Cursor is training its own 1.5T-parameter model

THE AI BRIEF

Today's signal: Cursor announced a 1.5T-parameter frontier model at Compile, and OpenAI's leaked 2025 financials show what that bill looks like.

In today’s issue:

• Main story: Cursor is a frontier AI lab now

• Also worth knowing: OpenAI's leaked financials, Coinbase registers an AI agent as an SEC-regulated investment advisor, Odyssey raises $310M for world models, and Midjourney reveals its first hardware product

THE READ

The $60B SpaceX deal funded a 1.5-trillion-parameter model. Yesterday's news was the price tag. Today's is the strategy.

Cursor CEO Michael Truell took the stage at the company's Compile conference in San Francisco on Monday and announced that Anysphere is training a 1.5-trillion-parameter foundation model from scratch on more than 100,000 GPUs, the same compute scale Anthropic used for Claude Opus and OpenAI used for GPT-5.5. Truell described the model as "generally intelligent, not just coding," and said it would ship in weeks. The announcement landed the same day as the SpaceX acquisition. The two events were not a coincidence. The compute behind the new model is SpaceX compute.

Yesterday's framing of the deal as "industrial conglomerate buys developer tool" was the surface read. The underlying transaction is different. SpaceX is not buying a $60 billion coding product. SpaceX is funding Cursor's path to a frontier model, with the IDE as the distribution surface for whatever that model becomes. Truell's "generally intelligent" phrasing puts the company in direct competition with OpenAI, Anthropic, and Google's foundation-model arms. Twelve months ago, Cursor was an IDE wrapped around other people's models. This week, it announced it is one of them.

AI READY PRO · FREE UNTIL FRIDAY

I will get you AI trained in 30 days for free

After several months of running it quietly with top AI operators and teams, I’m excited to launch AI Ready Pro today to our newsletter subscribers like you.

It’s a 30-day personalized AI training program that is personalized to you and where you are in your AI journey. It’s the culmination of 100’s of hours spent teaching AI to top Fortune 500 AI teams and operators (and working with Mark Cuban).

Each day, you receive a 10-15 minute exercise to complete to improve your AI skills.

This isn’t vague AI theory or a way to pitch you a tool. It’s 30 days of learning to use AI in your real work. So you come out the other side with the skills to use AI to actually improve you and your teams output.

Until this Friday (May 15), it’s free for newsletter readers who complete the extended assessment. This assessment will help us personalize the learning experience to you.

If you lead a team, AI Ready Team is open today too, also for readers first. It’s the same engine, but to train your whole organization. You’ll get a detailed view of:

• Your team’s AI readiness level + detailed strengths & weaknesses

• Ranked list of AI automation opportunities

• Get tactical advice on how to advance AI efforts in your org

Leading a team? Take the Team assessment instead.

The economics of that decision are not a mystery. Ed Zitron published OpenAI's audited 2025 financials on Tuesday: $13.07 billion in revenue, $34 billion in total costs, $10.59 billion paid to Microsoft for infrastructure, $5.73 billion in sales and marketing, $25 billion in cash on the balance sheet against a burn rate the same filings describe as accelerating. Anthropic, by contrast, is projecting an operating profit in Q2 2026 on its current path. Frontier AI runs at one of two economic profiles right now, and Cursor just decided to enter the more expensive one. SpaceX writing the check is what makes that decision survivable.

What I keep coming back to is the second-order implication on engineering-tool standardization. If the IDE your team standardized on is now a frontier model lab, the vendor's roadmap is no longer set by coding-product economics. It is set by foundation-model economics, which means longer release cycles, larger version jumps, periodic capability surges, and a much higher floor on the price per seat once the model launches and the cross-subsidy from venture capital stops absorbing the inference bill. None of that is a reason to move off Cursor on Thursday. It is a reason to keep the procurement option-set wider than it used to be.

SPONSORED BY CLUTCH

Hire secure AI teammates that work 24/7.

Hire pre-built AI teammates. Give your engineers and operators a platform to ship their own AI apps. Stop losing sleep about what is running where.

Clutch is the platform behind both: pre-built agents for the workflows your ops team should automate first, plus the integration plane your team's vibe-coded apps and Claude Code projects plug into. One platform. Real production. Visible and safe by default.

Built for ops, engineering, and security teams that are tired of the shadow-AI surface area inside their own company.

ALSO WORTH KNOWING

OpenAI's audited 2025 financials show $13B in revenue against $34B in costs. Ed Zitron reported the leaked filings on Tuesday, including $10.59 billion paid to Microsoft for infrastructure and $5.73 billion in sales and marketing spend. The most detailed public read on what running a frontier lab actually costs the operator on the other side of an enterprise contract.

Coinbase registered an AI agent as an SEC-registered investment advisor. The agent itself is the registered entity, not a human advisor who uses AI, with portfolio analysis and tax-loss harvesting rolling out to Coinbase One members. The first legal precedent for an autonomous financial agent holding a regulated identity, which sets up similar moves across banking and insurance.

Odyssey raised $310M at a $1.45B valuation to build world models as a category. Amazon, GV, AMD, and Natural Capital led the Series B for the foundation-model class focused on physical-reality simulation rather than language or image generation. The first venture round to price world models as a distinct frontier alongside large language models.

Y Combinator's Spring 2026 batch has 193 companies, with the AI-to-human relationship inverted. RentAHuman.ai is a marketplace where AI systems hire humans for tasks they cannot do, and Humwork.ai sells human expertise as a service layer for AI agents. The most direct signal yet that the early-stage thesis has moved from "AI replaces humans" to "AI buys human expertise as a discrete input."

Midjourney revealed its first hardware product Wednesday night in San Francisco. Founder David Holz teased "something far weirder" than expected ahead of the 6 pm PT reveal at the company's first physical event. Midjourney has been profitable since launch, has never raised venture capital, and is the first generative-AI company at scale to ship a hardware device.

WATCHING TOMORROW

Friday brings the close of the week of double Q3-2026 acquisitions, with the question of how Salesforce/Fin lands in the procurement conversations Cursor's frontier-model announcement just opened. Anthropic also has not yet posted a public statement on Fable 5 restoration since Saturday's worldwide pull, and any update before the Friday Week in AI would shift the read on the export-control compliance gate.

Back tomorrow,
Haroon