The bill and the keys: SpaceX borrows 40 billion for chips, Nous Research raises 90 million, ICANN reopens domain extensions

Three stories, one bill. A rocket company borrowing like a small country to buy chips. An open source agent that went viral with developers, raising 90 million dollars to sell to enterprises. And the internet's front doors, redrawn for the first time in fourteen years. That day, I talk about the bill and the keys.

Signal one: SpaceX wants to borrow 40 billion dollars to buy chips

The Financial Times reported on Tuesday that SpaceX is negotiating a 40 billion dollar raise to buy Nvidia AI chips. Reuters and Bloomberg confirm the outline: about 10 billion in bank loans and 30 billion in investment-grade debt. Apollo Global Management would lead, with bond giant Pimco among the lenders in talks. Talks are early and could still fall apart. If the deal moves forward, closing is expected in 2027.

I read the number slowly. Forty billion dollars, borrowed, to buy chips. One company, one order. It would be one of the largest debt financings ever tied to the AI rush. A semiconductor purchase starts to look like a national infrastructure project.

Why the appetite

SpaceX is no longer just a rocket company. Earlier this year, Elon Musk folded xAI into SpaceX, bringing the Grok chatbot and the Colossus supercomputers under one roof. Last month Musk said the Colossus 2 data center could more than double its Nvidia chip count by December. And on the August earnings call he said the company would build exclusively on Nvidia hardware, calling the Vera Rubin architecture "the best architecture."

SpaceX went public in June with a record 86 billion dollar IPO. The market has already priced in the ambition. This financing is the plumbing.

SpaceX is not alone at the bank

• According to Bloomberg, Broadcom's Wall Street syndicate is assembling 60 billion dollars of fresh AI chip financing, partly to benefit Anthropic.
• In August, Nvidia partnered with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR on financing platforms meant to mobilize more than 500 billion dollars for AI infrastructure.
• Morgan Stanley estimates the AI rush will need 1.5 trillion dollars of external financing by 2028.

What changes

The substance of AI competition has changed. The race has been fought on balance sheets for several years already, and this financing makes it visible. When one company borrows 40 billion to buy chips from a single supplier, the question for every other builder becomes: who can finance the furnaces?

A word on debt versus equity. Raising equity means selling part of the company. Borrowing means committing to repay, with interest, whatever happens. Financing chips with debt assumes those chips will produce enough steady revenue to service it. It is a bet on the future profitability of AI, carried by lenders who want their money back.

For brands downstream, concentration is the story: one dominant chip architecture, one dominant supplier, hundreds of billions in financing platforms. Your AI strategy is now exposed to credit markets, whether you know it or not. If credit tightens, the price of the AI you buy will move.

Signal two: the open source agent goes enterprise

The Wall Street Journal reports that Nous Research raised 90 million dollars at a 1.5 billion valuation. Robot Ventures led. Nvidia, M12 (Microsoft's fund) and Samsung joined, alongside Union Square Ventures, Menlo Ventures, 1789 Capital and others. Total funding is now about 158 million dollars.

The product

It is called Hermes, an open source AI assistant that went viral with developers this year. It browses the web, edits files, writes code and learns new skills from use. It can run remotely, around the clock, from Telegram or Discord.

• 22.7 million downloads since February.
• About 214,000 stars and nearly 40,000 forks on GitHub.
• According to OpenRouter, it has overtaken OpenClaw, the other open source agent phenomenon, in usage.

For readers who do not live on GitHub: a star is a mark of interest, a fork is a copy someone made in order to modify the project. Forty thousand forks means forty thousand teams or individuals who wanted to adapt the tool. That is a real usage signal.

The bet

The bet behind the round is the enterprise. Nous argues that companies will want open source architecture because it gives them control of their own data and less dependence on specific vendors. Chief executive Dillon Rolnick frames it as a principle: companies should have a level of ownership over their own destinies. And safety, in his view, comes from inspectable code that anyone can read, download and modify.

Two honest notes

Management forecasts annualized revenue above 100 million dollars by year end, up from 36 million in mid-September. That is the company's forecast. Actual numbers come later.

And the round itself is a sequel. TechCrunch reported in July that Nous was finalizing at least 75 million dollars at the same valuation. That round never closed. This is the closed version, larger.

What changes

The day before, I talked about Mistral selling sovereignty as availability: an open model nobody can switch off. Nous sells the same move in a different register: control of the agent running inside your company.

The skeptic's question is the revenue line. Open source projects generate enormous developer affection without capturing much money. Nous has to turn 22.7 million downloads into durable enterprise spending, from 36 to 100 million annualized, in three months. If it works, the open model proves it has an enterprise distribution channel. If it fails, the downloads were the whole story.

I see a classic brand lesson here. Free adoption builds awareness and preference. Converting it into revenue requires something else: support, guarantees, an accountable counterpart. What an enterprise really buys is someone to call when things break.

Signal three: the internet has new front doors

On October 7 at 18:00 UTC, ICANN, the body that manages the domain name system, published the list: 1,616 applications for new generic top-level domains (gTLDs, like .app or .shop). It is the program's second round, fourteen years after the first. In 2012 there were 1,930 applications.

The calendar matters now

• Until October 21: applicants can switch to a predefined replacement string.
• November 17, String Confirmation Day: ICANN publishes confirmed strings and contention sets, meaning who is fighting whom for the same word.
• Until March 16, 2027: community input and objections.
• First half of 2027: a draw sets the processing order.

The 2012 precedent

It is a cautionary tale. Only 20% of applications came from major brand holders. Contested words were expensive: .app drew thirteen competing applicants, .news seven, and conflicts were settled by negotiation or ICANN auction. Google applied for 101 strings, Amazon for 76. Apple, Microsoft, McDonald's and Nike applied. Coca-Cola, Facebook, Disney, eBay and Adidas sat out.

The "dot brand" experiment, a company owning its own extension like .apple, largely remained a curiosity. The promised "reset of the internet" never came. Most brands stayed on the sidelines, and the internet did not come to them.

What changes

For brand owners, this is the front door question. A domain is a piece of territory: where people find you, what they type, what they trust. Fourteen years ago, brands underplayed the first round and paid for it in defensive registrations and auctions.

This time the stakes are sharper, because the address space is about to collide with AI. Brand agents already answer inside ChatGPT. Agents pick merchants on people's behalf. And now, new front doors that anyone can apply for. An agent looking for your brand could land on an extension you do not control.

The first episode of this series already told how OpenAI named its agent dots while the dot.com domain, owned by xAI, redirected to Grok. The name is an asset. The domain is its lock.

What I would do this week, for any brand that matters:

1. Read the list of 1,616 strings and spot anything touching the brand name, its category or its flagship products.
2. Set up monitoring of the contention sets published on November 17.
3. Prepare an objection if needed before March 16, 2027.
4. Decide coldly whether a brand extension makes sense. For most brands, it does not. For those building an ecosystem of services or agents, the question deserves to be asked.

The cost of missing the round goes beyond the price of a domain. It means letting someone else open a branch under your name.

The thread of the day

SpaceX borrows like a country to own compute. Nous sells openness as the enterprise safety story. ICANN reopens the map of names.

Three different markets, one pattern: the infrastructure of the next decade is being claimed, in debt documents, open source licenses and domain extensions. The window is measured in months. Read the paper before it closes.

The noise filtered. The signal kept.