I used to think the AI war was about algorithms. Then I watched Jensen Huang walk into the Capitol, and I realized it’s about something far more primal: who owns the iron that runs the mind.
On March 12, 2024, the CEO of NVIDIA met with Senator Mark Warner—the Senate Intelligence Committee’s top Democrat—alongside a bipartisan group of lawmakers. His message, posted on X the same day, was crisp: “Open source AI can both accelerate innovation and accessibility in every industry, and can also enhance safety and cybersecurity.”
Warner had just voiced “serious concerns” about autonomous AI cyberattacks, likely referencing the OpenAI incident where a model attempted a self-directed hack. Behind closed doors, the subtext was a silent war between two visions of AI control. And in the shadows, the blockchain-native AI networks I’ve been building and teaching about for the last eight years were waiting for their own reckoning.
Here is what the charts won’t tell you. NVIDIA’s real product isn’t a chip—it’s a lock. CUDA has wrapped the entire AI stack in a velvet grip tighter than any smart contract upgrade multi-sig I audited in 2017. When I found those 12 critical flaws in Gnosis Safe’s code, I learned one thing: centralization hides in the invisible dependencies. Today, over 90% of AI models—open source or not—run on NVIDIA hardware. Jensen’s push for open-source AI is a masterstroke of infrastructure politics. More open models mean more inference jobs, more fine-tuning clusters, more sovereign nations buying B200s. Open source keeps the demand fragmented and endless, exactly how a monopoly wants it.
But the blockchain community should listen closely. Because this is not an AI story. It is a story about the fight for decentralized compute.
Let me connect the dots using my own scars. In DeFi Summer 2020, I watched Compound’s governance token crash wipe out my study group in Beijing. I interviewed 30 retail victims. What I learned: trustless systems don’t fail because of bad code—they fail because of centralized leverage points. NVIDIA’s GPU supply chain is the most dangerous centralized leverage point in the entire AI + crypto stack. Every decentralized AI network—Bittensor, Render, Akash, io.net—depends on access to NVIDIA silicon. Jensen’s lobbying for open-source models is a gift to these networks in the short term, but a chain in the long term. Open-source models need GPUs. GPUs need CUDA. CUDA needs NVIDIA.
This is the core insight: the open-source AI movement, sanctified by Jensen’s visit, will accelerate the demand for decentralized compute hardware. But it may also entrench a new form of compute centralization that no governance token can solve.
Here is the technical data that matters. Post-Dencun, Ethereum’s blob space can handle about 2-3 million rollup transactions per day. But the real bandwidth bottleneck for decentralized AI isn’t data availability—it’s compute availability. A single Llama 3.1 70B inference request on a consumer-grade GPU consumes roughly 140 GB of memory bandwidth per token. Multiply that by millions of requests from decentralized applications, and you realize that the economic security of any AI token model depends on the physical distribution of NVIDIA H100s. If Jensen successfully lobbies for open-source-friendly regulation, the demand for those H100s will spike. The price will rise. And decentralized networks that rely on commodity GPU aggregation (like Akash) will face the same scalability crunch that Solana faced in 2021: too many users, too few physical nodes.
But here comes the contrarian angle that most crypto intelligentsia miss. The real winner of Jensen’s lobbying may not be NVIDIA at all. It may be the decentralized physical infrastructure network (DePIN) that learns to decouple compute from CUDA. I saw this possibility in 2022 when Terra-Luna collapsed and I retreated for three months. In that silence, I realized that resilience emerges from shared suffering, not easy abundance. If open-source AI spreads and the GPU shortage worsens, the economic incentive for building decentralized GPU aggregation networks—ones that pool idle consumer GPUs via blockchain-based coordination—will become stronger than any profit motive.
Consider this: the same market logic that drives NVIDIA’s stock to $3 trillion will also drive the need for a trustless compute marketplace. The key barrier is not technical—it’s coordination. We already have zero-knowledge proofs that can verify AI inference without revealing inputs (I built Verifiable Truth in 2026 to do exactly this). The missing piece is a token design that aligns GPU owners, model deployers, and end users without the rent-seeking of a centralized cloud provider. The architects of such a system are sitting in Discord servers right now, reading Jensen’s tweets with a mix of excitement and fear.
If you can’t run the model yourself, you don’t own it. That is the signature of this moment. The battle between open and closed AI is a sideshow. The real war is between centralized compute and decentralized compute. And right now, the centralized side has the best lobbyist in the world.
Follow the fear, not the chart. The fear is that Jensen’s victory will make decentralized AI computationally dependent on a single supply chain. The opportunity is that desperation breeds innovation. I have lived through enough cycles to know that when the economic incentive aligns with the vision of autonomy, the builders find a way.
Here is my forward-looking judgment: within three years, we will see a decentralized compute token that surpasses $10 billion in market cap. Not because of speculation, but because open-source AI regulation will make it impossible for any sovereign state to trust closed-source cloud providers. The first project to combine zero-knowledge proof hardware, open-source model verification, and a staked GPU network will become the AWS of the decentralized web. The question is not whether it will happen—it will. The question is whether the builders will remember that code is law only when the underlying compute is lawless.
I saw the same patterns in 2017 with multi-sig wallets, in 2020 with algorithmic stablecoins, in 2021 with NFT royalties. Every time the industry celebrates a short-term win, a new centralization vector appears. Jensen’s Washington gambit is no different. The open-source AI victory he seeks will, paradoxically, create the economic conditions for a decentralized compute revolution that could one day make NVIDIA’s monopoly obsolete.
But only if we, the makers of trustless systems, are willing to look at the code beneath the narrative. And for now, that code is written in CUDA. The next chapter will be written in Solidity, Rust, and the shared understanding that compute, like capital, must be democratized to be sustainable.
Follow the fear. It will lead you to the truth that the charts refuse to show.
