New York State just proposed a 15% profit-sharing on AI data centers. That's not a tax. It's a ledger entry. The ledger keeps score. And the score says Big Tech's energy appetite is a liability they've been hiding in plain sight.
Gas fees don't lie. People do. But here, the gas isn't Ethereum—it's the megawatts that power a single GPT-4 training run. A run that cost roughly $100 million in electricity alone. That's 100,000 MWh. Enough to power a small city for a month. And the state regulator just decided to audit that bill.
I've been watching this regulatory wave from my Prague apartment. The same apartment where I once watched a Uniswap flash loan attack unfold in real-time, analyzing failed transactions as the gas pool clogged. The same apartment where I audited the Terra oracle code and predicted the 90% depeg. Now I'm auditing a different kind of code: the energy contracts between states and hyperscalers.
Context: The Hype Cycle Meets the Grid
AI data centers are the new crypto mines. They consume massive amounts of energy, produce heat, and generate speculative value. The difference? Crypto miners were transparent about their energy use—every hash is publicly verifiable on-chain. AI training runs are opaque. No one knows how much electricity Google's TPU v4 clusters actually draw. The data is proprietary. The cost is hidden.
States like New York, Virginia, and Arizona are waking up. They see the power purchase agreements (PPAs) that Big Tech signed with utilities. They see the grid strain. They see the local residents paying higher rates to subsidize the infrastructure for data centers. The revolt isn't about climate change—it's about cost transparency. The old model: Big Tech builds a data center, promises jobs, and gets tax breaks. The new model: states demand a cut of the revenue.
This is a direct parallel to the crypto mining tax debates. In 2022, New York passed a moratorium on proof-of-work mining. The argument was environmental. But the real motive was energy accounting. The state wanted to know who was consuming the power and at what price. Now they're applying the same logic to AI.
But here's the twist: AI data centers are not crypto mines. They don't have a public ledger. They don't have a transparent consensus mechanism. They are black boxes. And regulators hate black boxes.
Core: The Systematic Teardown of Profit-Sharing Proposals
Let's dissect the mechanics. A profit-sharing proposal on AI data centers typically means the state takes a percentage of the revenue generated by the compute sold to customers. For example, if a data center charges $500 per hour for an H100 GPU cluster, and runs at 80% utilization, the annual revenue per server is $3.5 million. A 15% tax becomes $525,000 per server. Multiply by 10,000 servers—that's $5.25 billion in potential state revenue.
But the numbers don't add up. The cost of energy for that cluster is roughly $1.2 million per year (assuming 10kW per GPU, 1,000 servers, $0.10/kWh). The profit margin before energy is ~$2.3 million. After energy, it's $1.1 million. A 15% profit-sharing on that $1.1 million is only $165,000 per server. That's a far cry from $525,000. The state's projection is based on gross revenue, not net profit. That's a fiction.
Code is truth. Intent is fiction. The state's intent is to capture a share of the AI boom. But the code—the actual energy contracts, the depreciation schedules, the tax loopholes—will reduce that share to a trickle. I've seen this before. In 2021, I analyzed the tax incentives for Bitcoin miners in Texas. The state offered a 10% property tax abatement. The actual savings were 2% after legal fees and compliance costs. The same pattern will repeat here.
Minted nothing, promised everything. The state promises to fund schools with AI data center profits. But the reality is that the profits are already allocated to debt payments, equipment leases, and energy contracts. The state will get a check, but it will be small. The real winners are the utilities, who get guaranteed revenue from the data centers, and the hyperscalers, who pass the tax to their customers.
Let me give you a concrete example from my own audit work. I looked at a proposed data center in Loudoun County, Virginia. The developer claimed a $2 billion investment. I analyzed the PPA. The energy cost was 12 cents per kWh, but the actual delivered cost was 18 cents after transmission fees. The developer's profit margin was 8%. The county's proposed 5% profit-sharing would yield $80 million over 20 years. That's $4 million per year. For a county with a $2.5 billion annual budget, that's noise.
The mechanical cruelty of the system is that the state's regulatory apparatus is designed for a previous era. It assumes a linear relationship between energy consumption and profit. But AI data centers are not linear. They are variable. They can throttle compute, move workloads to other regions, or use on-site renewable generation to avoid the tax. The state's proposal is a static tax on a dynamic system. It will fail.
Contrarian: What the Bulls Got Right
But the bulls have a point. The energy appetite of AI is real. A single ChatGPT query costs 10 times more energy than a Google search. By 2027, AI data centers could consume 10% of global electricity. That's a system strain. The state's profit-sharing is a blunt instrument, but it's a signal. It says: we see the cost, and we want accountability.
The bulls are right that the current regulatory vacuum is unsustainable. Without some form of cost transparency, data centers will continue to externalize their energy impact. Local communities will pay higher rates, grid infrastructure will degrade, and the public will revolt. The profit-sharing proposal, even if flawed, forces a conversation about who benefits and who pays.
I've interviewed regulators in Prague. They told me off the record: "We don't understand the technology. But we understand the power bill." That's honest. The state's approach is not sophisticated. It's a crude accounting measure. But it's better than nothing. The alternative is to let the data centers run wild, like the early days of crypto mining in upstate New York—where a single mining farm consumed 15% of the town's electricity and paid nothing extra.
Another blind spot I missed: the profit-sharing could actually incentivize data centers to be more energy-efficient. If the state takes a cut of revenue, the operator has an incentive to reduce energy costs to increase net profit. That could lead to better cooling systems, more efficient chips, and load shifting to off-peak hours. The tax becomes a nudge. The mechanical cruelty of the proposal might accidentally produce a positive outcome.
But don't expect transparency. The ledger remains hidden. The state will audit the data center's books, but the data center will cook the books. They'll create subsidiaries in Delaware, lease the equipment from a Cayman Islands entity, and sell the compute to a shell company in Singapore. The profit will disappear. The state will get a fraction. The cycle continues.
Takeaway: The Energy Ledger Is the Next On-Chain Frontier
Forward-looking thought: The state's fight with Big Tech over data center energy is the precursor to a larger movement. Energy accounting will become the new compliance standard. Just as crypto forced transparency in financial transactions, AI will force transparency in energy consumption. The state will demand an immutable record of every megawatt consumed. The data center will need to prove its energy source. That's a blockchain use case.
I've already seen startups building on-chain energy certificates. Their pitch: "Prove your green credentials with a public ledger." The state's profit-sharing proposal will accelerate this. Data centers will need to track their energy usage in real-time, verifiable by third parties. The gas fees won't be on Ethereum—they'll be on the grid. And the ledger will keep score.
Code is truth. Intent is fiction. The state's intent is to tax AI. But the code—the actual energy consumption, the profit margins, the legal structures—will determine the outcome. I've been watching this ledger for 15 years. The pattern is always the same. The regulator arrives late, proposes a blunt instrument, and the market adapts. The profits are hidden. The costs are externalized. The public pays.
But this time, the energy is too big to hide. The sheer scale of AI data centers makes them visible. The state's profit-sharing is a first step. It's not a solution. It's a signal. And signals, like gas fees, don't lie.
— Oliver Lee, Prague, 2025