Andrej Karpathy, the former Tesla AI lead and early OpenAI cofounder, is pushing the AI industry toward a fundamentally different power structure. Rather than letting a handful of companies control cutting-edge artificial intelligence, Karpathy envisions a future where capable AI models become shared infrastructure that the entire industry can access and build upon. His perspective adds crucial weight to an emerging debate about who should control AI's future as companies race toward profitability and public markets.

What Does Karpathy Actually Mean by "Common Working Space" for AI?

Karpathy's proposal centers on making mature, capable AI models available as a shared foundation rather than proprietary assets locked behind corporate walls. In a recent conversation with Sarah Guo, he explained his reasoning with historical perspective.

"Centralization has a very poor track record in the past, in my view. There are a lot of pretty bad precedents in economic and political systems. So I want there to be a thing that's maybe not at the edge of capability because it's new and unexplored. But I want there to be a thing that's behind and is a common working space for intelligences that the entire industry has access to. That seems to me like a pretty decent power balance for the industry," Karpathy stated.

Andrej Karpathy, Former Tesla AI Lead and Early OpenAI Cofounder

This distinction matters. Karpathy isn't arguing that companies should open-source their latest, most advanced models. Instead, he's proposing that slightly older, proven models should become industry commons. Think of it like how the internet itself became shared infrastructure, or how Linux became the foundation that countless companies build upon. The idea is to create a baseline that prevents any single company from becoming the sole gatekeeper of AI capabilities.

How Does Karpathy's Approach Differ from Sam Altman's Democratization Push?

OpenAI CEO Sam Altman recently published a manifesto calling for AI democratization, arguing that "power in the future can either be held by a small handful of companies using and controlling superintelligence, or it can be held in a decentralized way by people." While Altman and Karpathy share the same concern about power concentration, they're approaching the solution from different angles.

Altman's focus is broader, encompassing policy, governance, and ensuring the public and governments have a voice in AI development. Karpathy, by contrast, is talking less about policy and more about the practical mechanics of open-source AI. He's focused on putting the technical benefits of capable models into the hands of developers and organizations across the industry, which would accomplish the democratization mission through infrastructure rather than regulation.

Both perspectives are complementary. Altman is setting the philosophical and policy framework; Karpathy is describing the technical architecture that would make that vision real. OpenAI President Greg Brockman reinforced this sentiment, stating that "we need this broad conversation. We need lots of people to be aware that if this technology is going to come and change everything for everyone, people need to participate in that. It can't be something that's done off in secret by one centralized group".

Why This Matters Now, Before These Companies Go Public

The timing of this conversation is critical. Right now, while OpenAI, Anthropic, and other frontier labs are still private, they can theoretically choose to prioritize decentralization over maximum profit extraction. But once these companies go public, the financial incentives shift dramatically. Shareholders will demand that companies maximize returns, which typically means consolidating competitive advantages and maintaining proprietary control.

Karpathy's proposal essentially asks: what if we built the power-balancing mechanisms into the industry's infrastructure before market forces push everyone toward centralization? The challenge is that all of this remains theoretical. OpenAI has published principles and policy documents, but it's unclear how these companies will actually transform their operations to live up to these ideals.

Steps Toward Building Decentralized AI Infrastructure

• Establish Shared Model Commons: Create industry-wide access to capable but not cutting-edge AI models that serve as a foundation for smaller companies and researchers, preventing any single entity from controlling baseline capabilities.
• Implement Transparent Governance: Develop clear policies about which models become shared infrastructure, when they transition from proprietary to commons, and who makes those decisions, with input from policymakers and the broader industry.
• Support Open-Source Development: Invest in and promote open-source AI projects that allow developers worldwide to build, customize, and deploy AI systems without dependence on proprietary platforms controlled by a few companies.

The real test, according to industry observers, is whether OpenAI and other frontier labs will actually resist the consolidation pressures that come with going public. Altman has stated that OpenAI "will resist the potential of this technology to consolidate power in the hands of the few," but holding companies to that commitment will require sustained pressure from the industry, policymakers, and the public.

Karpathy's vision represents a pragmatic middle ground. He's not asking companies to give away their most advanced work or abandon competitive advantage entirely. Instead, he's proposing that the industry adopt a tiered approach where mature, proven models become shared infrastructure while frontier labs continue pushing the boundaries of what's possible. This could create what he calls "a pretty decent power balance for the industry" without requiring companies to sacrifice innovation or profitability.

Whether this vision becomes reality depends on whether the leaders of Anthropic, Google, Meta, xAI, Microsoft, Amazon, and other frontier labs embrace similar principles before their financial incentives shift. The conversation is happening now, but the hard part is just beginning.