) suggests he thinks both superpowers are optimizing for speed and scale while ignoring the fundamental architecture problems that make today's models unreliable for high-stakes decisions.

This matters because the race dynamics are driving massive resource allocation. TSMC is building new fabs. NVIDIA can't make enough H100s. Startups are raising at billion-dollar valuations to buy [compute](https://wire.fourthweb.ai/tag/ai-infrastructure/). And Marcus is saying all of that might be pointed at the wrong target. If neither architecture actually gets to trustworthy general intelligence, then the winner of the compute race just has the biggest pile of unreliable software.

The tie scenario Marcus predicts isn't a stalemate. It's something messier:

- Both countries develop powerful but fundamentally limited AI systems
- Neither achieves the decisive "AGI breakthrough" that justifies the investment
- The real winner might be whoever figures out the next architecture first, not whoever scaled the current one biggest

### The Implication

If Marcus is right, the strategic play isn't maximum acceleration. It's placing smaller bets on alternative approaches while everyone else is buying more GPUs. For companies, that means questioning whether you need foundation model scale or whether narrow, reliable AI solves your actual problem. For governments, it means rethinking whether chip export controls matter if the chips are being used to scale an architecture that hits a ceiling anyway.

Watch for signs that the major labs are hedging. If [OpenAI](https://wire.fourthweb.ai/tag/openai/), [Anthropic](https://wire.fourthweb.ai/tag/anthropic/), DeepMind, or Baidu start funding research that looks nothing like transformer scaling, that's the signal that insiders agree with Marcus. The race might not have a winner. But it definitely has an exit.

### Sources

[Bloomberg Tech](https://www.bloomberg.com/news/articles/2026-10-15/us-china-locked-in-unwinnable-war-on-ai-nyu-s-marcus-says?ref=wire.fourthweb.ai)