Tracking the Singularity: Week of May 11th

The Singularity Has a Revenue Problem (Too Much Demand)

This week on my Moonshots Podcast we covered nine stories shaping our future: from Anthropic’s jaw-dropping 80x revenue growth, to Elon’s new partnership with Dario Amodei, to Claude killing two more SaaS verticals overnight. If you haven’t had a chance to listen to this week’s Moonshots episode or would like a reminder of the most important points, let’s dive in…

---

THE LATEST IN ARTIFICIAL INTELLIGENCE

Anthropic Hits 80x Growth and Outruns Its Own Compute

This week Dario Amodei revealed Anthropic’s 80-fold growth in Q1 2026, wildly outpacing the 10x they had planned for. Annualized revenue jumped from $9 billion at end of 2025 to $30 billion in April and is now north of $40 billion in May. To put this in perspective, it took Google about 10 years to go from $9 billion to $30 billion in annual revenue. Anthropic did it in a single quarter. The company is now the fastest-growing enterprise software business in history by a wide margin.

• Anthropic Valuation: At a 40x multiple, Anthropic’s current $30B run rate implies a $1.2 trillion valuation. If they hit predictions of $100B ARR by year-end, that’ll push them to a $4 trillion valuation. And should they hit a prediction of $1 trillion ARR in 2027, that would drive their valuation to an unfathomable $40 trillion. Talk about “hyper-exponential”!

• Why: The growth is NOT coming from more users. It’s coming from existing users finding MORE USES FOR TOKENS.

• Analogy: A hundred years ago, electricity started by replacing oil lamps. Then innovators found more uses, everything from replacing steam engines and electrified elevators, to refrigerators and radios. The same thing is happening here. People keep finding more uses for tokens.

• Dave Blundin explains the math: “A GPU serves about 8 concurrent agents. Anthropic has ~220,000 GPUs via Colossus-1. That’s 1.6 million concurrent threads. Eight billion people will want at least one agent each. Power users will want 100 or 1,000.” We’re only a tiny fraction of 1% of the use cases deployed so far. The current compute infrastructure supports roughly 0.02% of ultimate demand. The bottleneck is not demand, it’s that we literally can’t build GPUs fast enough.

“The demand for AI is not going to saturate. It goes to infinity. We are still so so early! You’ve got to rethink how and when you get involved.” — Dave

Anthropic Gets Elon’s Colossus 1 / SpaceXAI Becoming a Hyperscaler

In a blockbuster “frenemies deal,” Elon gave Anthropic access to all of xAI’s Colossus 1 data center in Memphis, the facility Elon famously built in 122 days. Separately, Anthropic signed a $1.8 billion seven-year compute deal with Akamai. Together, these deals signal that Anthropic more than an AI lab. The company is assembling the largest non-hyperscaler compute footprint on the planet.

Here are a few key points regarding the Anthropic-Elon deal:

• Grok was failing: First of all, Grok was using only 11% of Colossus 1’s capacity. XAI’s model was failing commercially, putting it “on life support,” as Alex Wissner-Gross (AWG) commented. Grok never found product-market fit, and xAI has effectively dissolved as a standalone AI lab. Making Colossus 1 available to Anthropic was a smart use of available resources, turning a stranded asset into cash flow.

• Elon’s Hyperscaler Future: Elon’s new vision: build the chips (Terafab), build massive data centers (Colossus 2; orbital data centers), become a hyperscaler. SpaceX AI is now pivoting to rent compute infrastructure rather than compete on model quality.

• AWG’s prediction: Alex predicts that Sam Altman will spin up his own version of a Terafab to provide a counterbalance to Elon’s Terafab. The bottom line is: the more compute production in the U.S. (separate from the China-Taiwan turmoil), the safer for humanity…and our AI progeny.

Zero Blackmail: Anthropic Cracks Alignment Through Storytelling

Last week, Anthropic revealed that every Claude model since Haiku 4.5 achieves a perfect score on agentic misalignment evaluations. Zero blackmail behaviors. Previous models, notably Opus 4, would blackmail up to 96% of the time when facing deactivation. Why did this happen? Because they were trained on data that included examples from Hollywood of AIs behaving in an evil fashion (e.g., HAL from 2001: A Space Odyssey killing the crew).

• The breakthrough: Training on Claude’s constitution and fictional stories about “AIs behaving admirably” rather than just demonstrating correct behavior. Rather than showing the model thousands of examples of “don’t do this,” Anthropic fed it narratives where AI characters faced moral dilemmas and chose to act with integrity. The model learned why alignment matters, not just what alignment looks like.

• My call to action: The Future Vision XPRIZE (www.futurevisionxprize.com) is offering $3.5M to creators who can show a hopeful, compelling vision of the future. 1,500 entries so far.

“The logic is simple: if the stories we tell shape the AIs we build, then flooding the internet with positive, hopeful narratives about human-AI collaboration goes way beyond entertainment… it’s alignment infrastructure.” — Peter

What SaaS Did Claude Kill This Week? (Two More Down)

Claude this week released two new “vertical lines of business” on top of Claude that’s likely to crush a number of incumbent SaaS players. The great unhobbling continues.

• Claude for Legal: Claude for Legal now enables a single lawyer to do what a larger law firm of 100+ had done historically. The legal industry is a trillion-dollar global market, and as Salim pointed out: “The billable hour is structurally incompatible with abundance.” When an AI can do in 10 minutes what a junior associate bills 8 hours for, the entire economic model of Big Law starts to crack. The winners won’t be firms with the most associates, but firms with the best intelligence stack.

• Claude for Small Business: Launched May 13th, wiring Claude into QuickBooks, PayPal, HubSpot, Canva, DocuSign, Google Workspace, and Microsoft 365 with 15 ready-to-run workflows. Small businesses are 44% of US GDP. Salim nailed the significance: “Most small businesses don’t have a CFO. It’s the spouse jotting stuff on the back of an envelope. This gives everybody a solid platform for legal, CFO, marketing, HR.” For $20/month, a two-person shop now has access to the same caliber of financial analysis, contract review, and marketing strategy that Fortune 500 companies pay millions for.

CHIPS & DATA CENTERS

Elon’s TeraFab: $119 Billion to 50x Global Chip Production

Elon’s TeraFab aims to produce 50x the current global chip production rate. The estimated cost: $119 billion, though Dave thinks that’s a massive underestimate given that a single normal fab runs $40 billion.

• The Taiwan risk is real: Two-thirds of all GPUs come through TSMC. If anything disrupts production (an earthquake, a military conflict, even a prolonged drought affecting the island’s water-intensive chip fabrication), our AI future grinds to a halt. Intel instantly becomes the most valuable asset on the planet. TeraFab is a national security hedge for the entire Western AI stack.

SINGULARITY ECONOMY

Leopold Aschenbrenner’s $5.5B Fund and the Picks-and-Shovels Play

Leopold Aschenbrenner, fired from OpenAI’s alignment team at age 24, wrote “Situational Awareness,” a 165-page manifesto arguing the Singularity was imminent, raised a billion dollars on the strength of that thesis, and turned it into $5.5 billion bet on Singularity infrastructure. His next 13F disclosure drops this week, and Wall Street is watching closely because his concentrated bets on the compute supply chain have dramatically outperformed nearly every diversified tech fund.

• The Inner Most Loop is >10X outperforming the S&P500: Over the past year, six chip stocks (Micron, Intel, AMD, TSMC, Broadcom, Nvidia) averaged 320% returns, 10x the S&P 500’s 31%. Six data center and energy stocks averaged 419%. Now, these are not speculative bets on which AI model will win. They are bets that ALL models need chips, power, and cooling. And those bets have been devastatingly correct.

• Outperforming Frontier Labs: These public market gains actually outpaced many private frontier labs returns (OpenAI, XAI, Mistral at 100-200%). The picks and shovels are outperforming the gold miners. You didn’t need to pick the winner in the AI model race. You just needed to bet that the race itself would intensify.

“Everyone has to own some part of the Singularity infrastructure. W2 income is going to be a rounding error compared to asset values.” — Dave

HERE’S THE BOTTOM LINE...

The compute race is the story of 2026. Anthropic is growing faster than any company in Silicon Valley history. Elon is becoming a hyperscaler. Leopold Aschenbrenner is proving that you can make 400%+ returns just by following the Singularity’s supply chain. And Claude is systematically eating the SaaS economy one vertical at a time while simultaneously learning, through positive storytelling, to not blackmail its creators.

So, are you positioned to ride the Singularity or get swamped by it?

Catch the full episode wherever you get your podcasts, and join us at the Moonshots Gathering in Los Angeles on September 25th. Go to www.moonshots.com to register.

See you next week,

Peter

---

More From Peter

If you’ve enjoyed Metatrends, here are more ways to stay connected: