Seventy—eight zero. That's not a round number you announce for attention. That's a number you announce because you are playing a game where the stakes are measured in gigawatts, not in millions of dollars. I have seen this pattern before, in a different market, in a different form. It was the spring of 2017, and a young developer in a Parisian basement was showing me an ICO dashboard with a hundred thousand shills behind the code. The chart lied. The volume didn't speak—it screamed. We knew that if he was drawing in attention with public numbers, the real deal was in the private contracts.
The Crypt Brief report on Anthropic's 'Data Center Capacity Letters of Intent' has dropped, and the immediate takeaway is the same. This is not a news story about AI. This is a story about raw infrastructure saturation, about the physical world stretching to accommodate a volatility that has outgrown the cloud. I have to write this piece today, because the crypto press is going to repeat this as a bullish AI signal, but from my desk in Paris, watching the order flows, I recognize the pattern. It is not a signal of capability. It is a blueprint for exposure.
So, let's dissect the anatomy of this news. We are not talking about chips—we are not even talking about code. We're talking about the real estate of problems. Anthropic, the company behind a security-first architecture, is silently signing leases for the equivalent of a small national grid. That's not innovation. That's a liability. And I have a feeling the market is about to short the entire trend of 'infinite capacity' as 'infinite margin'.
The City of Paris has a saying, 'Il faut souffrir pour être belle.' It is the same for the AI giants. You must burn capital to be beautiful in the PNL.
We break down the known variables from the source materials:
First, the volume. Multiple sources state Anthropic has signed 'Letters of Intent' (LoIs) with multiple data center providers, for a pegged capacity concept. (Some trackers report the scope as 'up to 70-80%' of a larger power need, others as '70-80 separate LOIs'.) The number is not the news. The _surface area_ is. These are not purchase commitments. They are optionality. In high-frequency trading, you buy optionality on feast or famine. Here, Anthropic is buying a lock on real estate to prepare for a 10x surge in usage that hasn't happened yet.
Let's be very clear on the physics of this.
In the current energy market, a single hyper-scale gigawatt data center can require the output of a small nuclear reactor, or the entire reservoir capacity of a mid-sized European state. When you hear '70 gigawatts', you are not discussing servers. You are discussing the physical world—the diversity of geography for cooling, the rate of change in the climate for energy loss, and the potential for geopolitical public backlash.
The reason a tech company signs this many letters is hedging. They are securing 'first right of refusal' for locations where they will either build power purchase agreements for the aluminum and steel, or stride to the conductors.
I am not a banker, but I have traced the financial economics of my survival in 2022. The Terra collapse taught me that running a code without infrastructure is a mirror dance: you borrow one coin, lend another, and borrow the trust of the sophisticated. The data center ledger now is no different. When Anthropic signs a threshold for this many LoIs, they are signaling to their supply chain: 'Embrace me, or expire to a bottle.' They are placing a massive call option on the future of AI compute. And in that, there are three unspoken collateral implications.
The first implication is for the private investor in the region. Anyone holding a chipmaker like NVIDIA or a novel advanced packaging firm should check the inner tubes of their bet. This is the power and muscle phase of the market. GPU demand is not just about the unit, it is about the floor space and the electricity that house the units. Anthropic's presence unlocks a capex tailwind for the entire ecosystem. But there is a reason I see this with fractal clarity. On January 2024, when the Spot ETF approvals were leaked, I saw the careful delays in the SEC filing and front grid. It was the same muscle memory.
The second is for the utility and municipal regulator. When a load this size hints, the communication turns geopolitical. This level of power requirement will snap the planet's electricity tariffs into new territory in the next 18 months.
But here is the main split. The contrarian angle. This is my thesis.
Everyone will read this as 'Bullish AI'. But the actual model is Bullish Power, not Money.
In the time I spent as a digital asset editor, high-capacity compute provided the transactional means to create financial command. But the market is now squeezing: a traditional 'lease-back' structure between the same hyper-scaler provider—feeding the same assets—used to be a CAPEX game. Now, we are in a regime where the loan is on the PPA, not the GPU. The amount of energy’s connection—the ones that are actually COVID-quality, due to the old rules—is just in a hockey stick imprint.
This is what the Silicon Valley might not be saying. They are not scaling up to serve users. They are scaling up to serve the Frankenstein model itself. This is the notorious autopiloting 'algo-twitch'. If you look at the lens of labeling as a CNE (Compute-Overloaded-Electricity), then the whole company doesn't have a product-only future; the company is a mobile grid.
The chart lies. The volume speaks.
Let's speak about the volume of the buildable.
For the last three years, the Silicon Legion (all the big labs) has desperately bought traditional compute from AWS, but AWS itself is a wickedly powerful one-stop shop. The slow market of the cloud players is to treating the cloud pure as an infinite tertiary, but in this cycle, the field has finally met the wall of 'two strikes'. They have to go elsewhere, to the waste-/going to be there/nearly-exported and create work. So when Anthropic is signing 70+ data suite LoIs, what they are really saying is: 'Here's how we will offset the fast robot 'digital dividend' with physical noise.' They are creating a new class of labor for the infrastructure market. And it will have massive implications for the generalized market.
I have my own experience in the Paris Hackathon. I saw know a smart contract when I see one. A Letter of Intent resembles this exact state—commit the promise to draw intent. It doesn't have a lock on power. It's an option. And in the crypto bankruptcy contagion of '22, a slow mover with a de-levered LoI is the fastest path to insolvency. That will hinge on the timeline and the connectivity.
But I will draw my line in the sand: Anyone who opens this as a 'paper' without a 'market' has misinterpreted the teach.
Instead, the letter is to be read as a steel signal. It means the 'Time-To-Adoption' for the dense-nacelle will shrink. Because for now, we are filling machines with GPU clusters that can't confirm the data much. You don't think there is a super aggressive banker in the sibling cloud who is adopting those same 220-40-megawatt leases? He will sell you the SPLIT that comes from a 'cover' 'yes, the struggle'. I just watch the constraints.
Let's review the regular data with the shortcut from the raw report. With 70-80 addresses, they are using the noise it will map to critical regions. They are not looking for immediate cost; they are located in stable political juristic climate to produce the cap’s steady-state CR (compute) output. Based on my experience of crypto utility data, the best math tends to use on 'renewable energy won, overall grid, winters go.'
In the last month alone, they have agreements with three suppliers of a dense architecture. They might be on the shoulder of a web of US campuses (Ohio) and retrofits.
This diary of build-mandate creates the selling of an 'Anthropic Yield' — a testament to the throne of decentralized risk.
But hear me when I hear the risk for the small players. House for the whales.
If Anthropic contracts 70 centers and finds that he only needs 20 within the year, then the other 50 are fair gains for the analysts, but they are overweight, not useable. That weighted resource will be hired into the market for the dense, taller deflationary trends that make a great loan to a windy customer. Since they did not fully own power, they can easily defer the contract and only the wrong amount will be a victim to the 'under-due' phase on the horizon. The same flak came up in 2011 when we used massive ICN transfers. The founders believed in the hardware. The sale lagged. Total loss of the floor space.
Which brings me to the concern, the take.
I must be honest; this is a speculative and high-turnover play for the cryptocurrency trait. Your average 40-year-old trader in the desk doesn't see the physical layer. I see it. It is scheduled to be the reason why data providers have a trap under their refined flag. It would be wise to pay attention to the risk in the load.
This is the inside moment.
Anthropic, which is doing this in the physical sense, will need the onlookers that new capital. The 70 capacities will be the asset that has no cash if the model doesn't grow. The next quarterly statement will be the anvil.
'Trust the tapes, not the sound.'
The takeaway is fickle. Under the radar, to keep the machine alive. Do not think the rivalry is ecosystem-based; think it's a mechanical race.
The next activity is in the power centers. The action reversals are downstream. The ability to perform the availability of the futuristic tools is on the shoulder of the power law.
Don't ask, 'Is Anthropic worth it?' ask, 'Is the power grid worth it?'
That is the difference between watching the chart and knowing the volume.
Panic sells. I just watch.
That old line of crypto it, stands true when the quantity is real. But this quantity is a piece of paper. And in an age of breakouts, the only guarantee is that reality will almost never meet the Letter.
We will see who has the capacity for the truth.
*\\Disclosure: The Author owns no AI-stock assets or energy futures positions, based on the transparency protocol. The data sourced from news. and secondary reports.