In the middle of what many in Silicon Valley are calling an AI gold rush, one of the field’s insiders is warning that the money is badly out of sync with the science. Jenny Xiao, an OpenAI researcher turned venture capitalist, argues that mainstream investors are three to five years behind the frontier of artificial intelligence research, and that this lag is distorting which startups get funded and at what price. Her critique cuts against the exuberant mood around generative models and suggests that the most important AI companies of the next decade may not be the ones currently soaking up the largest checks.
Xiao’s vantage point is unusual: she moved from an economics and AI PhD track into a researcher role at OpenAI before crossing over into venture capital. From that perch, she is now trying to rewire how capital flows into the sector, betting that a deeper reading of the research papers, rather than the pitch decks, will be the real edge in the current “AI summer.”
The researcher who walked out of the lab and into VC
Before she started warning about the gap between capital and code, Xiao was immersed in the academic side of the field. She dropped out of a PhD program that combined economics and AI to take a researcher position at OpenAI, giving her a front row seat to the rapid progress in large models and reinforcement learning that has defined the past several years of AI development. That experience, and the sense that the commercial world was misreading what the labs were actually discovering, eventually pushed her toward investing, where she now frames herself as a translator between cutting edge research and the venture market, as detailed in reporting on Xiao.
To operationalize that thesis, she founded Leonis, a firm built explicitly around the idea that the best AI investments will come from people who can read and interpret the latest papers rather than simply chase visible traction. Xiao has described her mission as bridging the divide between the “AI summer” narrative that dominates public markets and what the technical community is actually saying about the limits and likely evolution of current systems, a theme that runs through multiple accounts of Jenny Xiao of.
“Three to five years behind”: what Xiao says investors are missing
When Xiao says investors are three to five years behind the latest AI studies, she is pointing to a structural lag between when breakthroughs appear in arXiv preprints or conference proceedings and when those ideas filter into mainstream venture theses. In her view, many generalist funds are still calibrating their expectations to research that was state of the art several training runs ago, which leads them to overestimate the defensibility of today’s model-centric startups and underestimate how quickly capabilities and cost curves are shifting. Reports on her comments describe her warning that the current wave of enthusiasm is anchored in an outdated picture of what models can and cannot do, a gap she highlights in interviews about Jan.
That lag matters because AI is not behaving like previous software cycles. Xiao and other specialists argue that the pace of improvement in large models, and the way open research diffuses across labs, means that moats built purely on access to a given model or API are fragile. Yet capital is still pouring into companies whose main pitch is a thin wrapper around a general purpose system, a pattern that has been flagged in coverage of OpenAI researcher turned. Xiao’s contention is that if investors were reading the same studies as the researchers, they would be far more cautious about paying premium valuations for businesses that could be leapfrogged by the next training run.
Inside Leonis Capital’s bet on emerging AI talent
Xiao’s critique of the market is not abstract, it is embedded in how she is deploying money. Leonis Capital has raised a dedicated pool of capital to back early stage AI founders, with reporting on Takeaways noting that the firm secured 25,000,000 dollars for a new fund aimed at uncovering the next OpenAI. That fund is structured to move quickly on technical teams spinning out of top labs and universities, often before they have polished commercial plans, on the theory that the right researchers will find product market fit faster than polished founders who are a step removed from the frontier.
Accounts of the raise describe how the new vehicle is backed by limited partners who are comfortable with the idea that the most important AI companies may look underwhelming by traditional SaaS metrics in their first year, and that the real signal lies in the depth of the team’s work and its alignment with where the research is heading. Coverage of the fund explains that it is designed to help Leonis Capital serve early stage founders better, including by accepting that the most promising AI-native businesses may not resemble the clean subscription curves that investors grew used to in the cloud era. A separate summary of the same raise underscores that the strategy is to identify emerging talent that could build the “next OpenAI,” a phrase that appears in descriptions of Bloomberg AI.
Why the AI hype cycle looks different from SaaS
Part of Xiao’s argument is that investors are applying the wrong mental model to AI. In software as a service, the playbook was to find recurring revenue, layer on sales and marketing, and assume that incumbency and switching costs would protect margins. AI, she and others argue, behaves more like a fast moving research field than a stable enterprise category, with breakthroughs and cost collapses that can render a once defensible product obsolete in a single model release. A detailed write up on venture capital highlights this contrast explicitly, noting that unlike Software as a Service, AI products are tightly coupled to underlying model performance and hardware constraints.
That mismatch between playbook and reality is feeding what Xiao sees as a distorted hype cycle. Reports on her comments describe her saying that it is “AI summer,” with capital and attention flooding into the space, but that some of the loudest boosters are not closely tracking what the latest studies are saying about what is ahead. Multiple summaries of her remarks, including one focused on Startups and another on Chai Disco, emphasize her view that the disconnect is not just about valuations, it is about which problems are being tackled at all. In her telling, too much money is chasing incremental wrappers around general purpose models, while underfunded researchers are working on areas like alignment, efficiency and domain specific architectures that could define the next wave of value creation.
Signals of a bubble, and what could pop it
Xiao’s warning lands in a broader context of anxiety about an AI bubble. One report on OpenAI’s financial relationships with outside investors cites an MIT survey of 300 economists that found widespread concern that speculative capital is inflating valuations across the sector, and that the hype around generative models is keeping the bubble inflated. That survey is referenced in coverage of a separate controversy around OpenAI’s funding, which notes that the MIT survey found that the majority of respondents saw clear signs of overvaluation. In that same piece, Xiao is quoted again as “Jenny Xiao of Leonis,” reinforcing that her critique of investor myopia is part of a larger debate about whether AI is in bubble territory.
Her specific claim that investors are three to five years behind the research has been repeated across several outlets, including detailed write ups that identify her as Jenny Xiao of and others that refer simply to Jan. Additional summaries on Jan 16 and Jan 17 repeat the same three to five year figure, underscoring how central it has become to the conversation about AI investing. Separate coverage of Jenny Xiao of in the context of OpenAI’s financial entanglements shows that her warning is being heard not just in venture circles but in broader debates about how AI should be funded and governed.
More from Morning Overview