Lightning AI is a platform for building machine learning and publishing Lightning Apps that glue together lifecycle tools. To learn more about Lightning AI, Pulse 2.0 interviewed Lightning AI CEO William Falcon.
William Falcon’s Background
Falcon grew up in Venezuela and came to the United States when he was a teenager to pursue a higher education.
“As an undergraduate student at Columbia University, I co-founded (and was CTO, CAIO) of NextGenVest, an AI startup focused on helping students finance college over SMS (the company was acquired by Commonbond). I was also a software engineer at Goldman Sachs and developed over 8 iOS apps as a product manager and iOS engineer. I graduated magna cum laude from Columbia University, earning a B.A. in Computer Science and Statistics, with a Math minor, in 2018. In September 2018 I began my Ph.D. at NYU in deep learning (CIVLR lab),” said Falcon. “As a Ph.D. intern at Facebook AI Research, I worked under Yann LeCun. During this time, I was developing the open-source AI framework I created, called PyTorch Lighting. I open-sourced it while at Facebook and it quickly became a tool of choice for AI researchers and ML engineers in the industry working on state-of-the-art artificial intelligence projects.”
PyTorch Lighting – which is also known simply as Lightning – has quickly emerged as one of the most widely used AI frameworks for its ability to accelerate and simplify the training of machine learning models. And while at Facebook, Falcon was among the first to realize that deep learning should be done in the cloud and this strategic bet was a runaway success from the start. As of now, the open-source Lightning framework has been downloaded over 48 million times and is used across more than 10,000 organizations for its ability to quickly and cost-efficiently train ML models at scale.
In 2019, Falcon founded Lightning AI (initially called Grid.ai) to bring AI within reach of any team or organization. And Lightning’s mission is to enable researchers in academia, machine learning engineers, individual entrepreneurs, and software developers in the industry to create and operate applications, products, tools, and services powered by AI without the resources of big tech.
Formation Of Lightning AI
While Falcon was a Ph.D. intern at Facebook AI Research, he started working on the PyTorch Lightning framework due to the complexity of using PyTorch.
“I released it to the open-source community two months later. As companies in every industry began putting PyTorch Lightning into production, I realized that the fragmentation of the AI ecosystem was a massive challenge that was slowing the adoption of AI. New ‘missing’ parts of the AI stack kept emerging—things like feature stores and data versioning, for instance—and later we saw the need for experiment managers and the ability to use multiple hardware accelerators without code changes, features we pioneered in 2020,” Falcon added. “Each missing piece of the ML puzzle puts a massive drag on the overall pace of AI innovation. An incomplete and fragmented stack makes it more time-consuming and costly than it needs to be to build AI. Just getting a model ready to be pushed into production takes hundreds if not thousands of developer hours spent purely on infrastructure. I founded Lightning AI (which was Grid.ai at the time) in October 2020, and a few months later we secured our initial funding from Index Ventures to evolve the framework.”
In June 2022, Falcon introduced the Lightning Platform to enable users to collaborate while building artificial intelligence tools and products by handling each part of this process, ranging from training models to deploying those models in real-world services. And by giving users a powerful yet simple platform on which to manage the conception, development, and deployment of their AI-powered product, the Lightning Platform lowers the barrier to entry for individuals and organizations looking to access AI technology. The same year, the company rebranded from Grid.ai to Lightning AI with additional funding.
Challenges Faced
What were some of the challenges that Falcon faced in building the company? Falcon told me that one of the first challenges was selecting the right VC partners.
“Many founders focus on locking in the deal terms without also ensuring the VCs understand the company and its vision, or share the same values. I didn’t want to make that mistake and have been extremely fortunate to partner with the right investors,” Falcon acknowledged. “To date, Lightning AI has raised a total of $58.6 million and is backed by Index, Bain, First Minute Capital, and the Chainsmokers’ Mantis VC. Also, Lightning AI’s success is directly related to the talents of its employees and partners. In the extremely competitive AI space, we have to be aggressive in identifying and hiring the most qualified team that can scale our tech and keep pace with market demand. I am proud of the leadership team we’ve assembled, with talent from Cohesity, Confluent, Meta, and more.”
Core Products
What are Lightning AI’s core products and features? Lightning AI’s core product is the Lightning AI platform, which unifies the AI/ML development and infrastructure lifecycle from start to finish.
“The platform features an easy-to-use and modular interface that makes AI and ML accessible to more people, not just big technology companies. Users can access the hardware they need to make ML tools run on the cloud without dealing with complex computer engineering. Lightning’s simple interface allows users to group together ML tools (“Components”) into production-ready AI applications (“Lightning Apps”) that can be used in academic or enterprise settings. By abstracting away complex engineering, it allows teams of any size to work on ML applications, accelerating work completion from months to days,” Falcon explained. “The Lightning AI platform is built on and extends the open-source Lightning framework. This framework quickly became one of the most widely used deep learning frameworks because it offers a quick, simple, and significantly less costly way to train machine learning models in the cloud. The core innovation of the LIghtning platform is its transformation of the complex development of ML technology into a simple process. We separated a typical AI workflow into discrete, open-source components that are freely available on the Lightning Component Gallery. Our users include pharmaceutical companies, healthcare organizations, and technology companies including NVIDIA that want to implement AI technologies as quickly and efficiently as possible. We eliminated the burden of managing processes like hyperparameter optimization and model serving infrastructure by turning them into modular Lightning Components.”
Users now have the ability to simply select from a number of prebuilt, open-source Lightning Apps that they can use right out of the box or extend into a solution that suits their specific needs. And rather than reinventing the wheel whenever a researcher or enterprise needs, for instance, a recommender system or the ability to search for things inside a video, those functionalities are available to users directly from the Lightning App Gallery.
Lightning AI’s platform makes it possible to both develop AI-powered solutions and implement them in a real-world setting quicker and easier than ever before. And by focusing development on specific verticals like healthcare, finance, and academic research, the company is able to widen the accessibility of AI technology beyond the narrow remit of large-scale organizations willing to invest millions of dollars and years of work into developing these solutions.
Evolution Of Lightning AI’s Technology
How has Lightning AI’s technology evolved over time? The Lightning AI platform was built on and extends the Lightning open-source framework (aka PyTorch Lightning).
“We’ve continued to innovate our technology with the recent launch of PyTorch Lightning 2.0, which introduces a stable API, offers a host of powerful features with a smaller footprint, and is easier to read and debug. We also introduced Lightning Fabric to give users full control over their training loop. This new open source library allows users to leverage tools like callbacks and checkpoints only when needed, and also supports reinforcement learning, active learning and transformers without losing control over training code,” Falcon shared. “Lightning Fabric is powerful for training LLMs like Lit-LLaMA, an independent implementation of Meta AI’s LLaMA that is open source under the Apache 2.0 license (making it easier to adopt for deep learning projects that use similar licenses). Using Fabric, teams can train and fine-tune LLMs much faster than traditional methods.”
Biggest Milestones
What have been some of Lightning AI’s biggest milestones? Falcon pointed out that there have been many milestone moments since Lightning AI was founded.
“However, our defining moment was the development of PyTorch Lightning, our open-source AI framework used by more than 10,000 organizations and downloaded more than 48 million times. It is one of the fastest-growing open-source projects in history, noted for its simplicity, modularity, and sustainability that make the developer experience faster, more powerful, and more efficient,” Falcon shared. “The evolution of our open source portfolio, including the release of PyTorch Lighting 2.0 and the new open source Lighting Fabric, means that Lightning AI can support a wider range of individual and enterprise developers as advances in machine learning are growing exponentially. Other milestones are our funding rounds, totaling $58.6 million (more on that below). The capital allows us to continue our mission of accelerating the development of an AI-powered world.”
Customer Success Story
When I asked Falcon about a customer success story, he cited a couple of examples:
1.) Mars Petcare diagnoses up to 700 cases of cancer in pets per day. Using Lightning, Mars Petcare has been able to seamlessly debug and inspect models and imaging to support a more reliable and objective evaluation of cancer in pets, including automatically processing and interpreting 15,000 x-ray images each day, with a whopping 99% increase in processing time to remove significant bottlenecks that could mean the difference between the life and death of a dog with cancer.
2.) SyntheticGestalt — an AI startup based in London and Tokyo — is scaling accelerated drug discovery. Using Lighting AI, the company created an automatic system to make valuable drug discoveries powered by machine learning algorithms and molecular simulations to validate potentially effective drug treatments. And with Lightning, the startup was able to run 15,000 confirmations (the largest set to date) in just one day compared to the 40 days it would have taken without the platform.
Funding/Growth
Lightning AI raised $58.6 million in funding to date. And last year, usage of the Lightning framework skyrocketed with a 42% increase in month-over-month adoption.
“We also grew our team last year to keep up with market demand by appointing a new VP of engineering, VP of product, VP of UX and design and chief engineer of a newly formed PyTorch team,” Falcon informed.
Total Addressable Market
What is the total addressable market (TAM) size that Lightning AI is pursuing? “AI breakthroughs in academia and industry have accelerated in recent years, fueling skyrocketing demand for AI-powered applications. Therefore, our market opportunity is vast,” Falcon analyzed. “The global artificial intelligence (AI) market size is expected to hit US $1,591.03 billion by 2030, according to Forrester. Looking at AI software, hardware, and services in the aggregate, IDC forecasts that worldwide revenues will surpass $500 billion by 2024.”
Differentiation From The Competition
What differentiates Lightning AI from the competition?
“The vision I’ve been pursuing since my time at NYU has always been to build something like an operating system for artificial intelligence, that allowed all the disparate pieces of the AI ecosystem to work together. No solution like this existed,” Falcon reflected. “In the same way you don’t need to know anything about the internal combustion engine to drive to the grocery store, why should you have to know about Kubernetes, cloud infrastructure, distributed file systems and fault-tolerant training to simply bring your AI project to life? Current solutions hand you the disparate pieces of a working car and hope that you’ll be able to assemble them into something that you can use to take a drive.”
This is why after two years of laser focus on solving the issue of fragmentation in machine learning development, Lightning introduced the Lightning AI, enabling researchers in academia, machine learning engineers in industry, and everyone in between to bring to life end-to-end ML systems in a matter of days rather than the years of work that currently requires.
Future Company Goals
What are some of Lightning AI’s future company goals? “Artificial intelligence (AI) can address many of the issues humanity faces across the globe, but its application is currently hindered by a fragmented ecosystem and a surfeit of mismatched tooling. As a result, many of the cutting-edge implementations of AI technologies are limited to organizations with the budget and technological expertise necessary to operate them,” Falcon concluded. “We want to continue to accelerate the widespread adoption of AI, regardless of an organization’s size or the resources at its disposal. Our core innovation – the Lightning platform – is open-source for precisely this reason.”
