What Meta-Scale AI Deal Could Mean; Cursor’s Explosive Revenue

Before we get to today’s column, here’s something to noodle on.

On Friday, word spread widely across the industry that an unnamed public company was going to invest a cool $10 billion in an unnamed private AI firm. On Saturday afternoon, Scale AI’s name popped up as the alleged recipient. By nighttime, Bloomberg reported Meta Platforms was the public company in question. 

The question was: Why?

The quality of Meta’s models is arguably behind that of at least half a dozen other major AI developers, if not more, according to researchers and customers. Those developers include two from China (DeepSeek and Alibaba) who seem to be beating Meta in the open-source AI race. 

Meta needs to change things up.

Scale AI, meanwhile, has been growing nicely in the market for hiring armies of people to produce data that models—including Meta’s—can use to improve their knowledge and abilities. 

These days, the race is on to produce data on solutions to problems in advanced scientific fields, which major AI developers think they need to produce artificial general intelligence—or at least PhD-level AI. That includes hiring PhDs to spend hours writing up answers to tough problems and explaining it in a way that models can understand.

Commissioning this kind of work isn’t cheap! (More on that here.)

But $10 billion? Perhaps Meta wants Scale to hire tenured faculty, not just PhDs!

In any case, the deal doesn’t sound so different from the way many other large companies have made equity investments in AI firms that are also either their customers or vendors. Most notably, Microsoft committed more than $13 billion to OpenAI, which is spending that money on Microsoft cloud servers. And OpenAI has helped fund CoreWeave, which in turn is renting out precious AI servers to OpenAI. 

Everybody wins!

On to the main event…

The most talked-about thing in artificial intelligence and venture capital circles last week wasn’t a new model or a CEO quote about AI’s potential impact. It was the bananas growth of AI coding assistant Cursor in the first half of the year.

Anysphere, the company behind the assistant, which launched two and a half years ago, said Thursday it is generating $500 million in annual recurring revenue. Presumably that means it is generating a little over $40 million a month in revenue from subscriptions. 

The company didn’t say how fast it is growing, but thanks to my colleague Natasha’s recent profile of the company and why engineers love their product, we know Anysphere was generating around $200 million in ARR just two months ago. In other words, its revenue graph just formed a hockey stick.

Corporate deals Anysphere struck with the likes of Uber help explain the sudden acceleration. Cursor is also heavily used by engineers at Amazon Web Services, my colleague Anissa says, and Natasha confirmed Snowflake recently signed a deal for thousands of its workers to use the assistant.

Anysphere’s ascent mirrors that of its bigger San Francisco cousins (and competitors), OpenAI and Anthropic. For comparison, OpenAI was generating $1 billion in annualized revenue about eight months after launching ChatGPT and has projected nearly $13 billion in actual revenue this year. Nobody comes close to that kind of growth at that scale. 

Anthropic, for its part, tripled its annualized revenue to $3 billion from $1 billion in the first five months of this year—not too shabby!—thanks in small part to payments it receives from Anysphere.

Anysphere pays OpenAI and Anthropic for access to their large language models (which are great at coding) to power its assistant. The assistant helps engineers with generating and editing code and debugging, and it eventually aims to automate everything an engineer might want to do.

It’s sometimes easy to forget that OpenAI and Anthropic’s own rise was powered by engineers who use their chatbots as coding tools or used their models to make their own coding assistants. Over time, OpenAI and Anthropic added more features to make their services a little more Cursor-like. 

OpenAI is also in the process of buying a Cursor rival, Windsurf, for $3 billion—presumably after earlier sussing out Cursor’s interest in such a deal. That means OpenAI is now the Cursor rival to watch, along with a few others including GitHub Copilot. 

OpenAI’s leaders believe launching an automated software engineering “agent,” which would handle large projects the way a senior staff engineer would, is a key pillar of achieving artificial general intelligence, as we covered here. No word yet on when that product is coming.

It’s safe to say the market for coding assistants is gargantuan and there’s room for multiple winners and products. That opportunity is surely one reason Anysphere on Thursday raised $900 million in new capital (at a $9 billion valuation).

The company is feverishly developing its own coding-focused models, some of which are already in use and would save the company money in the long run because it wouldn’t need to pay Anthropic and OpenAI. But developing models isn’t cheap!

Competing with big money-losing firms like Anthropic and OpenAI could be a slog, even though those firms’ goals (and expenditures) go far beyond building the best coding models.

One question is whether Anysphere’s gross profit margins are going to be much better than those of its crosstown frenemies. OpenAI, for instance, this year projected a 48% gross profit margin, which basically reflects server costs for revenue-generating products as a percentage of revenue. That’s not great for a software business, but the company says that metric will rise by 20 percentage points over the next few years as revenue grows. (If you know any of Anysphere’s projections, we are all ears.) 

Thanks to Natasha, we know that as of late March, Anysphere had just 50 employees. So overall, we imagine Anysphere is still operating more efficiently than the bigger firms.

We can’t wait to see where the AI coding race goes from here.

Policy Watch

The U.S. Senate Commerce Committee released its rewritten text of President Trump’s “Big Beautiful” budget reconciliation bill. The new bill still includes a provision that prevents states from passing their own laws governing AI for 10 years, with a twist.

Instead of an outright ban, the moratorium on AI laws is now a condition for states to access broadband funding. Congress authorized over $42 billion of broadband funding in 2021, but under the new budget bill, states would lose access to those resources if they passed AI legislation.

Senator Ted Cruz, who chairs the Commerce Committee, previously said the moratorium on state AI laws could face scrutiny in the Senate for not relating directly to the federal budget. It is not yet clear whether tying the measure to broadband funding will help it clear that hurdle.

While Cruz supports blocking state AI laws, some of his fellow Republicans have other ideas.

“I’m not voting for the development of skynet and the rise of the machines by destroying federalism for 10 years by taking away state rights to regulate and make laws on all AI,” wrote Marjorie Taylor Greene, Republican Congresswoman for Georgia, on X.

Due to AI, Greene’s manufacturing-heavy district with “one of the lowest unemployment rates in the country could go to one of the highest unemployed districts in the country,” she said. She added that she would vote against the bill—when it goes back to the House for review after passing the Senate—unless the AI measure is removed.

People on the Move

Richard Fontaine, the CEO of Washington, D.C. think tank the Center for a New American Security, is joining Anthropic’s Long-Term Benefit Trust. The trust is a financially independent body that elects some of Anthropic’s directors. For instance, the trust elected Netflix co-founder Reed Hastings to Anthropic’s board last month. Fontaine’s appointment is the latest example of AI companies attempting to deepen their national security credentials.

Here’s what else is going on…

Deals and Debuts

See The Information’s Generative AI Database for an exclusive list of private companies and their investors.

Meta is in talks to invest up to $10 billion in Scale AI, Bloomberg reported.

Plus Automation, a self-driving truck company, will go public in a $1.2 billion merger with a special purpose acquisition company. 

Anysphere, the company developing the Cursor coding assistant, is raising $900 million at a $9 billion valuation, in a funding round led by Thrive Capital, with participation from Andreessen Horowitz, Accel and DST Global.

Voxel, which uses AI to analyze security camera footage to prevent workplace accidents, raised $44 million in a Series B funding round led by NewRoad Capital Partners, with participation from Eclipse, Rite Hite, Tokio Marine, MTech, HG Ventures and Whitestone.

Simetric, which uses AI to help financial institutions reconcile their transaction records, raised $30 million in additional Series B funding led by Goldman Sachs.

Sema4.ai, which develops software to help companies make AI agents, raised $25 million in Series A extension funding, with participation from Snowflake Ventures, Rocketship VC, MVP Ventures, Cox Enterprises, Mayfield and Benchmark.

Obvio, which develops AI-powered cameras to automatically issue driving tickets, raised $22 million in a Series A funding round led by Bain Capital Ventures, with participation from Khosla Ventures and Pathlight Ventures.

Thread AI, which develops software to help companies use AI, raised $20 million in a Series A funding round led by Greycroft, with participation from Index Ventures, Scale Venture Partners, Plug-and-Play, Meritech Capital and Homebrew.

Skyral, which uses AI to simulate the outcomes of public policies, raised $20 million in a Series A funding round led by NOIA Capital, with participation from Accrete Capital.

Lendurai, which develops AI-powered drones, raised $6.4 million in seed funding, with participation from Expeditions Fund, HCVC and Vsquared Ventures.

Mistral is on track to generate over $100 million in annual recurring revenue this year and has signed or is close to signing multiple contracts worth at least $100 million over three to five years, the Financial Times reported.

AMD has hired the team behind startup Untether AI, which develops chips for running AI models, CRN reported.

EleutherAI released the Common Pile v0.1, a large dataset for training language models that is freely available. The 8 terabytes of data were collected in collaboration with the University of Toronto and Vector Institute, Hugging Face, the Allen Institute for Artificial Intelligence, Teraflop AI, Cornell University, MIT, CMU, Lila Sciences, poolside, University of Maryland, College Park and Lawrence Livermore National Laboratory.

Google added the ability for Gemini users to schedule instructions for Gemini at a later time.

Thank you for reading the AI Agenda Newsletter! I’d love your feedback, ideas and tips: [email protected].

If you think someone else might enjoy this newsletter, please pass it forward or they can sign up here.