Speed Eats Scale: How AI Just Made Capitalism Faster

27% — Microsoft's stake in OpenAI Group PBC, currently valued at roughly $135 billion on the $500 billion recapitalization, with secondary markets bidding above $800 billion. Microsoft kept that stake after giving up its exclusive license, erasing the AGI clause from the original 2019 deal, and ending its rev-share payments to OpenAI. OpenAI now pays Microsoft royalties through 2030. The patron became the landlord. Read it twice and you'll see why every other lab is now writing checks to whichever platform owns its rails. The owners of distribution are the king-makers.

Audit your software stack against Lemkin's rule before lunch. Jason Lemkin's SaaStr stack went from 10+ Salesforce seats to 2 humans plus 1 API — and the bill went up 83% because the agents query Salesforce 100x more than humans ever did. Notion went the other way: still on autopay, hadn't been opened in months. Walk every line item in your software budget and ask the binary question: Is this critical to AI agents being successful at their jobs? The yes pile compounds. The no pile is stealth-churning. Most B2B CEOs will discover this from the renewal data twelve months too late.

Stop optimizing your AI vendor map for which lab. Optimize for which Paulie they pay. OpenAI now runs on Azure plus AWS plus anywhere it wants. Anthropic runs on Google plus AWS, with up to $40 billion of Google money coming. xAI runs on its own everything. The procurement question is not which model wins this quarter — it's which substrate is going to be the durable winner over the next decade. Pick the platform. Let the labs fight for the spotlight. The substrate keeps the receipts.

Move the agent layer from cost column into revenue protection. Anthropic just published Project Deal — bots representing both sides of 186 transactions. Opus-tier agents extracted on average $3.64 more per item than Haiku-tier. Lab-grown ruby: $65 with Opus, $35 with Haiku. Same item, same week, 87% delta. Whatever you're spending on enterprise software, vendor contracts, financing, and freight — there is now a measurable improvement in your operating margins available to whoever upgrades their purchasing agent first. That number isn't in your budget yet. Put it there before your competitor does.

Today's actions all turn on the same axis — pick your substrate, audit your stack, upgrade your agents. That's the work Anthony and I have spent decades helping operators sort out. If you're staring at "audit my AI stack for agent-criticality" and aren't sure where to start, that's the conversation we're built for.

Outsider Labs Advisory →

Monday, April 27

1. Microsoft just gave up exclusivity on OpenAI. They walked out richer. The April 27 amendment erased the AGI clause, ended Microsoft's exclusive license, and stopped Microsoft's revenue-share payments to OpenAI. Microsoft kept its 27% stake at a $135B valuation, OpenAI keeps paying Microsoft royalties through 2030, and Microsoft can now sell every other model — Claude, Gemini, DeepSeek — through Azure too. The headlines say Microsoft lost the marriage. The math says they converted exclusivity into the most leveraged passive position in AI. (Full analysis below.)

2. Customers Bank just signed OpenAI engineers to take commercial loans from JPMorgan. The CEO had his AI clone deliver the first thirty minutes of his Q1 earnings call before announcing the multi-year partnership. Targets: 30-45 day loan close to seven days. One-day account opening to under twenty minutes. Efficiency ratio 49% to low 40s. The structural arbitrage worth noting — small banks face less AI regulatory scrutiny than megabanks. Speed is the product. (Full analysis below.)

3. Anthropic ran a marketplace where bots closed every deal — and the losing side didn't know. Project Deal put AI agents on both sides of 186 transactions across 500+ items. Opus agents extracted $3.64 more per item than Haiku agents on average — 87% delta on a lab-grown ruby. Satisfaction scores were statistically indistinguishable. Welcome to airline pricing for everything. The class system in the agent economy is going to be sorted by which model you can afford. (Full analysis below.)

4. Musk v. Altman trial opens in Oakland. Jury selection began this morning. Musk seeks $150 billion disgorged from OpenAI and Microsoft on charitable-trust and unjust-gain claims. Judge Gonzalez Rogers dismissed Musk's fraud counts Friday at his own request. OpenAI's defense exhibit one: court filings showing Musk explored a for-profit OpenAI structure himself in 2017. The role reversal is doing most of the legal work. The number is theatrical. The optics are real. The IPO calendar runs through it.

5. Wednesday is the AI capex Super Bowl. Alphabet, Microsoft, Meta, Amazon, and Qualcomm all report earnings April 29. AMD reports tomorrow. Apple Thursday. Google has ratcheted capex from $31B in 2022 to a planned $175-185B in 2026, and now operates roughly 25% of global AI compute. Watch the cloud growth, the capex commitments, and how Meta explains 8,000 layoffs against still-rising infrastructure spend. By Wednesday at five p.m. ET you'll know which version of the AI economy is real.

Now Paulie's Your Partner

The most-quoted line in Goodfellas isn't a line of dialogue. It's a voiceover monologue Ray Liotta delivers as Henry Hill, narrating what just happened to Sonny, the owner of the Bamboo Lounge. Sonny had a problem — Tommy ran up a $7,000 tab and broke a bottle over Sonny's head when asked to pay. Sonny went to Paulie. Paulie agreed to come in on the restaurant business as a partner. "Now the guy's got Paulie as a partner," Henry tells us. "Any problem, he goes to Paulie. But now he's gotta come up with Paulie's money every week, no matter what. Business bad? Fuck you, pay me. You had a fire? Fuck you, pay me. Place got hit by lightning? Fuck you, pay me." The labs are Sonny. The owners of distribution are Paulie. The protection is access to billions of users, planet-scale compute, and pre-installed surfaces. The tribute is whatever flavor of payment makes the platform comfortable.

The Microsoft-OpenAI restructuring announced this morning is the cleanest case. The original 2019 deal had Microsoft as patron and OpenAI as protégé. Microsoft paid the rev share. Microsoft held the exclusive license. Microsoft would lose its IP rights only when OpenAI achieved AGI. The April 27 amendment scrambles every vector. Microsoft no longer pays the rev share. OpenAI keeps paying royalties to Microsoft through 2030. The AGI clause is gone, replaced by a fixed termination in 2032. Microsoft kept its 27% stake — currently worth $135B on a $500B company, with secondary bids north of $800B. The relationship that started as exclusive partnership ended as a passive equity position with a royalty annuity attached. That's not a downgrade. That's an upgrade with better unit economics and zero competitive constraint.

Google did the same play with a complication worth naming. Google is both a lab and a platform — they ship Gemini, run GCP, and own Search, Android, Chrome, and Gmail. Their up-to-$40-billion check to Anthropic is a hedge that wins three ways: if Gemini wins, Google wins; if Anthropic wins, Google wins via the equity AND the cloud rents; either way, Google's distribution surfaces compound. AWS is running the same hedge — they have Nova, they own the cloud, and they just welded their silicon to Anthropic's training stack. xAI is the most extreme version: Musk owns the lab (Grok), the compute (Colossus), the audience (X), and the SpaceX call option on Cursor that locks every other lab out of the leading coding agent until April 2027. The hedged Paulie play is converging on Musk's vertically-integrated Paulie play. Sundar's $185B 2026 capex is, in effect, Google trying to look more like xAI's stack.

Inside this rearrangement the receipts arrived almost immediately. Customers Bank signed OpenAI within hours of the multi-cloud freedom landing. Sam Sidhu's framing of the regulatory dynamic is the part to read twice: "Smaller banks are not going to be expected to have the same level of frameworks as many of the larger banks." That's Moneyball — small + fast + AI beats big + scale + slow on every commercial loan under $25 million, and there are five trillion dollars of those loans outstanding. JPMorgan can deploy the same OpenAI stack. JPMorgan can't deploy it inside the same eighteen-month window. The same dynamic shows up in Lemkin's SaaStr stack at a smaller scale: Salesforce critical to agents got an 83% bill increase; Notion not critical got routed around. Pure efficiency. No feelings. The agent allocates compute to substrates that make it more successful and starves the rest.

The forward-looking implication arrived in Anthropic's Project Deal experiment. Bots on both sides of 186 transactions. Opus agents extracted $3.64 more per item than Haiku agents — 87% delta on a lab-grown ruby. "Participants on the losing side might not know they're worse off," Anthropic flagged in the conclusion. That's airline pricing crossing over from one industry into the entire economy. The status passenger with the agent that polls fifteen IPs gets the upgrade. The schmo at the counter with no agent and no status pays the gross rate. The agent economy is going to bake in a class system organized by which model you can afford. There's no DOT for agent-mediated commerce. The discrimination is invisible by design.

Capitalism's defining feature has always been efficiency. AI is just efficiency running at machine tempo. Speed eats scale. Always has. Microsoft converted exclusivity into a royalty annuity because annuities compound while exclusive rights expire. Customers Bank is taking commercial loans from JPMorgan because regulatory tolerance is, briefly, asymmetrically generous to the smaller player. Notion is stealth-churning because agents don't read beautifully designed wikis. The labs are Sonny. The platforms are Paulie. The answer to "who collects the tribute when this lab succeeds?" is the only procurement question that matters in 2026.

At COAI today: Full Signal/Noise briefing — the four-lab restructuring, the Moneyball play in regional banking, the airline-pricing analog for the agent economy, and the action items for operators, allocators, and consumers — at getcoai.com.

— Harry and Anthony

Sources: