GM. This is Milk Road AI, where we follow the capital so you don't have to.

​Here’s what we’ve got for you today:

Nexo is back in the U.S. - and new clients get 30 days of Wealth Club Premier perks! Higher yields, lower borrowing rates, and crypto cashback - start here.

THE MOST EXPENSIVE ALL-IN IN BUSINESS HISTORY

There's a scene near the end of Jurassic Park where John Hammond is sitting alone in a darkened restaurant, eating bowls of melting ice cream.

The power is out, the dinosaurs are loose, and everything is falling apart.

Ellie Sattler looks at him and says, "You never had control. That's the illusion."

Hammond takes a spoonful of ice cream and replies, "Spared no expense."

It's the most devastating line in the movie because the whole film's subtext is that he didn't spare no expense on what actually mattered. 

Nedry, the single person running the park's entire computer network, was underpaid and resentful, and no backup generators in a hurricane zone. 

The park had never even completed a safety inspection.

He spent lavishly on the tour cars, the food, the visitor center, and skimped on the thing, keeping the whole operation from collapsing.

On March 31, 2026, OpenAI channeled its inner Hammond.

They closed what the Wall Street Journal called "Silicon Valley's largest-ever funding round," $122B in committed capital at an $852B valuation.

Amazon pledged up to $50B, Nvidia committed $30B, SoftBank co-led the round, and Microsoft's historical contributions had already exceeded $13B. 

And in a move that is genuinely unprecedented for a pre-IPO company, OpenAI also pulled in over $3B from individual retail investors through banking institutions and announced plans to land in multiple ARK Invest ETFs managed by Cathie Wood.

That last part deserves a double-take. 

Normal companies don't open their private rounds to retail before going public. OpenAI is essentially building a public market proxy before the actual public listing. 

They're letting your financial advisor into the party early, and these are not passive investors chasing a return multiple. 

Amazon needs OpenAI's models running on AWS, Nvidia needs OpenAI consuming its GPUs at scale, and SoftBank needs a centerpiece for its Vision Fund III narrative. 

Every single one of these checks is a strategic infrastructure lock-in.

The move nobody saw coming

So you just raised $122B. What do you buy next? How about a talk show?

Two days after closing the largest private funding round in Silicon Valley history, OpenAI acquired TBPN, a daily three-hour tech livestream run by John Coogan and Jordi Hays that has become required viewing for Silicon Valley insiders. 

Guests have included Sam Altman, Mark Zuckerberg, and White House tech policy officials. 

Google's Gemini, a direct OpenAI competitor, was literally one of its advertisers.

The price was undisclosed, but rumors suggest it exceeded $150M, the message, however, was clear.

Fidji Simo, CEO of Applications at OpenAI, announced it internally with a line that tells you everything: “The standard communications playbook just doesn’t apply to us.”

Translation: OpenAI doesn't want to be covered by the media anymore, they want to be the media.

The show will sit under OpenAI's strategy organization, reporting to chief global affairs officer Chris Lehane. 

Simo promises "editorial independence"; Coogan and Hays will keep booking their own guests and making their own editorial calls. 

Sam Altman himself said he doesn't "expect them to go any easier on us,” which is exactly what you'd say if you just bought the show that covers you.

Look, the cynical read is obvious. 

OpenAI just purchased a credibility wrapper, a show that has interviewed competitors, criticized the industry, and built its entire reputation on independence. 

The sponsor banner disappeared from the stream the same day the deal was announced.

But here's the non-cynical read: OpenAI is about to IPO.

They are trying to become the operating system for global AI, and they just realized that owning the distribution layer for the conversation about AI is as strategically important as owning the compute layer.

Which brings us to the part that should make your brain short-circuit.

Around the same time they closed this monster round and bought a talk show, they killed Sora, their viral AI video app the entire internet was obsessed with, and started dramatically narrowing the products they're willing to build at all.

How does that make any sense?

Here's the thing most people misunderstand about the AI arms race: You cannot just buy compute on demand. 

And the market is already reflecting it, H100 rental prices are rising again, back to Q3 2024 levels, meaning compute is getting more expensive, not cheaper.

Lead times for data-center GPUs are running 36 to 52 weeks right now. 

You can be willing to pay any price in the world, and Nvidia still can't hand you a rack of chips tomorrow morning. 

TSMC's advanced packaging capacity is a hard physical ceiling. High-bandwidth memory is in a structural undersupply that cannot be resolved by throwing money at the problem.

So OpenAI raised $122B, and that money was essentially pre-committed before the wire transfer cleared. 

Every dollar is already allocated to chips, data center buildouts, and model training. 

According to Business Insider, OpenAI's internal teams hold daily triage meetings to decide which projects receive GPU time that day.

Think about what that means. The most-funded AI company in human history, sitting on $122B, is rationing compute like a college student rationing ramen during finals week.

That's why Sora got killed, video generation is 10 to 100x more GPU-demanding per output than text. 

The head of Sora, Bill Peebles, said it: "We have been quite amazed by how much our power users want to use Sora, and the economics are currently completely unsustainable."

Translation: people loved Sora, but it was financially ruinous to run. 

Every hour of GPU time keeping Sora alive was an hour stolen from ChatGPT and Codex products with a clear revenue path into the expected 2026 IPO, so they killed the thing everyone loved to fund the thing Wall Street wants.

Under Fidji Simo, OpenAI is executing a hard pivot, building only what is AGI-critical or revenue-generating and killing everything else. 

This is the rational response to three simultaneous pressures:

Hammond spent lavishly on everything visitors could see and skimped on the things that kept the dinosaurs in their cages.

OpenAI just raised the most money in Silicon Valley history. Bought a talk show and still can't run all their products.

THE MOST EXPENSIVE ALL-IN IN BUSINESS HISTORY (P2)

Here's where things get genuinely interesting for anyone who owns a 401(k).

Within the next 12 to 18 months, the public markets are expected to absorb three of the most expensive IPOs in history, simultaneously:

The combined float from these three, depending on the percentage of shares offered, could realistically demand billions in fresh public market capital. 

At the same time, 2026 has already opened with what analysts are calling a great rebalancing, where investors are rotating out of high-multiple tech into value, utilities, and healthcare.

So where does all the money come from?

The most natural answer, it may get rotated out of the very Magnificent Seven stocks that underwrite the rest of the tech sector. 

Passive index funds that must trim existing positions to make room for newly added constituents, growth hedge funds rotating out of illiquid secondaries into newly liquid IPO shares. 

Norway's GPFG, GIC Singapore, and Abu Dhabi's MGX, all of whom have already participated in OpenAI and Anthropic private rounds, will be able to absorb some allocation, but their reallocation is slow and heavily governed. 

And retail ETF inflows will amplify demand through index inclusion the moment these companies list publicly, since retail demand is already seeking AI exposure through any available channel.

The risk is the feedback loop. 

Selling existing Big Tech positions to fund the IPOs depresses the very stocks that have been underwriting the broader tech sector's valuations. 

Microsoft, whose cloud backlog is nearly half tied to OpenAI commitments, has already seen $440B wiped from its market cap on concerns about whether OpenAI can deliver on its obligations. 

That's before a single share of any of these three companies has traded publicly.

Tech CapEx as a percentage of U.S. GDP has now hit 7.2%, exceeding both the dot-com peak of 6.4% and the post-pandemic work-from-home surge of 7.1%.

BCA Research's chief strategist has warned that the AI investment splurge will likely hit empirical limits by the end of 2026.

If that timing coincides with three mega-IPOs demanding market attention simultaneously, the collision could not be more delicate.

The David vs. Goliath nobody saw coming

Now let's talk about the most underreported story in AI: Anthropic is beating OpenAI on every metric that actually matters for long-term survival.

OpenAI sits at an $852B valuation while Anthropic closed a $30B Series G in February 2026 at $380B, up from $61.5B just twelve months ago. 

The gap looks massive, but look under the hood, and it gets very complicated very fast.

OpenAI is projected to burn $143 to $150B before reaching profitability, which isn't expected until 2029 or 2030. 

Anthropic expects to reach positive cash flow by 2027 or 2028, burning roughly $20B to get there. 

There's also a sneaky accounting distortion buried in the headline revenue numbers. 

OpenAI reports revenue on a net basis, deducting the roughly 20% revenue share paid to Microsoft before reporting a number.

Anthropic reports revenue on a gross basis, including the hyperscaler revenue share in its top line before it flows out as an expense. 

Bank of America analysts projected in March 2026 that Anthropic could owe cloud providers $6B+ in revenue-sharing agreements this year alone. 

Shift that to net reporting, and Anthropic's headline ARR drops materially.

On an apples-to-apples basis, the revenue gap between these two companies is almost certainly larger than the headlines suggest.

And yet Anthropic's gross margin is projected at 50% in 2025 and 77% by 2028, compared to OpenAI's roughly 33%. 

Since both companies crossed $1B in annualized revenue, Anthropic has grown at 10x per year while OpenAI has grown at 3.4x. 

Anthropic went from $1B to $9B over 2025 and hit $19B by March 2026.

Over 300,000 businesses now use Claude, with 80% of revenue coming from enterprise and API channels. 

Claude Code alone hit an estimated $2.5B run-rate by early 2026, with SemiAnalysis projecting it could account for more than 20% of all daily GitHub commits by year's end.

Inside Anthropic, the numbers are just as striking. 

Claude Code's codebase is 90% AI-written, which is proof of compounding capability. 

Weekly active users doubled since January of this year, while business subscriptions quadrupled in the same period. 

Claude Cowork, their full desktop automation product, was built in approximately a week and a half, largely using Claude Code itself. 

Anthropic engineers report 200% productivity growth by internal measures, with 59% of all employee work now involving Claude. 

When they launched the Computer Use feature, allowing Claude to navigate computers autonomously, it triggered another subscription surge.

OpenAI, meanwhile, is scaling headcount, building a super app, and holding daily GPU triage meetings.

Two different bets on the same future

OpenAI is betting on becoming the next iOS. 

Their super app strategy merges ChatGPT, Codex, and the Atlas browser into one unified desktop experience with third-party app integrations across Expedia, DoorDash, and Uber, plus e-commerce and advertising potential baked in. 

It's the iPhone App Store playbook applied to AI.

Their GPT-5.3-Codex model, described internally as "our first model instrumental in creating itself," is a genuine milestone in recursive AI development. 

Sam Altman has characterized AGI as imminent and frames OpenAI's capital requirements as proportionate to the size of that prize.

If the consumer AI platform becomes the next operating system, OpenAI is positioned to own it. 

They have 900M weekly active users, providing a continuous feedback loop for model improvement that no competitor can match.

Meanwhile, Anthropic went deep on enterprise instead. 

Rather than chasing consumer mindshare, they pitched Claude as infrastructure for developers and businesses, and it’s working.

Here's the honest take on who wins: OpenAI is more likely to reach AGI first, not because it's smarter, but because it has structured access to more compute, more capital, and more real-world usage data than any other entity on Earth. 

But Anthropic is the more likely company to still exist and thrive in the aftermath, because its capital efficiency and profitability trajectory give it structural resilience that OpenAI currently lacks.

So you've got one company burning roughly $150B to own the consumer layer, and another burning $20B to own the enterprise layer. Both are pre-IPO and pre-profitable. 

And both are about to ask the public markets to price a future that hasn't happened yet.

The verdict

Is this the next internet? Or the largest capital bubble ever created?

Probably both, and that's the point.

The underlying technology is real. AI is already capturing 61% of all global venture capital. 

Enterprises adopting AI are seeing cash-flow margin expansion at 2x the market average. 

The hyperscalers are cash-flow-positive and directing real capital into infrastructure with real business rationale. 

But the pricing of that transformation? $852B for a company losing $14B annually is 42x annualized revenue, far beyond even the most generous software multiples. 

$380B for a company that was worth $61.5B twelve months ago and a $1.75T SpaceX IPO. 

These reflect a classic bubble dynamic, the present value of a genuinely transformative future, discounted at near-zero rates of skepticism. 

When every dollar of future AI value is already priced in today, the only way forward is flawless execution. 

Any shortfall, slower monetization, and compute cost increases, regulatory intervention reprices the entire sector simultaneously through index reweighting and sector-wide multiple compression.

The railroad got built. The Internet was built. The infrastructure being laid right now is real and permanent. 

But the history of technology says the infrastructure winners survive, and the pioneer companies that rushed to IPO at peak valuations often didn't.

The market will spend the next decade sorting out which of these AI companies was Amazon and which was Pets(.)com, and the answer will not be obvious until long after the capital has already been deployed.

The build is happening regardless, and the only question is who owns it when the dust settles.

Alright, that's it for this edition of Milk Road AI. We want to hear from you.

BITE-SIZED COOKIES FOR THE ROAD 🍪

Microsoft launched three new AI models for text, voice, and images. The move strengthens its push to compete with OpenAI and Google.

Tesla’s cheaper EVs aren’t boosting sales, with Q1 deliveries missing expectations. Growth remains weak as the company lacks a new mass-market model.

Truecaller has reached 500M monthly users globally. Growth is driven by demand for spam and scam protection.

MILKY MEMES 🤣

ROADIE REVIEW OF THE DAY 🥛