Gaming Worlds Could Be The Answer To AI’s Data Problem

Welcome back to The Prompt.

Artificial intelligence is facing an existential crisis. Powerful AI models are gated by the quality of data they’re trained on. But even after scraping the internet for all its public data, paying some of the most intelligent humans to label and annotate data and generating troves of synthetic data, frontier labs are coming up short on data to improve model performance. And so the question remains— where will the data for the next major advancement in AI come from?

Two former Stanford PhD students, Sharon Lee and Fan-Yun Sun, are trying to answer that question at their new startup, Moonlake AI. The pair are developing AI software that can quickly create visual simulation environments and “interactive 3D worlds” that might serve as the bedrock to create data for reasoning models— systems that carry out multiple steps to solve complex problems. The idea is that the tool will organically be used by people to generate 3D worlds for gaming, animation, filmmaking or even for education, which in turn would create data that can be used to train more advanced models, said cofounder Sun, who previously worked as a researcher at Nvidia, building virtual worlds to train and evaluate robots.

“We know that data is very scarce right now,” Lee said. “And we believe these large scale interactive worlds are the next paradigm that allows you to scale the data infinitely.”

Coming out of stealth today, Moonlake AI has raised $28 million in seed funding from AIX Ventures, Nvidia Ventures and Threshold Ventures. Lee said the program could also be used by AI researchers in fields like robotics to create digital simulations and verify if a task has been correctly completed. “For example, if a robot interacts with the blender in a kitchen in the environment we create, you can see if this solid becomes liquid and the fruit juice is blended and that would be a successful task,” she said.

Moonlake AI isn’t the only company using artificial intelligence to generate 3D worlds. World Labs, cofounded by esteemed Stanford faculty Fei Fei Li, who is known as the godmother of AI and was Lee’s mentor, is also working on spatial intelligence and creating interactive virtual worlds. In August, video generation startup Runway launched Game Worlds to let people create interactive games.

Let’s get into the headlines.

BIG PLAYS

OpenAI announced users will be able to buy products directly through ChatGPT from sellers on Etsy and Shopify. Some 700 million people already use the chatbot to search and compare products. OpenAI is reportedly planning to introduce ads within ChatGPT as it tries to monetize its star product. That’ll be important for the AI goliath, which has been spending billions buying compute from companies like Oracle and CoreWeave. The company booked $4.3 billion of revenue in the first half of 2025 on $2.3 billion in cash burn, The Information reported, but has committed to spend up to $500 billion on its 10 gigawatt AI infrastructure project called Stargate in the next four years. OpenAI projects $13 billion in revenue this year.

CHIPS + COMPUTE

AI cloud compute giant CoreWeave signed a $14.2 billion contract with Meta to provide the social media behemoth with computing power through 2031 to train new models. OpenAI also recently expanded its contract with CoreWeave and now plans to spend $22.4 billion on its infrastructure. The deal comes amid an industry-wide infrastructure arms race as leading AI companies splurge astronomical amounts of money on AI data centers and advanced silicon GPUs. Investors and tech companies would need to book some $2 trillion in annual revenue through 2030 to profitably fund its AI investments, according to a report by Bain and Company.

TALENT RESHUFFLE

Elon Musk’s businesses have experienced a significant churn among senior level leaders in recent months, the Financial Times reported. At xAI, Musk’s AI startup, both the chief financial officer and general counsel recently left, citing long hours of work as a key concern. The departures have been driven by burnout and because some staffers are fed up with Musk’s politics. AI talent has become a need-to-have for startups scrambling to win the AI race, which means there’s plenty of opportunity for employees to seek greener pastures.

AI DEALS OF THE WEEK

Axiom Math, an early stage startup that plans to train an AI model that can solve challenging mathematical problems, raised $64 million in seed funding at a $300 million valuation, I reported this week.

Up and coming legal AI startup Legora is in talks to raise more than $100 million at a $1.8 billion valuation, my colleague Iain Martin and I reported.

Periodic Labs announced it has raised $300 million in seed funding from Andreessen Horowitz and others. Cofounded by ChatGPT co-creator William Fedus, the startup plans to train artificial intelligence models for scientific discovery in areas like semiconductors, magnetism and superconductivity.

DEEP DIVE

When Nvidia, the $4.4 trillion juggernaut that makes the GPU chips undergirding the AI frenzy, needs to monitor its systems for threats and suspicious activity, it turns to a small startup named Grafana Labs to help it scour access logs and analyze vulnerabilities. And the chip maker isn’t alone. Uber, Anthropic and Adobe are among the other giants tapping Grafana’s tools — like a dashboard for operations across an entire business.

Now $6 billion-valued Grafana, known as an observability platform, on Tuesday said it hit $400 million in annualized revenue. The uptick in sales has been thanks in part to the growth of Grafana Cloud, which lets customers use its platform without setting up their own infrastructure like servers or storage, CEO Raj Dutt told Forbes.

The need for good observability services has ballooned in the AI era, especially as vibe coding from generative AI models has become more common. As people increasingly rely on AI to write code and build products, it’s become harder to keep track of what’s going on under the hood, when it’s being done and who’s doing it, Dutt said. “Software today is starting to look more like a living organism,” said Dutt. “It’s being built and shipped faster than ever — sometimes sloppily. That makes it critical to understand how your app is performing in production.”

Read more on Forbes.

MODEL BEHAVIOR

Tilly Norwood, an AI-generated character dubbed as an “AI actress,” received backlash from SAG-AFTRA, a union that represents over 160,000 (human) actors. The organization alleged the synthetic performer was generated using actors’ performances without permission or compensation. It also noted that the use of AI-generated figures could risk “jeopardizing performer livelihoods and devaluing human artistry.”