Follow the Money: Amazon’s $200B AI Spend by the Letter, OpenAI’s $100B Ad Target for 2030, SpaceX’s $5B AI loss pre-IPO & More.

MP’s Key Takes Today:

1. Jassy’s $200B AI capex isn’t a bet — it’s a declaration. Amazon is building a chip empire inside a cloud company. Custom silicon at $20B+ annualized, triple-digit growth. That’s the story easily missed.

2. OpenAI just declared Ad Ramp-up. Up from $2.5B in ad revenue this year, projecting $100B by 2030. Anthropic ran a Super Bowl ad saying “no ads ever.” That’s the defining fork of the two AI Sibling Rivals.

3. SpaceX posted ~$5B in losses from AI/xAI spending — heading into a June IPO at $1.5T+. When a rocket company loses money on AI, you know the AI infrastructure is included.

Jassy’s $200B Declaration

Andy Jassy dropped his fifth annual shareholder letter this week, and the headline number is the highest amongst peers: $200B+ in capex for 2026, the lion’s share going to AI infrastructure. That’s roughly 4x what Amazon spent in 2023, and the biggest single-year infrastructure bet in tech history. GeekWire summed it up perfectly in their headline: “Not on a hunch.”

The headlines focused on the number. The story less understood is the chip business hiding underneath it.

Graviton, Trainium, and Nitro — Amazon’s custom silicon stack — are now generating over $20B annualized, growing triple-digits year over year. Trainium3 is shipping and nearly fully subscribed. Jassy hinted in the letter at a $50B external chip business. Amazon isn’t just a cloud company pouring money into GPUs. Amazon is becoming an AI chip company hiding inside a cloud company. And Jassy took direct aim at Nvidia, Intel, AND Starlink in the same letter — that’s full-stack competitor positioning, not polite coexistence.

There’s a second signal I want to flag: AWS-backed Anthropic just hit $30B ARR and employees refused to sell at the $350B tender offer that closed this week. They’re betting on $500B+ at IPO. When insiders won’t sell, that’s the strongest signal you can read. (AI-RTZ #999, #1007, #1048)

MP angle: Amazon always builds for itself first, then sells to everyone. That’s exactly how AWS started — internal infrastructure that became an industry. Having covered the internet wave at Goldman, this is the Intel-in-the-80s playbook, except Amazon already owns the cloud distribution. Watch this space.

OpenAI gets in line on Ads behind Google and Meta

Axios broke this exclusively today: OpenAI is on track for $2.5B in ad revenue this year, with an internal ramp projecting $11B in 2027, $25B in 2028, $53B in 2029, and $100B by 2030. The ad pilot hit $100M ARR in under two months.

That growth curve isn’t startup-scale. It’s Google-scale.

The pitch to advertisers is seductive: chatbot ads beat search and social ads because users literally tell you what they want. No inferring intent from cookies — they just say it out loud. And the model assumes 2.75B weekly users by 2030, which is consumer-internet-platform ambition on par with Facebook’s best years.

But here’s the tension: ads undermine AI’s core promise. The whole pitch for an AI assistant is that it works for YOU, not for advertisers. This is the original sin of the consumer internet, repeating itself with a different logo.

Which is why Anthropic’s Super Bowl ad earlier this year — the one promising Claude stays ad-free forever — looks less like branding and more like strategic positioning. OpenAI = scale + ads. Anthropic = premium + trust. Two companies, two opposite bets, and the fork is now visible to everyone. This will be the defining strategic divergence of the AI era, and it’s going to matter enormously when OpenAI starts its IPO roadshow with three revenue legs: subscriptions, API, and ads.

MP angle: This is Facebook circa 2012. Build the user base, monetize with ads. The open question is whether AI users will tolerate ads the way social media users did — or whether the intimacy of conversational interfaces makes ads feel qualitatively different. Anthropic just handed itself the anti-ad positioning on a silver platter. (AI-RTZ Ep 42, #1048)

When a Rocket Company Loses Money on AI

The Information reported today that SpaceX posted roughly $5B in losses last year, driven almost entirely by AI/xAI spending post-merger. Pre-merger SpaceX was doing ~$8B profit on $15-16B revenue. Post-merger, xAI’s ~$1B-per-month burn rate has dragged the whole entity into the red.

xAI’s quarterly revenue is around $107M. Against billions in losses. That’s not a business — that’s an AI research lab stapled to a rocket company.

And this is the entity heading into a June IPO targeting $1.5-1.75T+ valuation, with banks pushing a $50B raise. Elon is pushing for 30% retail allocation — three times the typical 5-10%. (Another mega meme stock besides Tesla anyone? When insiders want retail investors to carry that much of the float, ask why. And they’re working hard to push index funds to include it early into passive investment funds with no typical 4+ quarter seasoning as public companies. Force-feeding the Geese for foie gras. And We are the Geese.

Meanwhile, Jassy mentioned Amazon Leo today — a direct competitor to SpaceX Starlink ambitions. All backed by Prime’s 200M+ households and AWS.

MP angle: Called the SpaceX/xAI merger math back in RTZ #987 and #1007. The merger was always about xAI needing SpaceX’s cash, not the other way around. xAI is burning $1B/month competing against Anthropic ($30B ARR), OpenAI (now a profitable ad business), and three hyperscalers spending $200B+ each. The $5B loss is the headline. The real question for June IPO investors: is xAI’s AI business EVER going to rank amongst the top companies? Or is Elon selling a space company at an AI company’s valuation? (AI-RTZ #987, #1007, #1048)

Also Worth Noting

The AI gadgets race is heating up as a sideshow to all this infrastructure spending. Meta’s Ray-Ban glasses now have 3M+ installed base, selling faster than they can ship, with facial recognition “Name Tag” features coming. Google’s Android XR glasses are back from the Google Glass grave with Warby Parker and Samsung partnerships. Apple has three AI wearables in development and 2B+ iPhones as a distribution base when they’re ready. OpenAI and Jony Ive are pushing the boldest swing — a screen-free, behind-the-ear device codenamed Sweetpea, now delayed to 2027+. Market projection: 3.3M → 13M units by 2026. The post-smartphone form factor war is on, and we’re still in inning one. (AI-RTZ #487, #728, #849, #863, #866, #926, #944, #1002)

And on the public-sentiment side, AI keeps scoring as the most negatively received tech wave in my three plus decade tech career. Nine distinct fear factors (variable utility bill shock, cognitive degradation, copyright theft, misinformation, environmental guilt, always-listening devices, dependency, political vacuum, and latency speeds) are all hitting simultaneously.

I’ve covered many tech waves, and this one IS different. AI skipped the “harmlessly mediocre” phase. It went from gimmick to “could replace me” in six months.

Which is the bigger red flag — OpenAI putting ads in your AI, or SpaceX losing $5B on AI before going public? Drop your take in the comments.

Subscribe to AI-RTZ for the daily Substack and follow AI Ramblings Daily on YouTube for the video breakdown every morning.

That’s it for today’s AI Ramblings Daily, #52. Stay tuned.

Prefer shorter cuts? Here’s the buffet.

Not in the mood for the full 21-minute episode? Pick your attention span preference. One 5-minute deep-dive on the AI glasses race, plus five bite-size clips on the sharpest moments from today’s show.

🥽 Deep dive: AI Glasses — Hype vs. Reality (5 min)

AI Glasses: Hype vs. Reality — Six Companies, Six Strategies, One Race

Meta, Google, Apple, OpenAI + Jony Ive, Amazon, Snap — six different bets on the post-smartphone form factor. Who’s leading, who’s following, who has the boldest swing, and what could kill the whole category. The 5-minute version of the gadgets conversation from today’s full episode.

⚡ Quick hits (under 1 min each)

9 Reasons AI Is the Most Feared Tech Wave Ever (0:48) Why is AI the most negatively received tech wave of my career? It’s not just jobs. Nine distinct fears — utility bills, cognitive decline, creative theft, always-listening devices, raw speed — all landing at once.

Why AI’s Open-Ended Use Cases Are the Real Adoption Problem (0:42) The hardest thing about AI isn’t the tech. It’s that AI can do almost anything — so users don’t know what to ask for. We don’t have the vocabulary yet. That’s the real adoption curve.

Are AI Glasses Actually the Next Big Thing? Six Companies Say Yes (0:57) The 57-second version of the AI glasses race. Meta ships 3M+ Ray-Bans. Google is back with Android XR. Apple plays the long game. OpenAI + Jony Ive made the boldest swing of all.

The AI Early Adopter Divide Nobody’s Talking About (0:53) The AI divide isn’t age or income — it’s curiosity. Early adopters are already running circles around everyone else. The gap will look like the 1996 internet gap, but much faster.

Meta’s Ray-Bans: Meta’s Real Shot at Mainstream AI (0:57) Meta’s Ray-Ban glasses are selling faster than they can make them. Facial recognition is coming. THIS is Meta’s real mainstream AI play, not the Muse Spark model.

MP’s Key Takes runs daily on AI Ramblings Daily — three differentiated calls on the day’s biggest AI news. Track them over time. See how many become conventional wisdom.

Follow MP on X/Twitter at

(NOTE: The discussions here are for information purposes only, and not meant as investment advice at any time. Thanks for joining us here)