🙏 The AI Trinity. Hype Is Real. The Math Is Not. Amen.
$25 million checks. A papal encyclical. And the inconvenient truth that AI tokens already cost more than the humans they were supposed to replace.
Sometimes last week, I’ve seen under my own eyes the unravelling of a wire transfer of $25 million into Anthropic’s cap table via one of our SPVs. This is probably the largest transaction I’ve been involved with so far.
It was not the only one. It was not the largest one the private market has seen in 2026. And that scares me.
This has become the rhythm of the AI capital cycle: announcements of rounds so large they barely register as news anymore.
Anthropic at $900 billion valuation.
OpenAI filing a confidential S-1 at $1 trillion.
SpaceX listing on Nasdaq in June at $1.75 trillion.
Three companies. Combined valuation: $3.65 trillion. More than the GDP of Germany.
4 years ago, I signed up as an investor in Climate Tech, a movement to mobilize capital into innovations that would save the planet. And somehow now the AI hype is hard to avoid, even for myself, and I feel kind of bad about it. All this capital, 2.7 trillion dollars last 12 months alone went to AI, and for what? What could have we done with this if it was funnelled to climate tech?
I want to be clear: I am not writing this as a skeptic of AI. I am writing it as someone who has spent the past 4 years watching capital flow into the technologies that shape civilization, someone who believes this particular AI cycle deserves an honest accounting.
Because last week, while the wire transfers to Anthropic were clearing, two other major things - for the century? - happened.
The Pope issued a 43,000-word encyclical calling for the disarming of artificial intelligence.
And meanwhile, Uber’s CTO quietly admitted his company had burned through its entire 2026 AI budget in four months.
Only one of those stories made the front page. The other is the one that matters more.
I. The Hype: $25 Million Checks and the Self-Fulfilling Prophecy ⚡
Last week, I witnessed a single family office wiring $25 million into a seller of Anthropic shares. And we’ve got other family offices in discussion for even larger checks for the same deal.
Let me show you how the capital cycle works in practice: Not in Wall Street, on Whatsapp.
Not because it has done a deep analysis of the unit economics, the token costs, or the actual enterprise adoption curves.
Just because:
No investment committee in the world gets fired for backing OpenAI, Anthropic, or SpaceX.
There is a lot of money (dry powder) sitting on bank accounts right now. A LOT. And the economy needs it to go somewhere…
This is the self-fulfilling prophecy of the American AI stack- and it is the most important and least discussed dynamic in the current cycle.
The logic runs like this:
The United States has declared AI its sovereign industrial priority.
The CHIPS Act, the AI Action Plan, the $800 million DOE grants for small modular reactors to power data centers, these are not tech investments, they are strategic declarations.
Family offices in the US, in Singapore, Private Equity funds in Europe, all face the same calculus: if you are not in the American AI stack, you are not in the future: Your LPs
2026: Not the AGI, But The $3.65 Trillion Moment.
SpaceX is targeting a $1.75 trillion valuation and a $75 billion raise. OpenAI is filing a confidential S-1 targeting a September listing at $852 billion to $1 trillion — which at the upper end would be the largest technology public offering in history. Anthropic is raising at a $900 billion valuation and targeting an October listing.
The three IPOs together could demand north of $200 billion from public markets. The entire US IPO market raised just $45 billion in all of 2025.
There is an estimated $8 trillion sitting in U.S. money market funds. SpaceX’s $75 billion raise represents roughly 1% of that. Institutional money has spent years accessing AI via proxy — buying Nvidia for chip exposure, Microsoft for its OpenAI stake. The moment pure-play labs are publicly available, that demand unlocks.
This is not bubble logic. This is momentum logic. And momentum logic is self-validating until it is not. The $25 million checks keep clearing because the $25 million checks keep clearing. No one gets fired for backing the biggest names. Everyone gets fired for missing them.
The Pope noticed.
II. The Pope Got Scared… Yet, He May Be Creating Exactly What He Fears 🔮
On May 25, 2026, Pope Leo XIV stood in Vatican City and personally presented Magnifica Humanitas — “Magnificent Humanity” — the first papal encyclical in history dedicated to artificial intelligence, signed on the same day that Rerum Novarum was released 135 years previously.
Pope Leo XIV called for the “disarming” of artificial intelligence, warning against “a race for ever more powerful algorithms and larger datasets” driven by “the desire to secure geopolitical or commercial dominance.”
Echoing the worker-centered concerns of Rerum Novarum, he warned that rapid automation could displace workers and reshape labor markets in ways that would risk leaving many in forced inactivity, undermining both human dignity and social stability.
The Vatican launch included remarks by the co-founder of Anthropic — which is currently locked in a legal battle with the Trump administration over access to its AI technology. The Vatican decided to involve Anthropic as part of its decade-long effort to engage Silicon Valley in dialogue.
And yet in his text,
🦸🏻♂️ Pope Leo repeatedly blasted the concentration of power and data in the hands of so few people in the private sector as a danger.
Anthropic sat on the stage as the Pope warned against companies like Anthropic (lol yeah for real). That detail tells you everything about where we are.
But here is the paradox that no one is naming directly.
The Pope’s warning, and the institutional, regulatory, political attention it generates , will accelerate the capital flow it is trying to slow!
When the Church issues a formal doctrine on AI, it confirms to every allocator in the world that AI is the defining civilizational technology of our era. That it requires the same moral and regulatory architecture as the industrial revolution, nuclear power, or the printing press. That it is not a product category but an epochal shift.
That confirmation is not a deterrent for capital. It is a green light.
The $25 million checks will keep clearing. in my Whatsapp groups.
The IPOs will price. The valuation multiples will hold or exceed their targets. Not despite the papal encyclical, but partially because of it.
When the most powerful religious institution in human history names your industry as the central moral challenge of the age, you have already won the narrative war.
Basically, the self-fulfilling prophecy of the American AI capital stack just got a 43,000-word endorsement from Rome.
🔮 So my forecast is that the AI bubble will not burst before at least mid 2027.
III. The Math That Nobody Is Talking About 💰
Here is where I want to slow down. Because the hype is real, but the math is not.
70% of enterprise AI agent projects will be shut down by the end of 2027…
For 77% of jobs: humans are still cheaper. The MIT study said it in 2018. The Uber CTO confirmed it in 2026.
🌬️ The architecture that closes financially for the next 15–20 years is not AI replacing humans.
Uber’s CTO Praveen Neppalli Naga told The Information his company had burned through its entire 2026 AI coding budget in four months. “I’m back to the drawing board,” Naga said, “because the budget I thought I would need is blown away already.”
Bryan Catanzaro, VP of Applied Deep Learning Research at Nvidia — the $5+ trillion company making the chips powering much of the AI industry — told Axios “for my team, the cost of compute is far beyond the costs of the employees.”
Read that again. The VP of Applied Deep Learning Research at Nvidia, the company that profits most from AI compute spending, is saying that compute already costs more than the people it is supposed to replace.
The replacement narrative is not an economics argument. It is a marketing one. The only architecture that closes financially today is human + AI together.
Gartner is now forecasting that more than 40% of enterprise AI agent projects will be shut down by the end of 2027, citing escalating costs and unclear business value.
A 2018 MIT study suggested that AI automation is economically viable in only about 23% of jobs. Ultimately, humans are still cheaper in the remaining 77% cases.
Some power users are racking up monthly token bills north of $150,000. “I probably spend more on Claude than my salary,” one software engineer in Stockholm told the New York Times.
The numbers get stranger. Within Meta, one employee created a dashboard called “Claudeonomics” so teams could see who generated the greatest AI usage, turning consumption into a competitive metric.
This is not productivity. This is performative consumption.
The companies that replaced humans with AI agents are not finding the savings they modeled. Short-sighted enterprises that have replaced humans with AI could find themselves paying five times as much to maintain the tech. The only winners in this scenario are the tech overlords.
Goldman Sachs forecasted that agentic AI could drive a 24-fold increase in token consumption by 2030, reaching 120 quadrillion tokens per month 🤮.
But Gartner found that cheaper tokens won’t translate to cheaper enterprise AI because agentic models require far more tokens per task than standard models, and increased consumption can outpace falling unit costs. This is kind of important, if you are still reading.
The pitch for enterprise AI was relentless cost reduction. The reality is: the costs moved, they did not disappear. They moved from human payroll to compute infrastructure. And compute infrastructure, at the token consumption rates the agents require, already exceeds the cost of the humans it was supposed to replace.
The Pope is worried about displacement. The CFOs are worried about the invoice.
They are both, in different ways, very right.
IV. The Only Way Forward: Human-in-the-Loop as the Economic Constraint + 🌏 Save the planet (…lol?)
Here is the conclusion the data is forcing not as a moral argument, but as a cost argument.
In short: “bro, don’t fire everyone for a few AI agents, just keep paying a few employees and give them Claude to doublecharge their work productivity.”
The AI-only model does not close financially at current token and electricity costs.
Full AI agent deployment, autonomous systems running 24/7, handling tasks end-to-end without human oversight, consumes tokens at a rate that, in most enterprise applications, exceeds the fully-loaded cost of the human workforce it replaces. This is not a transitional problem waiting for better hardware.
By 2030, inference costs are expected to fall nearly 90% — but aggregate consumption will rise so fast that total enterprise AI costs will continue to climb regardless of per-token price reductions.
The economic equilibrium, for the next fifteen to twenty years, is human-in-the-loop.
Not because humans are better than AI at most tasks.
Not because the 🦸🏻♂️ Pope said so.
Because the compute and electricity costs of running fully autonomous agents at scale do not clear the basic financial test of replacing a human worker.
The companies that will extract real value from AI in this cycle are not the ones replacing humans with agents. They are the ones building systems where AI handles the volume and humans handle the judgment. Where AI compresses the research and humans make the call. Where AI generates the output and humans own the decision.
This is not a consolation prize for humanity. It is the architecture that closes financially.
And it points to a new metric the financial world has not yet priced: compute credits.
We have spent the last decade building carbon credit markets, financial instruments that assign a cost to the externality of carbon emissions, forcing it onto balance sheets where it had previously been invisible.
The externality that the current AI cycle is generating is compute consumption, measured in tokens, expressed in electricity, ultimately constrained by the same energy infrastructure that every other industrial system runs on.
A single ChatGPT query consumes roughly ten times the electricity of a Google search.
A fully autonomous AI agent running 24/7 on enterprise tasks consumes orders of magnitude more.
The total electricity demand of AI data centers is projected to represent a growing and significant share of U.S. grid capacity by 2030.
The insane madness we are going into…
Here is the hidden cost of your AI agent slope doing half *** work.
⚠️ A few important numbers to put things in perspective:
460 TWh - Global data center electricity consumption in 2025 — rivals the annual consumption of FranceIEA · Sustainability Atlas 2025
800 TWh - IEA projection for data centers by 2028 — nearly double 2025 levelsIEA Energy & AI Report 2025
50 GW - Additional dedicated power Anthropic says the U.S. must build by 2027 to sustain the AI economyEarth911 · March 2026
Carbon credits forced carbon costs onto balance sheets. Compute credits, a framework that accounts for the true energy cost of token consumption, would force the same reckoning onto the AI economy.
The companies building toward solar sovereignty, nuclear baseload, distributed energy infrastructure, the vital industries the Quiet Renaissance is funding, are not adjacent to the AI economy. They are the constraint the AI economy is about to run into.
The Pope is asking for a more humane AI. The market is about to ask for a more energy-efficient one.
Both arguments lead to the same place: human intelligence and artificial intelligence operating together, within the energy envelope civilization can actually sustain.
At the end of the day I tell you what. I have a feeling that the Anthropic, the Nvidia, the Google, bref, the Magnificent 7 will make their money through infrastructure & datacenters rentals.
The big questions is, how many of these small, super young AI agents startups funded last 24 months will exist in 10 years. Less than 5%? Less than 1%?
Only God knows.
V. We Have a $5M Block. It Will Not Last the Week.
The $25 million commitment referenced earlier in this issue is real. The buyer is in our network. The allocation is confirmed.
We are co-investing alongside them.
WeTheAtlas has secured a $5 million layer 1 block SPV for co-investment alongside the previous investor mentioned in Anthropic’s pre-IPO round. Same terms. Same entry point. Last private pricing before the October listing.
This is secondary shares sales from a GP selling his stakes directly on the cap table.
To be direct about the math: Anthropic’s run-rate revenue crossed $44 billion annualised in May 2026. The company is on track to post its first operating profit this quarter. The IPO window opens in October. The valuation at listing will not look like the valuation today.
If you want to understand the details minimum ticket, structure, timeline, documentation indicate your interest now via WeTheAtlas and we will come back to you directly within 36 hours.
Commitment will close fast. We will not hold allocation for undecided capital.
→ https://wetheatlas.com/solutions/698a8b1b0b3211f8eabe0c2e
This is not an offer to sell securities. WeTheAtlas Advisors LLC (EIN. 37-2207031) facilitates introductions to co-investment opportunities for qualified investors only. All investments carry risk. Past performance does not guarantee future results.
Conclusion:
I have to apologize loosing my nerves writing this newsletter but some of the stuff I’m seeing here is really excruciating.
Fight the Good Fight.
Djoann
Sources & to go further down the rabbit hole as I did the past 2 weeks for you my friend:
Pope Leo XIV, Magnifica Humanitas, Vatican City, May 25 2026 ·. Rerum Novarum, Pope Leo XIII, May 15 1891 · SpaceX S-1, SEC/EDGAR, May 20 2026 · Reuters, Bloomberg, FT on OpenAI and Anthropic filings May 2026 · The Information on Uber AI budget, April 2026 · Axios on Nvidia compute cost statement · Gartner agentic AI project forecast 2026 · Goldman Sachs token consumption forecast · MIT automation viability study 2018 · Fortune on Microsoft AI cost report May 2026 · The Adaptive Economy research · June 2026
10 years ago I pushed Greenpeace board to examine the tech industry’s energy material and climate footprint . The A-F grades fluff up remote renewable & underestimate exponential growth of tech industry Ai segment demands for “hard” energy and materials. Sustana’s Global Cooling impact fund is a distinctive fast-growing AI infrastructure efficiency play for “growth at a reasonable price” investors who want a piece of next generation infrastructure that will be required post 2027 after the valuation bubble deflates- as Dolan predicts. We need a $20M anchor wire to do 1st close. contact: jonathan @sustana.co .