Why AI Companies Can Afford a $500 Billion Deficit

📝 The LLM Industry Is Digging a $500 Billion Hole. Here's Why That Might Be Fine. Nearly every transformational technology looks like a catastrophic blunder before it looks like a genius bet. Uber burned $32 billion. Tesla burned $9 billion. AI hyperscalers and LLM trainers are now on track to burn $500 billion. This has fueled a tremendous amount of AI Doomerism. Even with 800 million weekly active users, how could ChatGPT ever earn back this kind of investment? But a look back at the history of truly disruptive startups and companies suggests that, counter-intuitively, OpenAI could be on track to do just that. But how? Understanding the J-Curve You've likely heard the expression "you have to spend money to make money." This can be visually depicted on a chart as a "J-Curve." At first, a company does not have customers, or even a finished product, so they're not generating revenue. Still, they have to spend capital... Developing their product, hiring staff, going out and finding customers, and so forth. Cumulative losses compound over time, and the line on the profitability chart dips down. At SOME POINT (unless the company goes out of business), revenue kicks in, the product or service gets closer to product market fit, and it starts attracting exponentially more customers. Profit is assured. BUT the company is not free and clear yet; it's still in the hole from all those early expenditures, and must play catch-up. Over a long enough timeline, that curve turns into a J. A big dip while the company takes in investment dollars or loans and spends it, then a gradual rise out of the pit and into financial success. That big dip on the chart might look scary, but it's not a certain indication of inevitable collapse. Many great and successful companies have dropped into a dark void and then soared back out like the proverbial phoenix. The Uber and Tesla Precedents On TWiST, @Jason took a look back at two key examples from recent tech history: @Uber and @Tesla. Uber's accumulated deficit peaked at around $32.8 billion in 2022. But that same year, the rideshare startup's free cash flow (FCF) -- the amount they had left over after spending all that they needed to run the business -- turned positive, to $0.4 billion. Today, Uber FCF is around $7 billion, and their accumulated deficit is down to just $10 billion. Uber dug that hole in a number of ways, from expanding their infrastructure to building out a driver and rider network via subsidized rides. They were building a two-sided marketplace that did not exist until that point, and which now handles up to 30 million rides every single day. Tesla underwent a similar journey. The electric auto maker went roughly $10 billion in debt at its lowest point in the late '10s. They were building huge factories, developing new manufacturing processes, and scaling up production: all tremendously costly undertakings. The company didn't turn a profit until it was in Year 16. But this is how the J-Curve chart of their free cash flow looks today. Building out manufacturing capabilities and physical infrastructure like a Gigafactory is a massive undertaking, that can dig a company into a deep hole. But if the product does actually get built, and the business model is solid, that company will take a turn toward profitability over a long enough timeline. The Coming LLM J-Curve The J-Curve principle is widely understood by technologists and investors. So why do we still hear so much panic about investment in training AI models, chip manufacturing, cloud infrastructure deals, and data center build-out? Probably because the investment in LLMs and AI technology is so vast, across so many large companies and enterprises. Since the launch of ChatGPT in November 2022, roughly $250 billion has been invested in LLMs and related technology, by private companies including OpenAI, Anthropic, xAI, Mistral, and so on. On TWiST, Jason predicts $250 billion more will ultimately be put into the system, for a grand total of $500 billion. He further suggests that, by 2030, these companies will probably start making a collective $10 billion or so in profits. If that grows at a 50% annual rate, the industry would be back to break-even after 6-10 years. It theoretically could look a little something like this: There was a time, not all that long ago, when this seemed very unlikely. Sure, many people were regularly using chatbots, but there were reports that enterprise pilot programs weren't going well, and doubts about whether this tech would ever seen the kind of mainstream mass adoption required to make up these multi-billion dollar investments. But today is 34 AO (After @OpenClaw). The agentic revolution is HERE, it's impossible to get your hands on a Mac Mini, and founders are switching from Anthropic's Claude Opus 4.6 to cheaper models because they're draining so much compute on everyday tasks. It's no longer impossible to believe that LLM companies will eventually have everyone's business, and make enough profit to pull them out of their $500 billion hole. It's becoming less a question of HOW than WHEN. Nothing is Guaranteed! This is not financial advice! It's also worth pointing out that Tesla and Uber had relatively clear paths to profitability once they achieved the necessary scale. We already knew that people wanted to pay for taxi rides and new cars. It was just a matter of making the systems or products affordable enough, then moving enough units. LLM companies, by comparison, face a lot more open questions. Will inference costs continue to decline at an acceptable rate? Will current pricing levels hold, even as competition intensifies and open-source solutions become better and more widely available? Will profitability eventually pass to the LLM model providers themselves, or divert to the agents and services that get layered and built on top of them? Still, on a fundamental basis, it remains entirely possible that we're just in the pit of another promising J-Curve, staring up at a a seemingly-impossible cliff, wondering how we'll climb it. That's not necessarily the end of the story. Founders ship faster on Deel. Set up payroll for any country in minutes and get back to building. Visit deel.com/twist to learn more. http://x.com/i/article/20277499162110771…