☕ Anthropic went from $380B to $900B. In 3 months.

Tuesday, April 28, 2026 | Private Markets Intelligence

Good morning. This was the week AI valuations completely decoupled from traditional metrics. Anthropic went from $380 billion to $900 billion in three months—and investors are fighting to get in. Big Tech committed $650 billion to AI infrastructure in 2026 alone, raising $108 billion in debt to fund it. Elon took the stand in his $130B lawsuit against OpenAI, accusing Sam Altman of "stealing a charity." And SoftBank—fresh off losing hundreds of millions on AI pizza—announced a $100B robotics company targeting H2 2026 IPO.

If you're confused whether this is a bubble or the beginning of a trillion-dollar infrastructure buildout, you're not alone. Here's what actually happened this week and what it means.

Alberto, Jose, and the Founderscrowd team☕

This week's top 5:

Founderscrowd: Anthropic is fielding multiple unsolicited offers to raise $40-50 billion at a valuation between $850 billion and $900 billion. Board meeting in May to decide. This would be the company's final private round before an October IPO. In February, Anthropic raised at $380 billion. Three months later: $900 billion. That's not growth. That's a re-rating.

The numbers:

OpenAI's last round closed at $852 billion in March. If Anthropic accepts at $900B, it becomes the most valuable AI company in the world. The title OpenAI has held since 2019.

Why this matters:

This isn't about AI models anymore. It's about who controls the infrastructure to train and deploy them. Anthropic locked in massive compute deals this month:

Investors aren't betting on Claude being slightly better than ChatGPT. They're betting Anthropic controls enough compute capacity to out-scale everyone else. And unlike OpenAI (which reportedly missed revenue targets and is burning cash on Sora), Anthropic is growing 3-4x year-over-year on enterprise revenue.

The institutional investor who prepared to commit $5 billion still can't get a meeting with CFO Krishna Rao. Demand is that high.

Bottom line: Anthropic went from scrappy OpenAI competitor to potential market leader in 3 months. The shift wasn't gradual. It was a baton change. OpenAI's CFO is privately warning about funding shortfalls. Anthropic is turning away $5B checks. The AI race just flipped.

Founderscrowd: Amazon, Microsoft, Google, and Meta collectively committed $650 billion in capital expenditure for 2026—a 36% increase from 2025. Roughly 75% ($450B+) is tied to AI infrastructure: data centers, chips, power, cooling systems. This is the largest infrastructure buildout in tech history. And it's being funded with debt.

The breakdown:

Total 2026 capex: $650B+
2025 capex: $448B
Growth: 45% year-over-year

How they're funding it:

Big Tech raised $108 billion in debt in 2025 alone. JP Morgan projects $1.5 trillion in tech debt issuance over the coming years. Why? Because even with massive cash flows, AI infrastructure spending now consumes 94% of operating cash flows after dividends and buybacks (Bank of America calculation).

These companies are profitable. AWS grew 24%. Google Cloud grew 48%. Azure grew 39%. But AI capex is growing faster than revenue. Microsoft's Amy Hood said the company expects to stay "capacity constrained through 2026" even with $190B in spending.

Why this matters:

AI is no longer a software feature. It's a capital-intensive industrial buildout. Access to compute is becoming a strategic advantage, not a routine operating expense. The companies that can afford to spend hundreds of billions (Big Tech) are pulling away from everyone else.

For startups: If you can't afford your own infrastructure, you're paying rent to Amazon, Microsoft, or Google. And they're your competitors.

For investors: The returns on this $650B are speculative. AI services generate ~$25B in direct revenue today. That's a 26x spend-to-revenue ratio. Either AI revenue is about to explode (justifying the spend), or this is the most expensive bet in tech history.

Meta's stock dropped 6% after raising capex guidance. Microsoft and Amazon slipped on AI buildout costs. Alphabet was the only gainer (cloud strength). The market is skeptical that $650B in spending will generate proportional returns.

Bottom line: AI winners will be decided by infrastructure depth, not just model quality. Big Tech is spending $650B to make sure they control that infrastructure. Everyone else is either a customer or a competitor with no compute.

Founderscrowd: Elon Musk took the stand in federal court as opening statements began in his $130 billion lawsuit against OpenAI. Elon's claim: Sam Altman "stole a charity" by converting OpenAI from nonprofit to for-profit. OpenAI's defense: Elon sued because he "didn't get his way" after leaving in 2018.

What Elon wants:

This isn't a settlement negotiation. This is scorched earth.

What's at stake:

If Elon wins (unlikely), it sets a precedent that nonprofits can't convert to for-profits and make founders/investors billions. That would freeze every AI company's exit plans.

If Elon loses, it validates the nonprofit-to-for-profit playbook. Expect more companies to pull the same move.

What we're about to see:

Four weeks of courtroom testimony. High-profile witnesses: Sam Altman, Greg Brockman, potentially Ilya Sutskever (former Chief Scientist). And hundreds of pages of private messages—Slack, email, texts—between Elon, Sam, Greg, and the board from 2015-2024.

Every internal debate about mission vs profit. Every disagreement about safety vs scale. Every message about why Elon really left in 2018. All public.

If you thought the 2023 Sam Altman firing drama was messy, this is 10x worse.

Why this matters:

OpenAI's credibility is on trial. If internal messages show Sam Altman misled Elon about the nonprofit mission, it undermines OpenAI's entire "safety-first" narrative. If messages show Elon left because he wanted control (not safety concerns), his lawsuit collapses.

Either way, we're about to learn the real history of OpenAI. Not the sanitized blog posts. The actual fights, the actual motivations, the actual reason OpenAI went from nonprofit research lab to $852B for-profit company in 8 years.

Bottom line: This trial isn't just Elon vs Sam. It's a test case for whether nonprofit-to-for-profit conversions are legal when they make insiders billions. The discovery process alone could reshape how we understand OpenAI's entire trajectory.

Last week, Premium members got access to two deals: a fintech solving an $80B broken market (CEO has 3 exits, 40% NJ tax rebate) and a $20M revenue cutlery business preparing for a 2030 exit.

This week: A biotech play solving antibiotic resistance. NIH-backed, early-stage, high-risk/high-reward.

Premium members get:

$40/month. Cancel anytime.

Going to $100/month this week.

Upgrade to Premium

Founderscrowd: SoftBank is assembling a new company called Roze AI that would deploy fleets of autonomous robots to build data centers in the U.S. Target valuation: $100 billion. IPO timing: H2 2026. CEO Masayoshi Son has committed tens of billions to AI infrastructure. Roze extends that to the physical construction layer.

The pitch: AI data centers are the bottleneck. Labor is expensive and slow. Power costs are up 66% (we covered this Tuesday). If Roze can deploy robot fleets that build data centers faster and cheaper than human labor, that's a real unlock for AI scaling.

The problem: Masayoshi Son's track record is uneven. Six years ago, he sank hundreds of millions into Zume—an AI pizza startup that was supposed to use robots to deliver fresh-baked pizzas from trucks. Zume raised $445 million, peaked at $2.25B valuation, then collapsed spectacularly in 2020. Turns out robots can't make pizza profitably.

So when SoftBank announces a $100B robotics company targeting H2 2026 IPO, insiders are questioning whether a pricey new robotics spinout is worth the risk.

Why this might work:

Data center construction robotics solves a real problem. Meta, Google, Amazon, Microsoft are spending $650B on AI infrastructure. If Roze can cut construction time by 30-50% and reduce costs, there's massive demand.

SoftBank recently acquired ABB Robotics and has deep investments in energy, land, and infrastructure. The pieces are there for a vertically integrated data center construction company.

Why this might fail:

$100B valuation for a company that doesn't exist yet. No working prototypes. No proven deployments. Just a pitch deck and Masayoshi Son's conviction that robotics + AI infrastructure = massive opportunity.

If Roze actually IPOs in H2 2026 at $100B, it'll be one of the most speculative IPOs in history. If it doesn't, it'll be another Zume—hundreds of millions burned on robots that couldn't deliver.

Bottom line: Data center construction is a real bottleneck. AI needs infrastructure fast. But $100B for a company with zero revenue, zero customers, and a CEO who lost hundreds of millions on AI pizza? That's either visionary or delusional. We'll find out which in six months.

Founderscrowd: Here's the number that matters most: Big Tech is spending $650 billion on AI infrastructure in 2026. AI services are generating roughly $40 billion in direct revenue. That's a 16x spend-to-revenue ratio.

The math:

Combined AI revenue (estimate): $100-120B across all players
Combined AI capex: $650B
Gap: 5-6x spend vs revenue

Even if we're generous and say AI is generating $120B in total revenue across all companies, that's still a 5.4x spending overhang.

Why this matters:

Either (a) AI revenue is about to explode 5-10x in the next 2 years (justifying the spend), or (b) this is the most expensive infrastructure bet in history with unclear payback timelines.

Microsoft CFO Amy Hood said even with $190B in capex, the company expects to stay "capacity constrained through 2026." Translation: Demand is so high they can't build fast enough. That supports the bull case.

But Bank of America notes that hyperscaler capex now consumes 94% of operating cash flows after dividends and buybacks. Companies are issuing debt ($108B in 2025, $1.5T projected over coming years) to fund this buildout. That's not normal. That's betting the company on AI infrastructure returns.

The bull case:

AI revenue is growing 3-4x year-over-year (Anthropic tripled in 4 months). Enterprise adoption is accelerating (1,000+ customers spending $1M+ annually with Anthropic alone). Cloud growth is strong (AWS +24%, Azure +39%, Google Cloud +48%). The $650B spend creates the capacity to capture a $1T+ AI market.

The bear case:

AI models are commoditizing fast. Pricing wars are driving margins down. Most AI revenue today is cloud infrastructure (not AI applications). And the $650B being spent could take 5-10 years to pay back—if it pays back at all.

Meta's stock dropped 6% on higher capex. Microsoft and Amazon slipped. The market is skeptical.

Bottom line: The AI infrastructure buildout is either the smartest bet in tech history or the most expensive mistake. We'll know which in 2-3 years when we see if $650B in spending generated $1T+ in returns or created a massive stranded asset problem.

Anthropic board decision (May):

Big Tech Q2 earnings (July):

Elon vs OpenAI trial (ongoing):

SoftBank Roze IPO timeline:

AI revenue growth:

Alberto, Jose & the FC team.
Founderscrowd

P.S. If you're wondering whether Anthropic's $900B valuation is justified or peak bubble—join Premium. This week's analysis breaks down the revenue multiples, the compute capacity math, and the comparable exits. The answer isn't what you think.