OpenAI's revenue just shrank by $20 billion. Here's why.

Good morning and happy Sunday, crowd. ☕

This was the week the AI boom got its first real bill. One revenue number shrank by $20 billion overnight. One of the world's most valuable companies asked to borrow $40 billion, and its credit risk hit a record. And two big banks said AI will replace a big chunk of their staff.

Quick game first. 🎯

OpenAI just told investors its annualized revenue is "approaching" how much?

• 🅰️ $70 billion

• 🅱️ $50 billion

• 🅲️ $100 billion

Answer in story #1.

📋 Today's 60-second summary

• 💸 OpenAI's revenue reset: about $20B less than the number investors were passing around

• 🏦 AI runs on debt now: SpaceX wants to borrow $40B, and its credit risk just hit a record

• 🇨🇳 Manus raised $500M after China forced Meta to undo a $2B acquisition

• 🤖 AI comes for finance jobs: HSBC and DNB are cutting advisers and staff

• 🛒 Shopping agents: Meta, Walmart and Stripe want new rules. Amazon isn't playing.

⏱️ Read time: 6 minutes

DIT AgTech Doesn’t Need to Sell Hardware to Make Money.

One installation can generate years of recurring supplement revenue. 

provides the hardware free to ranchers through a three-year nutrition plan. The hardware gets installed once. The supplements keep generating revenue year after year. 

Then there’s the second paycheck. 

Methane reduction is becoming an economic opportunity for agriculture. 

Methane-reducing nutrition create carbon credits, giving ranchers another income stream while creating an additional revenue opportunity for . 

That means one connected livestock operation generates recurring supplement revenue, with carbon revenue as additional upside. 

We win when ranchers win. 

The model is already working in Australia, one of the world's toughest livestock environments, with 500+ units deployed and 370,000 head on the platform. 

Now DIT AgTech brings that proven model to the massive U.S. livestock market. 

For investors, the opportunity is getting in before the model scales into its next market.

^{𝘐𝘯 𝘮𝘢𝘬𝘪𝘯𝘨 𝘢𝘯 𝘪𝘯𝘷𝘦𝘴𝘵𝘮𝘦𝘯𝘵 𝘥𝘦𝘤𝘪𝘴𝘪𝘰𝘯, 𝘪𝘯𝘷𝘦𝘴𝘵𝘰𝘳𝘴 𝘮𝘶𝘴𝘵 𝘳𝘦𝘭𝘺 𝘰𝘯 𝘵𝘩𝘦𝘪𝘳 𝘰𝘸𝘯 𝘦𝘹𝘢𝘮𝘪𝘯𝘢𝘵𝘪𝘰𝘯 𝘰𝘧 𝘵𝘩𝘦 𝘪𝘴𝘴𝘶𝘦𝘳 𝘢𝘯𝘥 𝘵𝘩𝘦 𝘵𝘦𝘳𝘮𝘴 𝘰𝘧 𝘵𝘩𝘦 𝘰𝘧𝘧𝘦𝘳𝘪𝘯𝘨, 𝘪𝘯𝘤𝘭𝘶𝘥𝘪𝘯𝘨 𝘵𝘩𝘦 𝘮𝘦𝘳𝘪𝘵𝘴 𝘢𝘯𝘥 𝘳𝘪𝘴𝘬𝘴 𝘪𝘯𝘷𝘰𝘭𝘷𝘦𝘥. 𝘋𝘐𝘛 𝘈𝘨𝘛𝘦𝘤𝘩 𝘩𝘢𝘴 𝘧𝘪𝘭𝘦𝘥 𝘢 𝘍𝘰𝘳𝘮 𝘊 𝘸𝘪𝘵𝘩 𝘵𝘩𝘦 𝘚𝘦𝘤𝘶𝘳𝘪𝘵𝘪𝘦𝘴 𝘢𝘯𝘥 𝘌𝘹𝘤𝘩𝘢𝘯𝘨𝘦 𝘊𝘰𝘮𝘮𝘪𝘴𝘴𝘪𝘰𝘯 𝘪𝘯 𝘤𝘰𝘯𝘯𝘦𝘤𝘵𝘪𝘰𝘯 𝘸𝘪𝘵𝘩 𝘪𝘵𝘴 𝘰𝘧𝘧𝘦𝘳𝘪𝘯𝘨, 𝘢 𝘤𝘰𝘱𝘺 𝘰𝘧 𝘸𝘩𝘪𝘤𝘩 𝘮𝘢𝘺 𝘣𝘦 𝘰𝘣𝘵𝘢𝘪𝘯𝘦𝘥 𝘩𝘦𝘳𝘦: https://bit.ly/4bzuWCi​}

NEWS OF THE WEEK

⚡ TL;DR

• OpenAI told investors its annualized revenue is approaching $50B, about $20B below the figure circulating a week earlier

• The gap comes down to how revenue is counted, not a collapse in demand

• AI stocks tied to OpenAI, including Oracle, CoreWeave and Nvidia, sank on the news

🧠 Quiz answer: It was 🅱️, about $50 billion.

A week ago, investors were passing around a much bigger number. OpenAI's annualized revenue was reportedly approaching $70 billion, which would have put it neck and neck with Anthropic, last reported at a $65 billion run rate.

Then the Financial Times reported on Thursday that OpenAI had told investors the real figure is approaching $50 billion.

So where did $20 billion go? According to TechCrunch, the $70 billion figure came from investors trying to compare OpenAI directly with Anthropic. The two companies count revenue differently. Anthropic counts sales made through its cloud partners, Amazon and Google. OpenAI doesn't.

If that sounds familiar, it should. Last Sunday we covered Anthropic's leaked IPO filing, which showed 47% of its sales flowing through those cloud partners. OpenAI has argued that this way of booking revenue inflates the topline. This week, that argument stopped being an accounting footnote and started moving stock prices.

Here's what's wild. OpenAI is reportedly in early talks to raise $30 billion or more at a $1.4 trillion valuation. At $50 billion in annualized revenue, that's roughly 28 times sales. At $70 billion, it would have been 20 times. Same company, very different story.

And sentiment is shifting. When Axios's Dan Primack asked about 30 Bay Area founders which shares they'd rather own, two-thirds picked Anthropic, one-third picked SpaceX, and nobody picked OpenAI.

🎯 What to do with this:

• Investors: when you compare AI companies, ask whether revenue is counted gross or net. A $20 billion gap can hide in that one word.

• Founders: report your numbers the conservative way. A run rate that has to be walked back costs you more trust than a smaller number ever would.

🗳️ POLL: Which would you rather own today?

• 🤖 Anthropic

• 💬 OpenAI

• 🚀 SpaceX

⚡ TL;DR

• SpaceX is in talks to borrow about $40B for Nvidia chips, and its credit risk hit a record

• Broadcom is lining up $50B+ to finance custom chips for OpenAI

• With the 10-year Treasury yield above 5%, borrowing has never cost the AI boom more

Last year, the AI buildout was paid for mostly with cash and equity. This week made it clear that debt is now doing a lot of the work.

SpaceX is in early talks to raise about $40 billion: $10 billion in bank loans and $30 billion in investment-grade debt, with Apollo expected to lead, according to the Financial Times and Bloomberg. The money would reportedly buy Nvidia AI chips. SpaceX already had $38.4 billion of debt at the end of June, after a $25 billion bond sale alongside its IPO. If this package closes, that roughly doubles.

Bond investors noticed. The cost of insuring SpaceX debt for five years hit a record of about 197 basis points on Wednesday. In plain English: protecting $10 million of SpaceX debt now costs about $197,000 a year.

🔢 The rest of the AI debt pile:

• Broadcom is arranging a package of more than $50 billion to finance custom chips it builds with OpenAI, with the chips themselves as collateral

• Apollo and Blackstone are reportedly in talks to lend

• Oracle is negotiating its own deal

• The 10-year Treasury yield sits above 5% and briefly crossed 5.35% on Thursday, when the Nasdaq fell 1.25%

Billionaire investor Ray Dalio said this week that AI looks like a classic bubble that is "close" to bursting. His triggers: heavy borrowing for AI infrastructure and rising rates.

The IPO market is feeling it too. Deals priced at the top of their range are holding up, but Accelevation and Orion180 both priced below their ranges this week. And the biggest test is still ahead: Bloomberg reports Anthropic could start marketing its IPO the week of November 9.

🎯 What to do with this:

• Founders: with rates above 5%, every dollar of venture debt costs more. Model your loans at today's rates, not last year's.

• Investors: watch credit spreads, not just stock prices. Bondholders usually spot trouble first.

Traded VC

⚡ TL;DR

• AI agent startup Manus raised more than $500M, its first round since its sale to Meta was undone

• Meta agreed to buy Manus for $2B in December 2025. Beijing ordered the deal unwound in April

• Manus is now reportedly considering a Hong Kong IPO

Imagine selling your company for $2 billion, then having a government make you give the money back.

That's what happened to Manus, the AI agent startup founded in China and now based in Singapore. Meta agreed to buy it for $2 billion in December 2025. In April 2026, Beijing ordered the deal unwound, citing concerns about losing AI talent to Western companies. Manus went back to operating independently in August, and had to delete some user data as part of the separation.

This week it showed it can stand on its own. Manus raised more than $500 million, co-led by Boyu Capital and IDG Capital, with existing investors Tencent, HSG (formerly Sequoia China) and ZhenFund joining. The valuation wasn't disclosed, but it was reportedly in talks last month to raise at about $4 billion, double what Meta agreed to pay.

Manus was already generating more than $100 million in annual recurring revenue when it agreed to the Meta deal. Now it's reportedly weighing a listing in Hong Kong.

Here's the bigger lesson. Most founders think of an exit as a deal between a buyer and a seller. Manus shows there can be a third party at the table: a government. Where your company was founded, where your team lives and where your data sits can decide who is allowed to buy you.

🎯 What to do with this:

• Founders: your legal home is part of your exit strategy. Think about it early, before a buyer shows up.

• Investors: when a cross-border deal looks great, ask who could block it. Regulatory risk now cuts both ways between the US and China.

⚡ TL;DR

• HSBC plans deep cuts in its UK wealth business as it moves to AI, according to the FT

• Norway's DNB is cutting about 400 jobs and pointing to agentic AI

• Even VCs are changing: USV's Fred Wilson says 20–30% of the firm's legal work no longer involves a human lawyer

For years, "AI will take jobs" was a prediction. This week, banks started putting numbers on it.

HSBC plans widespread cuts in its UK wealth management business as it shifts toward AI, according to the Financial Times. About 70% of its financial advisers and roughly half of its managers and specialists are affected, with departures expected by the end of October.

In Norway, DNB said it will cut about 400 full-time positions, roughly 4% of its workforce, mostly in technology and services. Its reason was blunt: agentic AI now handles tasks people used to do by hand.

And venture capital isn't exempt. Union Square Ventures just raised $900 million: a $500 million core fund, up from $275 million in 2024, plus a $400 million opportunity fund. It cut its active partners to four. Co-founder Fred Wilson says 20–30% of USV's own legal work no longer involves a human lawyer. The firm's new motto for startups: "Obliterate, don't automate."

But here's the thing. The jobs disappearing first aren't factory jobs. They're well-paid office jobs built on reviewing documents, giving standard advice and moving information between systems. That is exactly where many AI startups are now pointing their products.

🎯 What to do with this:

• Founders: "Obliterate, don't automate" is the pitch investors want to hear. Don't sell a tool that saves a team 10%. Sell the result the whole team used to deliver.

• Everyone: the safest skills are judgment, relationships and decisions. Clients may still want a human for those, even as banks cut the rest.

🗳️ POLL: Would you trust an AI with your investment advice?

• ✅ Yes, if it's cheaper and good enough

• 🤝 Only with a human checking it

• ❌ No, I want a person

⚡ TL;DR

• Meta and Sierra proposed a new standard for how AI agents sign in and act at stores, backed by Walmart and Stripe

• Amazon cut off Meta's Muse agent last month

• The fight is over who controls the customer when an AI does the shopping

AI agents can now browse, compare and check out for you. The problem: nobody agreed on the rules.

On Tuesday, Meta and Sierra, the AI customer service company co-founded by OpenAI board chair Bret Taylor, proposed the Personal Agent Protocol. Partners include Walmart and Stripe. The idea is simple: give AI agents a standard way to sign in at a business, with customers choosing read-only or full access and businesses setting limits on what agents can do. A first version of the spec is due later this month. "It is kind of chaos until such a standard exists," Taylor said.

The biggest store in America isn't on board. Amazon cut off Meta's Muse agent on September 20, saying Muse didn't identify itself while browsing and appeared to capture customer credentials. Meta says Muse stores credentials securely and asks before sensitive actions like buying.

🔢 Why this fight matters:

• AI-sourced traffic to US retail sites is up 393% this year, according to Adobe

• Morgan Stanley projects agents could influence 20% of US e-commerce by 2030

• But only 3% of US adults say they'd trust an agent to complete a purchase, per an NMI survey

Here's the real stake. Whoever controls the agent controls the customer relationship. If your AI picks the store, the store loses the shelf space, the ads and the upsell. That's why Amazon wants agents playing by its rules, and why everyone else wants an open standard.

🎯 What to do with this:

• Founders in commerce: start treating AI agents as a sales channel. Make your products, prices and policies easy for an agent to read.

• Investors: the trust gap is the opportunity. The startups that make agents safe to pay with, through identity, permissions and fraud checks, could be the biggest winners.

⚡ Quick Hits

• 🏭 Schneider Electric agreed to buy PTC for about $22.6B in cash, or $205 a share, a 42% premium

• 🏟️ Arena raised a $200M Series B at a $3.1B valuation, co-led by Lightspeed and Khosla

• 🌱 Hone, founded by former Cognition and OpenAI staff, raised a $60M seed led by Benchmark and Index

• 🦾 Mecka raised a $60M Series B led by Sequoia to collect human-motion data that trains robots

• ⚛️ Universal Quantum raised more than $100M in a Series A co-led by DCVC

• 🧬 DeepMind spinout Isomorphic Labs is reportedly in early talks at a valuation of $40B or more

• 🇪🇺 Mistral previewed Large 4, a 1-trillion-parameter model, with open weights due October 27

• 🛰️ The FCC approved up to 15,000 SpaceX direct-to-cell satellites

• 🔐 Cybersecurity startups raised about $5.26B in Q3, nearly double the prior quarter, per the WSJ

• 🇵🇭 Mynt, the parent of GCash, priced its IPO at a valuation of about $7B, set to be the biggest ever on the Philippine Stock Exchange

❓ Quick answers

What is OpenAI's annualized revenue?

OpenAI told investors in October 2026 that its annualized revenue is approaching $50 billion, according to the Financial Times. That is about $20 billion below a roughly $70 billion figure that had circulated a week earlier, a gap linked to differences in how OpenAI and Anthropic count revenue sold through cloud partners.

Why is SpaceX borrowing $40 billion?

SpaceX is in early talks to raise about $40 billion, made up of $10 billion in bank loans and $30 billion in investment-grade debt, reportedly to buy Nvidia AI chips. Apollo is expected to lead. The cost of insuring SpaceX debt hit a record of about 197 basis points in October 2026.

Why did China block Meta's acquisition of Manus?

Meta agreed to buy AI agent startup Manus for $2 billion in December 2025. Beijing ordered the deal unwound in April 2026, citing concerns about losing AI talent to Western companies. Manus then raised more than $500 million in October 2026.

What is the Personal Agent Protocol?

The Personal Agent Protocol is a proposed open standard from Meta and Sierra, backed by partners including Walmart and Stripe, for how personal AI agents sign in and act at businesses. Customers choose read-only or full access, and businesses set limits on what agents can do.

💬 What This Week Tells Us

The bill for the AI boom is coming due, and it's arriving in four places at once:

• 💸 Revenue: the numbers are getting checked, and some are smaller than advertised

• 🏦 Debt: the buildout is now funded by borrowing, at the highest rates in years

• 🇨🇳 Politics: governments are deciding who can buy whom

• 🤖 Jobs: the savings AI promised are showing up as layoffs

None of this means the boom is over. It means the easy part is. From here, the winners will be the companies with real revenue, sensible debt and products people actually trust.

📣 Your turn

Is AI in a bubble? Hit reply with one word: yes, no or partly, plus one sentence on why. We'll share the best answers next week. 👀

📨 Know someone pricing an AI deal right now? Forward them story #1.

See how OpenAI, Anthropic and SpaceX stack up on our Private Company Index.

Hunt accordingly. ☕

Jose, Alberto & the Founderscrowd Team

Sources: TechCrunch and CNBC on OpenAI's revenue; Crypto Briefing on SpaceX's debt; Yahoo Finance on yields; CNBC on Manus; FinTech Futures on HSBC and DNB; Implicator on the Personal Agent Protocol; Tech Startups on this week's deals.

Disclaimer: Founderscrowd is a media company. This newsletter is for informational and educational purposes only and does not constitute financial, investment, or legal advice, or a recommendation to buy, sell, or hold any security. Private market investments are speculative, illiquid, and may result in the loss of your entire investment.