Despite huge investment, turning AI hype into sustainable profit is proving far harder than expected. Vast amounts of venture capital have gone into AI companies, or at least into companies that claim to be AI companies in their pitch decks. In 2025 no less than $192 billion was invested in AI startups, 61% of the total. In Q1 2026 that had risen dramatically to 80%, with $242 billion invested, a huge increase in absolute terms. So, how is all that investment performing so far?
It is a little early to tell, given the multi-year horizon of venture capital, but there are some troubling early signs. Monarch Tractor promised to put AI into tractors to improve efficiency on farms, and raised $240 million at a valuation of $518 million, launching its product in 2023. In early April 2026 it closed its operations and laid off all its staff. Early user Patrick O’Connor, a winemaker, claimed that it was dangerous in its self-driving mode: “I wouldn’t let anyone else around it”.
The British company called builder.ai entered insolvency in 2025 after a lawsuit revealed that its “AI development” was largely done by offshore engineers. Its AI engine, “Natasha”, turned out to be mostly a team of 700 engineers in India. It had raised around $500 million from investors on the promise of making custom software development “as easy as ordering a pizza”. In May 2023 it achieved a $1.5 billion valuation. By March 2025 whistle-blowers revealed that much of the revenue growth had been inflated, using “round tripping” with another company, whereby each company bought services from the other, artificially inflating the revenues of both. A US financial investigation began in May 2025. Bankruptcy followed in June 2025.
This was not the only case of an AI tool being essentially a bunch of offshore engineers standing on each other’s shoulders and wearing a trench coat. Amazon’s “Just Walk Out” stores promised a staff-less retail experience based on AI. In fact there was a team of a thousand offshore staff that reviewed camera footage in the stores and manually put together receipts for customers.
The UK legal-tech company Robin AI promised a mix of legal services and AI software, and in 2023 had a Series A financing round led by SoftBank. It gained over a dozen Fortune 500 customers but investors were disappointed in its growth, and a Series C funding round failed in late 2025. Most of its staff were sold off as a team to a rival in December 2025 in a distress sale.
Autonomous-driving company Ghost Autonomy raised $220 million and had a partnership with OpenAI. Its multimodal LLM promised to enhance the reasoning ability of self-driving systems. Truly autonomous self-driving turns out to be a hard engineering problem, and the company went bankrupt in April 2024. Digital AI advert generator company Icon.com became briefly famous for spending $12 million on a domain name but its product struggled and the company folded in March 2026.
The list of failed AI companies extends beyond the Anglo-Saxon world. French computer vision company Prophesee raised €126 million but went insolvent in October 2024. Volkswagen’s Cariad software unit, including a team for autonomous driving, has had well publicised struggles despite huge funding and resources.
The AI label has led to dubious companies raising money without ever really having a true AI product. Gameon raised $60 million from investors based on AI technology that mimics human conversation. The founders were indicted in early 2025 after they mostly spent the money on themselves, including a lavish wedding venue. Fintech company Tally raised $172 million at a $885 million valuation, aiming to consolidate credit card debt using “smart algorithms” and automation, though its perception as an AI startup was more marketing than reality. Effectively a capital-intensive lending business, it folded in 2024.
Of course, we would expect there to be a fairly high failure rate in any new industry, especially one awash with capital. But what about the AI frontier labs? Here the investor faith continues. OpenAI added $122 billion in funding at a valuation of $852 billion in a March 2026 funding round. Its rival Anthropic did a Series G round in February 2026, raising $30 billion at a $380 billion valuation. They are going to need every dollar. OpenAI lost $13.5 billion in the first half of 2025 and does not expect to be profitable until 2030. Anthropic saw strong sales growth in 2025, up from $1 billion to $9 billion, yet still burned over $3 billion in cash, hoping to break even in 2028. The issue for these companies is that not only are LLMs extremely expensive to train (ChatGPT-4 cost nearly $100 million to train) but they also incur inference costs with every single question that a user poses to the LLM. As I explored recently, this means that frontier models are currently losing money on most queries that run, even for those on paying subscriptions (at least 95% of ChatGPT users are using it for free). Anthropic at least has managed to make real progress in enterprise sales, capturing around three quarters of all new enterprise licences by February 2026.
Much of the revenue that AI companies are booking is circular. For example, 70% of Coreweave’s revenue comes from Microsoft, with much of the rest coming from Nvidia. So Coreweave buys Nvidia chips and then rents them back to Nvidia. OpenAI has a $22 billion deal to access Coreweave’s infrastructure, built on Nvidia chips. Nvidia, the one company in AI actually making money in spades, has invested in OpenAI and Anthropic, which in turn will buy chips from Nvidia.
At the moment most of the AI plates are still spinning in the air, as shown by the funding rounds of OpenAI and Anthropic. However, the increasing list of AI insolvencies may be a canary in the coal mine for the AI industry. At some point a company other than Nvidia needs to actually make some money, or there will be a day of financial reckoning. The AI industry needs to demonstrate that AI can generate durable, independent profits, and not just circulate capital within its own ecosystem. Until then, the question isn’t whether AI is transformative, but whether it is economically sustainable.
