AI may be the hottest thing since sliced bread. But that doesn’t mean it’s getting easier to develop and run. According to a recent Boston Consulting Group poll, 74% of organizations are struggling to derive value from their AI investments.
William Falcon, the creator of PyTorch Lightning, a popular open source AI framework, says that one of the biggest mistakes businesses make is underestimating the amount of legwork involved in AI orchestration. “Building your own AI platform today is like building your own Slack — it’s complex, costly, and not core to your business,” he told TechCrunch. “The value for enterprises lies in their data, domain knowledge, and unique models — not in maintaining AI infrastructure.”
Falcon, a former Navy Seal trainee and Facebook AI Research intern, started developing PyTorch Lightning while an undergraduate student at Columbia. The framework provides a high-level interface for the AI library PyTorch, abstracting away the code to set up and maintain AI systems.
After dropping out his NYU Ph.D. program, Falcon decided to team up with Luis Capelo, Forbes’ former data products lead, to commercialize PyTorch Lighting. Their venture, Lightning AI, takes the open source framework and layers enterprise-focused services and tools on top.
“We have thousands of developers single-handedly training and deploying models [with Lightning AI] at a scale that would have required teams of developers without Lightning,” said Falcon.
Lightning AI handles normally cumbersome tasks like distributing AI workloads across servers and provisioning the infrastructure needed to evaluate and train AI. The company’s flagship product, AI Studios, allows customers to fine-tune and run AI models in the cloud environments that they prefer.
Companies can even use Lightning AI to host AI-powered apps that run on private cloud infrastructure or their on-premises data centers. Pricing is pay-as-you-go, with a free tier that includes 22 “GPU hours” per month.
Falcon says that the goal of Lightning AI is to make AI dev “as intuitive as using the iPhone.” The platform has helped Cisco cut infrastructure setup time to two days, he claims, and enabled researchers at his alma mater, Columbia, finish hundreds of experiments in 12 hours.
“Most people don’t know this, but many of the world’s leading AI products have been trained or built on Lightning,” Falcon said. “For example, Nvidia’s suite of models, NeMo, was built using Lightning tools — Stable Diffusion by Stability AI is another.”
Certainly, Lightning AI has momentum. More than 230,000 AI developers and 3,200 organizations use the platform today, and the company recently raised $50 million in a funding round.
There’s competition, though. Comet, Galileo, FedML, Arize, Deepset, Diveplane, Weights & Biases, and InfuseAI offer comparable mixes of paid and free AI orchestration services.
Falcon, for his part, believes the market for managed AI solutions is big enough to support many players. And he’s likely not wrong. Per Fortune Business Insights, the machine learning operations industry vertical — Lightning AI’s vertical — could be worth roughly $13 billion by 2030.
With the recent $50 million investment, which was equally led by Cisco Investments, J.P. Morgan, and Nvidia, Lightning AI’s total war chest stands at $103 million. The New York-based, 50-person company plans to spend the proceeds on recruiting new customers, including government customers, and expanding the Lightning platform to new markets.
“With a lean, high-performance team and a 90%+ gross margin product,” Falcon said, “we are on track to reach $10 million to $20 million in annual recurring revenue by the end of next year and achieve profitability shortly after.”