Wall Street está vendendo IA para um país que a odeia

Key Takeaways

• The revenue is real: Anthropic hit $30 billion in Annualized Revenue Run rate (ARR) in early April 2026, up from roughly $1 billion at the end of 2024, representing approximately 1,400% year-over-year growth.
• The hatred is also real: 57% of registered voters say AI’s risks outweigh its benefits. Only 26% hold positive feelings about the technology. Nearly half of Gen Z, AI’s heaviest users, say it makes them afraid.
• The collision is imminent: Anthropic is targeting an October 2026 Initial Public Offering (IPO) at a potential $800 billion valuation. OpenAI is targeting Q4 2026. Combined, these two listings could exceed $1 trillion in market capitalization.
• The infrastructure is under siege: Maine became the first state to pass a data center moratorium. At least a dozen other states have introduced similar bills. Nearly half of US data center projects planned for 2026 have been delayed or canceled.

The $800 Billion Nobody Asked For

On April 10, 2026, a 20-year-old from Spring, Texas named Daniel Moreno-Gama threw a Molotov cocktail at the gate of OpenAI Chief Executive Officer (CEO) Sam Altman’s $27 million Pacific Heights home at 4 a.m. He was arrested an hour later at OpenAI’s headquarters, where he allegedly tried to smash the glass doors with a chair and told security he had come “to burn it down and kill anyone inside.” Two days later, two more people were arrested for firing a gun near Altman’s Russian Hill residence.

Two days after that, on April 14, the Maine legislature passed the first statewide data center moratorium in American history.

Four days after that, on April 18, Anthropic received investor offers valuing the company at approximately $800 billion — more than double the $380 billion at which it closed a $30 billion funding round just two months earlier.

Firebombs. State bans. An $800 billion price tag. The AI industry is living inside a paradox that has no historical precedent at this scale: the fastest revenue growth in the history of enterprise software is occurring inside a country that has turned against the product.

The Numbers That Don’t Add Up

Start with the financials. They are staggering.

Anthropic’s ARR crossed $30 billion in early April 2026. At the end of 2024, that figure was roughly $1 billion. That is approximately 1,400% year-over-year growth, a trajectory that makes Facebook’s pre-IPO growth look glacial. Goldman Sachs and JPMorgan are leading the underwriting. The target listing is Nasdaq, October 2026, with a raise exceeding $60 billion.

OpenAI is not far behind, with an ARR of approximately $24 billion as of early 2026. It just committed $20 billion to chip startup Cerebras for server capacity over three years, with warrants for up to 10% equity in the chipmaker. OpenAI is targeting its own IPO in late 2026.

Now look at the demand side: not enterprise demand, but public demand.

An NBC News poll conducted February 27 through March 3, 2026, surveying 1,000 registered voters with a margin of error of ±3.1 percentage points, found that 57% believe the risks of Artificial Intelligence (AI) outweigh the benefits. Only 26% hold positive feelings about AI. Forty-six percent hold negative feelings. A Quinnipiac University poll from March 30 reported 55% believe AI will do more harm than good in their daily lives, up from 44% just eleven months earlier.

The generational data is even more striking. Per Gallup, more than half of Generation Z in the US uses AI regularly, yet less than a fifth feel hopeful about the technology. About a third say it makes them angry. Nearly half say it makes them afraid. The demographic most likely to use the product is also the demographic most likely to resent it.

This is not a polling anomaly. This is a structural divergence between financial performance and social license.

The Infrastructure Is Already Under Siege

Revenue growth of 1,400% requires physical infrastructure. Specifically, it requires data centers, massive warehouses of Graphics Processing Units (GPUs) consuming between 20 and 150 megawatts of electricity each. And that infrastructure is being legislatively dismantled.

Maine went first. On April 14, 2026, the Maine Senate took a final vote on LD 307, enacting an 18-month moratorium on any data center requiring 20 or more megawatts of power. No state, local, or quasi-governmental agency in Maine can issue a permit for such a facility until at least November 2027.

The vote killed a $550 million proposal from Sentinel Data Centers to convert the former Androscoggin paper mill in Jay, Maine, a project that would have generated 800 to 1,000 construction jobs and 125 to 150 permanent positions in a community that lost its economic anchor when the mill closed in 2023. Governor Janet Mills, a Democrat, specifically asked the legislature for an exemption for Jay. The legislature refused.

Jay’s town government wanted the project. The state legislature killed it anyway. That is how toxic the brand has become.

Maine is not alone. Lawmakers in at least a dozen other states have introduced bills or resolutions that would pause data center development in some form, including Vermont, New Hampshire, New York, Virginia, Michigan, and Wisconsin. On March 25, Senator Bernie Sanders and Representative Alexandria Ocasio-Cortez introduced the federal AI Data Center Moratorium Act, which would halt all new data center construction nationwide until Congress passes laws protecting workers, consumers, and the environment. The bill is unlikely to pass, but it establishes the legislative template.

At the local level, the numbers are worse. Over the past two years, $18 billion in data center projects have been blocked and another $46 billion delayed due to community opposition, totaling $64 billion. A total of 142 activist groups across 24 states are now organizing against construction. Data center project cancellations surged in 2025 as local pushback intensified.

And the supply chain is failing independently of the politics. According to analysts at Sightline Climate, between 30% and 50% of AI data centers planned for deployment in the US in 2026 will be delayed or canceled due to transformer shortages, power grid constraints, and tariff-disrupted supply chains. Of 16 gigawatts of capacity slated to come online before the end of 2026, only about 5 gigawatts (GW) are currently under construction.

The infrastructure the IPO valuations depend on is being attacked from three directions simultaneously: democratic opposition, regulatory action, and physical supply constraints.

The Facebook Precedent (and Why It Doesn’t Fully Apply)

Wall Street’s response to the sentiment data will be predictable: “Facebook.”

Facebook went public on May 18, 2012 at a $104 billion valuation, the largest US tech IPO at the time, amid intense privacy backlash following years of scandals over user data handling. The stock cratered. It closed the first week at $26.81, tumbled below $18 by September 2012, losing more than half its IPO value. The Wall Street Journal called it a “fiasco.”

Then it became a trillion-dollar company.

The lesson Wall Street took from Facebook: public sentiment doesn’t matter. Revenue does. Institutional investors price cash flow, not Gallup polls. And on pure revenue metrics, Anthropic and OpenAI look extraordinary.

But the Facebook analogy has a structural flaw that bulls will not want to hear.

Facebook did not require physical infrastructure in communities that could vote to ban it. Facebook did not consume 4.4% of the national electricity grid with a trajectory toward 12% by 2028. Facebook did not face a single state-level moratorium on its facilities. Facebook did not require $4.3 billion in ratepayer-funded grid upgrades across seven states in the PJM interconnection region just to connect its hardware.

Facebook was a software product. It needed servers, but it did not need the industrial-scale energy infrastructure of a mid-sized city in every county it entered. AI does. And that infrastructure sits in someone’s backyard.

The more precise historical parallel is not Facebook. It is nuclear power.

The Nuclear Playbook

In the 1970s, nuclear energy was the technology of the future. The physics worked. The economics worked. Utilities were building reactors at scale. Then Three Mile Island melted down in 1979. Public sentiment turned. Chernobyl in 1986 made it permanent.

The technology was still sound. The economics were still viable. But no new nuclear plant was ordered in the United States for over three decades.

What killed nuclear was not physics. It was politics. Community opposition made siting impossible. Regulatory requirements, born from public fear, made construction costs spiral. Utilities stopped trying. The industry entered a multi-decade hibernation from which it is only now emerging, largely because AI data centers need baseload power that renewables alone cannot yet provide.

AI faces an uncomfortably similar fork.

The technology works. The revenue is real. Enterprise demand is growing at 1,400% year over year. But the communities that would host the infrastructure are saying no. The legislatures that govern the permitting process are saying no. And the voters those legislatures answer to are saying no by a 57-to-34 margin.

Nuclear power did not die because the reactors failed. It died because the towns refused to build them. AI’s revenue is not threatened by the technology failing. It is threatened by the infrastructure never arriving.

The Steelman: Why the Bulls May Be Right

There is a strong case that none of this matters for the IPOs.

First, institutional investors, the entities that actually drive IPO pricing, do not track public sentiment polls. They track enterprise revenue contracts, gross margins, and forward booking rates. On every one of those metrics, Anthropic is performing at generational levels. A company growing at 1,400% with enterprise contracts from Amazon, Google, Apple, and the federal government does not need the general public’s permission to generate returns.

Second, public sentiment toward technology has been negative before and tech stocks recovered. Post-Snowden in 2013, post-Cambridge Analytica in 2018. Both events triggered polling drops and regulatory scrutiny. In both cases, the underlying business models were unaffected. The stocks recovered within months.

Third, the backlash could actually benefit the largest players through regulatory capture. If moratoriums freeze new data center construction, existing facilities (owned by Amazon, Microsoft, and Google, the same companies buying Anthropic and OpenAI’s products) become more valuable. A supply constraint with fixed demand is not bearish. It is a moat.

Fourth, the AI Data Center Moratorium Act has zero chance of passing a Republican-controlled Congress. Interior Secretary Doug Burgum called the bill “complete capitulation,” stating it would be “like waving the surrender flag to China.” Democratic Senator John Fetterman agreed, writing “I refuse to help hand the lead in AI to China.” Maine is a state of 1.4 million people. Virginia, Texas, and Ohio, where the bulk of US data center capacity is being built, show no signs of following.

These are not trivial objections. Any honest analysis of the AI IPO landscape must contend with the fact that Anthropic is generating revenue at a rate that would rank it among the fastest-growing companies in the history of American enterprise, and that public polls are a poor predictor of stock market performance.

The Counter-Steelman: Why Sentiment Is a Leading Indicator

But the nuclear precedent cuts deeper than the Facebook one for a specific structural reason.

AI’s growth requires new physical infrastructure in new locations at continuous scale. Every quarter of sustained growth demands more data centers, more grid interconnections, more cooling water, more transformers. The revenue generated this quarter depends on infrastructure built this quarter. And the pipeline of that infrastructure is contracting, not expanding.

Consider the math. The US Department of Energy (DOE) projects that data centers could consume up to 12% of total US electricity by 2028, up from 4.4% in 2023. That trajectory requires the equivalent of dozens of new power plants. Those plants need to be sited somewhere. The communities at those sites are voting against them.

This is not a soft risk. This is a physical constraint on the revenue growth that justifies the valuation.

When a Sightline Climate analysis shows that only 5 GW of the planned 16 GW of data center capacity is actually under construction, that is not a sentiment problem. That is a capex delivery problem. And when 142 activist groups across 24 states are organized to block the remaining pipeline, the delivery problem has a political accelerant.

The bulls are right that institutional investors don’t read Gallup polls. But institutional investors do read permit denials, grid interconnection queues, and construction timelines. And those metrics are deteriorating.

What Happens Next

Here is what is structurally different about the AI IPO moment compared to every prior tech IPO cycle:

Factor	Facebook (2012)	AI (2026)
Physical infrastructure required	Minimal (servers)	Massive (data centers, grid, water)
Community opposition	None	142 groups, 24 states
State-level bans	Zero	1 passed, 12+ pending
Federal moratorium bill	None	Sanders-AOC bill introduced
% of national grid consumed	<0.1%	4.4% → 12% by 2028
Physical attack on CEO	None	Molotov cocktail, gunshots

The question is not whether AI companies are profitable. They obviously are. The question is whether the infrastructure pipeline can sustain the growth rate that justifies the valuation, given that the country’s political system is actively contracting that pipeline.

If Anthropic IPOs at $800 billion, it is being priced for a future in which $30 billion in ARR continues compounding at triple-digit rates for years. That compounding requires data centers. Those data centers require permits. Those permits require communities. And those communities are saying no.

The revenue is real. The growth is real. The technology works. But the social license to build the infrastructure that sustains the growth is evaporating in real time.

Nuclear power proved that a technology can be right about the physics and wrong about the politics — and the politics can win for decades. AI’s IPO architects are betting that this time is different.

The 57% of Americans who say the risks outweigh the benefits will get to vote on that.

What This Means for You

If you are considering buying AI IPO shares at listing:

• Evaluate the data center pipeline, not just the revenue growth. Ask: where will the infrastructure come from to sustain the growth curve that justifies the IPO price? If half the pipeline is already delayed or canceled, the forward revenue assumptions deserve scrutiny.
• Watch Maine. If Governor Mills signs LD 307, it establishes a template. If two or three more states follow in the next six months, the infrastructure constraint becomes material.

If you are in the AI industry:

• The backlash is not irrational. It is structural. Communities are paying higher electricity bills, facing water shortages, and receiving few permanent jobs in return. The data center tax arbitrage that shifts costs to ratepayers is well documented. Until the industry addresses the externalities, the opposition will grow.
• The Molotov cocktail at Altman’s house is an escalation, not an outlier. Physical security for AI executives is now a cost of doing business.

If you are in energy or utilities:

• The grid strike is becoming a grid revolt. Utilities that refused to build without 20-year guarantees now face communities that refuse to let them build at all. The infrastructure gap between AI demand and grid capacity is widening on both the supply side (utility reluctance) and the demand side (community opposition).

The Bottom Line

Anthropic grew its revenue 1,400% in a country where only 26% of the population views AI positively. OpenAI’s CEO had a Molotov cocktail thrown at his home. The first state data center ban just passed. A federal moratorium bill is on the congressional record. Nearly half the planned data center pipeline is stalled.

And Wall Street wants to sell you more than $1 trillion of it.

The technology works. The revenue is real. The hatred is also real. And the infrastructure the growth depends on is being fought at every level: town, state, federal, and now physical.

History offers two templates. Facebook proved that you can IPO into hatred and win. Nuclear power proved that you can be right about the physics and still lose for 30 years.

The AI industry is betting everything on being Facebook, not nuclear. The 57% will determine which.