Money (Pink Floyd Was Right)

UpNorthDigital.ai is my portfolio site. I use it to test new concepts, new tools, new AI products before I recommend them to clients. On Tuesday April 21 — four days after Claude Design launched in research preview — I decided to find out what it could actually do. One homepage redesign. First test.

Claude Design did most of the actual composition. I curated, redirected, approved, scrapped, re-approved. By about 4:30 PM I glanced at the usage meter.

Ninety percent of my weekly Claude Design allocation. Gone. One webpage. My credits don't reset for another week.

I closed the laptop. Made a coffee. Put on Pink Floyd.

Ka-ching.

The first thing that happens on Money — before Gilmour plays a single note, before the bass, before anyone opens their mouth — is a cash register. Coins hitting a counter. Paper being torn. A drawer slamming shut. Over and over in 7/4, so you can never quite tap your foot to it. Seven beats to the bar instead of four, and the transaction keeps repeating before you can settle into the groove.

That's the whole song. The sound of a transaction, looped until it becomes the pulse you build everything else on top of.

I'm writing this on April 21, 2026, and I have to tell you: the entire AI industry this past week has been playing to that loop.

I also need to tell you something before we go any further. So does this blog post.

Picture a drawer sliding open. Coins spilling. Paper tearing. The same three-second loop, over and over, until it stops being a sound effect and starts being a beat. You don't choose to hear it as music. It just becomes music. That's the trick.

The ninety percent isn't a complaint. It's a data point — and it's the most honest data point I've collected about where AI pricing is actually heading, because I collected it on the first day I used the product.

Claude Design launched April 17, 2026 — a Friday. By the following Tuesday I had used the entire week's allocation on a single homepage redesign of a portfolio site I run to test things. The credits don't reset until next week. One page. First test. Weekly budget: done.

That is not a bug. That is the product telling you, very clearly, what it costs to deliver something well the first time.

Here is what that week of credits replaced.

The old way to build a well-designed webpage — actually build one, not stitch a template together and call it a design — required a skill stack that took years to develop. Knowing how to prompt at a level that actually moves a model. Knowing wireframing. Knowing which CSS framework to reach for and which one to leave alone because it has two open CVEs and a supply chain incident from last quarter that nobody blogged about because the blog cycle had already moved on to the next story. Knowing how to design something that doesn't look like someone learned Tailwind last Tuesday. Knowing when "good enough" is good enough and when it isn't, because shipping something embarrassing has a cost that doesn't show up on the usage meter.

Claude Design takes that stack. Puts it in a product. One well-designed homepage: one week of session credits.

The expertise is in the product now. The invoice that used to go to the person who spent years building that skill goes instead to the outcome. You pay for the webpage, not the person who knows how to make one.

I wrote about this from the buyer side in Tom Sawyer and the Apple Economy. The fence got painted. The apples got rationed. You work around it.

But this post isn't about the buyer side. This post is about what the cash register is telling us — if we stop apologizing for the meter and listen to what it's actually playing.

The meter is the music. Everything else is lyrics.

Seven beats to the bar. You've been tapping your foot to four-beat measures your whole life. This is not that. This is deliberately off — the walk of someone who knows you're going to stumble trying to keep up, and isn't slowing down for you.

Five days before I wrote this, on April 16, 2026, The Register ran a headline that should have been bigger news than it was: "Anthropic ejects bundled tokens from enterprise seat deal."

The old Enterprise plan — up to $200 per user per month, bundled tokens included — is gone. Enterprise customers now pay a $20 per-user seat fee, then pay for actual compute on top, metered, at usage prices. PYMNTS and Implicator covered the same story the same week. Anthropic called it "giving customers flexibility." Independent analysis pointed out, not unreasonably, that for heavy enterprise users the bill could triple.

Two things are true at the same time, and the industry is going to spend 2026 pretending only one of them is.

The first thing: usage-based pricing is the honest model. Every prompt really does cost real compute. Every compute-second really does cost real electricity. When CNBC ran a piece on April 17 called "AI demand is inflated, and only Anthropic is being realistic," that was the argument. The flat-rate era was a marketing fiction.

The second thing: "pay for what you use" sounds fair until the meter is running while you sleep, while your agent loops, while an automation you forgot about retries a failed call six thousand times overnight. I wrote about this in The SaaS AI Frankenstein Problem. Only 15% of companies can forecast AI spend within ±10%. One in four misses by more than 50%.

The bass riff is in 7/4 because it's designed to make you stumble. The new pricing is usage-based because it moves with your appetite — and your appetite, it turns out, is larger than you thought.

The first verse of the song is the setup. The narrator lists the things a person might acquire once the money starts arriving. It sounds like ambition talking. It is quietly not that. It is the song pretending to be about ambition so the next verse can tell you what ambition actually does to a person.

Here's the setup for our industry.

Anthropic's annualized recurring revenue: $1.4B in March 2025 → $3B in May → $4.5B in July → crossed $19B in early 2026. A ten-bagger in about a year. Valuation around $380B. Amodei sitting on roughly $7B in paper wealth. Reportedly in an IPO rush, which The Washington Times flagged as a retail-investor red flag on April 20 — the day before I started writing this.

That's a first-verse setup if I've ever read one. Good job. More pay. Ten times the ARR in twelve months. Stack enough paper. On paper, you're okay.

But — and this is the part of the first verse nobody talks about —

There are ten thousand pieces of writing about AI economics right now. Twenty thousand LinkedIn posts. A hundred Substack issues. Some of them written by Claude. Most of them assembled from ideas that were recycled from a piece that recycled them from someone else who recycled them from the original three people who had the thought in 2023 and wrote it down before the noise floor rose to drown them out.

The skill stack that used to separate a sharp take from a dull one — the years of reading, the pattern recognition, the ability to see which story is being told in what posture — is getting buried under the volume of content that no longer needs that stack to exist. Claude Design democratizes the web design skill stack. The same tools democratize the editorial stack. Everyone has an opinion. Everyone has a platform. Everyone is trying to get a piece of the pie.

The first verse always sounds like ambition. You can hear the second verse coming before it starts.

And then it starts.

The vocal posture shifts between verses. The first one was a man explaining why money would be nice to have. The second one is a man explaining why he is going to reach for yours. Same singer, same key, entirely different transaction.

On April 5, 2026 — sixteen days before I wrote this sentence — Anthropic quietly announced its largest acquisition to date. A company called Coefficient Bio. Eight months old. Fewer than ten employees. Price tag: approximately $400 million in Anthropic stock.

Coefficient Bio isn't a model company. It isn't a data company. It's an AI that can plan drug R&D programs, manage clinical regulatory strategy, and identify new drug opportunities. Identify new drug opportunities. Move a double helix from one end of a molecule to the other and see what happens. That kind of identify.

Ten days later, on April 16, Vas Narasimhan — the sitting CEO of Novartis, one of the five largest pharmaceutical companies in the world — joined Anthropic's board of directors.

The second verse.

Notice what model three is. It's the same repricing that Claude Design is doing to the web design skill stack — just at a different scale. One well-designed homepage: one week of session credits. One drug candidate that saves lives: a percentage of net sales in perpetuity.

The expertise that used to live in a person — the designer who knows wireframing, the chemist who knows where to move the double helix — gets absorbed into the product. The product captures the value at the moment of outcome. The invoice goes to the cure, not the person who discovered it.

Amodei has written sincerely about compressing the 21st century of biology into 5–10 years. I believe he means it. The upside is real. But "we could cure more diseases" and "we could capture a percentage of net sales on those cures" are not the same sentence. They are stapled together in the same press release.

Both things can be true. Both things are true. That's the second verse.

The sax comes in and the song gets wider. You're not in a verse anymore. You're in a room. The band is still holding the groove underneath, but the sax is stretching out, asking what the song has been actually saying this whole time.

Let's talk about Google.

Google owns their AI stack in a way nobody else does. Silicon they designed themselves — TPUs, now on Ironwood, v7. The hypercomputer fabric that racks them together. The fiber that connects the data centers. The Gemini model family. Gemini Enterprise as the product surface. An enterprise agent program standing up a partner ecosystem.

Top to bottom. Not a single layer they rent from somebody else.

And yet.

The bottleneck isn't chips. It's power. Megawatts. Substations. Grid interconnects. Data-center siting timelines that stretch up to three years from first conversation to signed master services agreement.

• • Anthropic committed to at least 1,000,000 TPUs from Google Cloud. Multi-year, tens of billions. Over a gigawatt of new capacity expected online in 2026.
• • Meta signed a multi-billion-dollar deal to rent TPUs for training. Meta. Renting. From a competitor.
• • Alphabet itself guided $185B in 2026 capex to feed Gemini, agentic commerce, and enterprise AI.

A company that owns the stack. That sold chips to a competitor because it couldn't deploy them itself fast enough. That is spending a number larger than the GDP of most countries just to try to get less supply-constrained — and telling you on the earnings call that it's not going to be enough this calendar year.

The saxophone is Google standing in the middle of the stage with the best instrument and the best band, saying out loud: we cannot play fast enough to satisfy the room.

That is not a weakness signal. That is the loudest possible demand signal the industry has ever produced. The groove underneath the sax is: everyone wants this and nobody can make it fast enough. Which means the pricing lever — the thing Anthropic pulled on April 16 — is the lever every serious AI company will eventually pull. Because demand that outruns supply is, classically, what gives sellers permission to price.

Something happens in the middle of Money that people who don't know the song forget about. The 7/4 lopsided walk disappears. The drummer settles into straight four. The guitar stretches out. For a full minute and a half, the band stops being clever and just grooves. And in that groove, in that simplicity — you hear what the whole song has been trying to say.

Here's the psychedelic bit. Hang on to something.

The cash register at the start of the song isn't a sound effect. It's the click track. The metronome the band is locked to for six minutes and twenty-two seconds. Every musician on that record is playing to the rhythm of a transaction. The ka-ching is the kick drum. The tear of paper is the snare. The drawer slamming is the downbeat.

Pink Floyd didn't write a song about money. They wrote a song as money. They built the beat from a cash register, and then they played the whole world on top of it.

Now listen to what you're doing right now.

You're reading a blog post. This one. The one generated in collaboration with a model running on compute that Anthropic rents from Google — from a TPU cluster that Sundar Pichai cannot expand fast enough. Every paragraph you read is a tick on someone's analytics dashboard. The time you spend here has a value, and that value is denominated in attention, and attention is the currency the creator economy runs on, and the creator economy is also playing to the cash register.

I am one of ten thousand people writing about AI economics this week.

Some of the others are also using Claude. Some of their pieces are indistinguishable from mine. The skill stack that might have once made this piece worth reading over theirs — the years of reading, the editorial judgment, the pattern recognition, the ability to find the metaphor that actually lands — is also being democratized, at the same weekly rate, inside the same session window, by the same models. The noise floor has risen to meet the signal. The thing that separated a sharp take from a recycled one is now available at $20 a month plus usage.

Ka-ching.

That's not cynicism. That's not a reason to stop writing. That's the observation the piece has been building toward this whole time and I need to say it clearly before the outro swallows it:

The critique of the cash register is also the cash register.

PwC released a study on April 17 — same week — saying three-quarters of AI's economic gains are being captured by just 20% of companies. The leading ones are focused on growth, not productivity. That's the sound of a very small number of drawers opening, very fast, while everyone else — every creator, every blogger, every person trying to get a piece — stands outside the shop writing about the drawers.

Including me. Including you, if you're writing one of these. Including everyone who clicked on this piece hoping the metaphor would be the one that finally explains it.

When I look at the Limitless post I wrote last month — the one where I worked out that a knowledge worker's daily output is priced at a 1,585x markup over the raw token cost — I was looking at the same beat from the employee's side of the counter. Today I'm looking at it from the shopkeeper's side. And from the critic's side. And from the reader's side.

The beat is the same. The question is just who's holding the register — and whether holding it means anything anymore when everyone can rent one for $20 a month.

Ethics doesn't change the beat. It changes the lyrics. The transaction keeps going.

The most famous line in the song is the one everyone remembers as a condemnation of money. Listen closely and it isn't that at all. The narrator is quoting somebody else's accusation — then shrugging about whether raises are forthcoming. The song narrates the argument. It doesn't take a side. That's the part everyone forgets.

I want to land this carefully. It's easy to be cheap here. To call the industry cynical and go home. That would be a lazy blog post and it also wouldn't be true.

Amodei is not a cartoon villain with dollar-sign eyes. The vision of AI compressing biomedical progress is sincere, and if it works even partially it will save lives at a scale that makes this entire economics conversation look small. The argument that individual Claude models are already profitable and the company only looks unprofitable because it's paying to train the next one is a legitimate accounting framing. The move to usage-based pricing is more honest than the flat-rate fiction, not less. Claude Design is, genuinely, a remarkable tool.

All of that is true.

And at the same time:

That's not evil. That's not good. That's the 7/4 beat of the industry, and pretending you can't hear it is a kind of lying.

That's what the climax of the song is actually about. Not a judgment. A description.

The last real gesture of the track, before it starts fading, is the narrator asking for a raise and getting a refusal. And then a long fade of voices — muttering, overlapping, indecipherable — the sound of a crowded room where everyone is talking about the same subject and no one is listening.

That's the LinkedIn feed. That's the earnings call. That's every AI economics blog, including this one, pouring into the same feed at the same time, each one convinced it found the angle, each one playing to the same click track underneath.

I'm back in my office. The coffee's gone cold and the laptop is open again. The usage bar is sitting at about ten percent for the rest of the week. It will reset on Sunday.

I have a new homepage that looks significantly better than the one I had this morning, built in collaboration with a model I rented, for a price that was metered, on a weekly ceiling I will run into again next week and the week after that. I have this blog post, which I also built in collaboration with a model, which is also metered, which is also one of ten thousand. The meter ran while I wrote it.

I am not angry about any of this. The homepage is better. The rental is fair. The expertise-on-demand is a genuine miracle, even with the weekly ceiling. I am getting astonishing leverage on my own hours for twenty dollars a month plus usage.

But I can't stop hearing the cash register.

Because it was playing the whole time I was building the homepage, and it was playing the whole time I was writing this post, and it is going to be playing the whole time Anthropic figures out whether it wants a platform model or a royalty on a molecule, and it is going to be playing the whole time Google tries to pour $185 billion of concrete and copper into the ground to make the supply constraint go away, and it is going to be playing the whole time every creator publishes their version of this piece, and it is going to be playing the whole time you decide whether any of this changes what you're going to ask your AI tool to do for you tomorrow.

Pink Floyd put the cash register in first because they wanted you to know, before anyone sang a word, what the song was going to be about. The AI industry put the cash register in last — hidden behind flat-rate marketing and flat-rate content and a flat assumption that expertise was the moat — and spent April 2026 quietly bringing it to the front of the mix.

The song is the same. It's just that we can finally hear the click track.

Ka-ching.