The K-Shaped Divide: Why Societe Generale's AI Warning Is a Call for Decentralized Ownership
From the ashes of 2022, we planted seeds for 2030. But the soil is shifting. On August 12, 2025, Societe Generale released a report that sent ripples through the financial ecosystem: AI is accelerating a K-shaped economy, where the rich grow exponentially richer while the majority stagnate or fall. The report's core thesis—that AI rewards ownership of compute, models, data, and financial assets—isn't just another macroeconomic prediction. It's a fundamental critique of the current technological architecture. And for those of us in Web3, it's a call to action.
Let me unpack this. The K-shaped economy describes a divergence: one branch of the economy (the upper arm) benefits from AI-driven productivity gains, while the other (the lower arm) faces displacement and wage stagnation. Societe Generale, a European investment bank, isn't new to macro analysis. But their timing is telling. They've observed that the lion's share of AI's value creation is captured not by workers, but by asset owners—those who hold the GPU clusters, the training pipelines, the proprietary datasets, and the equity in AI giants like NVIDIA, Microsoft, and OpenAI.
As a Web3 community founder who has lived through the ICO idealism, the DeFi summer, and the bear market resilience, I see this pattern with painful clarity. In 2017, I wrote about Golem and Bitconnect as tools for social equity. Today, I watch the same concentration dynamics play out at a systemic level. The question is no longer whether AI will reshape the economy, but whether we can reshape the ownership structures of AI itself.
The report's strength lies in its clarity. It identifies four pillars of ownership: compute, models, data, and financial assets. Each is a lever of power. Compute is the most tangible—NVIDIA's H100 and B200 GPUs are the new oil, and their supply is tightly controlled by a handful of companies. Models, from GPT-4 to Claude, are trained on vast datasets that are themselves proprietary. Financial assets—the stocks and bonds of AI leaders—have surged, rewarding those who already hold capital. The result is a self-reinforcing cycle: the rich get richer because they own the means of production, and AI amplifies that ownership.
But the report also has blind spots. It neglects the role of open-source models like Llama, Qwen, and DeepSeek, which have begun to democratize AI capabilities. In 2025, we saw small teams fine-tune these models on modest hardware, achieving results that rival closed-source alternatives in specific domains. This is a counterforce—a potential antidote to the K-shaped divergence. Yet, the report's silence on this is telling. Open-source reduces the cost of using AI, but it doesn't inherently change ownership. The fine-tuning still requires data, and the deployment still requires cloud infrastructure. The open-source movement is a step, but not a revolution.
From my experience building 'Decentralized Hearts' in 2021, I learned that true empowerment requires more than access—it requires ownership. We taught 50 women to mint NFTs on Ethereum, but the real value creation was in the community itself, not the tokens. Similarly, AI's K-shaped effect can only be mitigated if we move beyond 'renting' AI capabilities to 'owning' the underlying assets. This is where Web3 enters the conversation. Tokenization of compute resources, data cooperatives, and decentralized AI model governance are not just experiments—they are necessary countermeasures.
Consider the compute layer. Projects like Akash Network and Render Network are already enabling peer-to-peer GPU sharing, but they lack the scale to compete with big cloud providers. However, the trend is clear: if we can tokenize compute capacity and allow anyone to contribute and earn, we can democratize the ownership of the most critical AI resource. Similarly, data DAOs could allow individuals to pool their data and receive royalties when it's used for training. This is not a silver bullet, but it's a direction.
Societe Generale's report also fails to address the policy dimension. The report hints at 'wealth concentration' without proposing solutions. As a financial analyst, I know that tax systems are designed for a pre-AI world. Labor income is heavily taxed, while capital gains are lightly taxed. AI will exacerbate this imbalance. A 'compute tax' or 'data dividend' could rebalance the scales, but global coordination is unlikely. The EU AI Act is a start, but it focuses on safety, not distribution.
Now, the contrarian angle: what if the K-shaped forecast is too pessimistic? The report assumes that AI's productivity gains will continue to flow to capital owners. But history shows that transformative technologies eventually diffuse. The internet, for example, initially concentrated wealth in a few ISP and platform companies, but over time, it enabled millions of small businesses and creators. AI could follow a similar trajectory. The marginal cost of AI inference is dropping rapidly—it's already 90% cheaper than two years ago. If this trend continues, AI could become a utility, like electricity, accessible to all. The K-shape might flatten as the lower arm gains access to powerful tools that enhance their productivity.
Yet, I remain skeptical. The internet's diffusion was aided by open standards and public infrastructure. AI, by contrast, is built on proprietary models and cloud lock-in. The open-source movement is fighting this, but it's an uphill battle. The real question is: will the decentralized alternatives scale fast enough to prevent a permanent K-shaped divide?
For the Web3 community, this is a moment of truth. We have the tools—smart contracts, tokenomics, DAOs—to create new ownership models. But we must act with urgency. The window for shaping AI's economic architecture is closing. In 2026, the first wave of AI-native regulations will solidify. If we don't embed decentralization into the core of AI infrastructure, the K-shaped economy will become a permanent feature.
As I write this, I'm reminded of a lesson from the 2022 bear market: resilience is the new utility. The projects that survived were those with strong communities and sustainable tokenomics. The same principle applies to AI. We need to build systems that are resilient to centralization, that reward contribution over capital, and that prioritize long-term social value over short-term returns.
From the ashes of 2022, we planted seeds for 2030. But those seeds must be planted in decentralized soil. Societe Generale's report is a warning, but also a gift. It forces us to confront the structural flaws in how we build AI. The next decade will define whether AI becomes a force for inclusive growth or a tool for entrenched inequality. The choice is ours, and the time to act is now.
Trust is built in the bear, sold in the bull. Visionaries plant trees they never sit under. Let's plant the seeds of decentralized AI ownership today.