Prodigy Research, a Y Combinator-backed AI research lab, claims it achieved 140% returns during its accelerator batch using a proprietary foundation model for quantitative trading. But the startup has yet to produce third-party audits or administrator-verified performance data, raising questions about claims that have circulated widely in tech and crypto media since August.
The announcement lands in a particularly fraught moment. U.S. regulators have spent the past year warning investors about AI trading algorithms "that promise unreasonably high or guaranteed returns," according to a Commodity Futures Trading Commission advisory that remains active. The SEC's 2025 Risk Alert, last updated in June 2026, lays out strict performance presentation requirements for investment advisers, including proper netting of fees and transaction costs. The gap between what Prodigy Research is saying publicly and what it's willing to show independently is wide enough to drive a truck through.
What the Company Is Saying
The firm's website states that its live systems "more than doubled our capital over our YC batch" and that the company has "never had a down week." The YC Launch page repeats the 140% figure, adding that returns came from delta-neutral strategies during a stretch when "major indices [were] flat or down."
Michael Wang, the CEO and a former researcher at DeepMind and Jane Street, fielded skepticism on X in August. When asked why a firm with genuine alpha would bother raising outside capital, he replied: "it takes capital to convert alpha to dollars!" His co-founder, Yuhua, previously worked at Apple and Salesforce.
Prodigy also claimed to have "beat top 10% Jane Street traders" on internal benchmarks. Huuush, an independent media outlet, pointed out on August 11 that Jane Street doesn't publish trader-level performance leaderboards, making this claim unverifiable. In other words, the only way to know if it's true is to take Wang's word for it.
The Documentation Problem
As of mid-September, no third-party audit, net asset value time series, or independent brokerage confirmations for Prodigy Research's performance had surfaced publicly, according to Huuush's analysis. The outlet called the claims "self-reported" and noted the absence of disclosed methodology, leverage ratios, or capacity constraints. Several crypto news aggregators repeated the figures in August, but none cited independent verification.
The company describes itself as training "the world's best foundation model for quantitative trading" and says it achieved "SOTA live trading" results, using the abbreviation for state-of-the-art. That kind of language is common in academic AI labs. It's less common in institutional finance, where performance claims typically come with pages of footnotes and administrator signatures.
SEC guidance on the Marketing Rule, updated this year, requires investment advisers to present performance net of fees and to follow specific protocols for extracted performance claims. The CFTC's advisory is blunter. It warns that "fraudsters are exploiting public interest in artificial intelligence to tout automated trading algorithms," a statement the regulator has maintained since March 2025.
A Crowded Field with Mixed Evidence
Prodigy Research isn't alone in claiming breakthrough AI performance in finance. Standard Signal, also in Y Combinator's Spring batch, reported a Sharpe ratio above 3 in live trading, with performance details available to qualified investors. Established quantitative managers have taken more measured stances, though. Man AHL published a paper in February describing two agentic AI approaches: Alpha Assistant, an interactive coding agent, and AlphaTrend, a predefined autonomous research pipeline. The firm emphasized that "humans review" the AI-generated strategies before implementation.
XTX Markets, a principal market-making firm, disclosed in a March presentation to the Bank of Canada that it operates "tens of thousands of H100/A100 GPUs" and "650 petabytes of usable storage" for deep-learning applications. The presentation argued that compute has become a "massive barrier to entry" for AI-driven market making, a comment that might explain why even successful trading firms seek outside capital.
Regulated AI-powered products have shown more modest results. Qraft's AIVI fund reported a year-to-date return of 31.94% through June 30, according to an SEC filing. Earthian AI's "Project Alpha-Index" gained 47.8% from late February through mid-July, the company said in a blog post. While these figures represent different timeframes and market conditions, neither approaches Prodigy's claimed 140%.
Timing and Market Context
The claims emerged against a backdrop of significant volatility in AI-themed trading strategies. Goldman Sachs reported that hedge funds suffered their worst monthly underperformance versus the S&P 500 in more than 20 years during July, according to coverage by Institutional Investor and other outlets published in August. The underperformance stemmed from heavy de-grossing of AI-linked long positions as what traders called the "crowded AI trade" unwound.
The HFRI Fund Weighted Composite Index rose 1.7% in August, with macro strategies leading at 4.11%, according to HFR estimates published in early September. The July reversal illustrated concentration risk in model-driven strategies, particularly those relying on similar AI signals. That kind of environment can produce outlier returns for contrarian strategies. It can also produce outlier returns from luck, especially over short time horizons.
U.S. equities volumes hit record levels in 2025. Average daily volume rose 44.6% year-over-year to 17.6 billion shares, Cboe reported in March. Off-exchange trading surpassed 50% of total volume for the first time in 2025, reaching 50.6%, according to World Federation of Exchanges data published in February. The growth in liquidity has made certain high-frequency strategies more viable, but it has also intensified competition.
Capital Intensity and Practical Limits
The capital requirements of AI-driven trading may explain why Prodigy Research is fundraising despite claimed profitability. A systematic review published in June in Science Direct found a "persistent gap between predictive accuracy and practical, net trading profitability" in machine-learning trading strategies. The paper cited execution costs, capacity constraints, and model degradation as primary culprits.
Financial institutions are scaling AI budgets broadly. NVIDIA's January survey of more than 800 financial services professionals found AI "usage at all-time high" with near-universal plans to increase or maintain spending. McKinsey reported in August that 44% of companies use AI at enterprise scale, up from 38% in 2025, with financial services among the leading sectors.
Global equity market capitalization reached $157.8 trillion in 2025, up 18.9% year-over-year, according to the SIFMA Fact Book published in August. The Cambridge Centre for Alternative Finance wrote in an April report that OpenAI was the most-used foundation model provider in financial services, though it cautioned about "self-reporting bias risk." That caveat applies to startup performance claims as well.
The Compliance Environment
Prodigy Research's marketing approach faces an evolving regulatory landscape. FINRA's 2026 Annual Regulatory Oversight Report, published in January, detailed generative AI supervisory expectations under Rule 3110. The Financial Conduct Authority in the UK issued a multi-firm review on "Frontier AI and cyber resilience" in September, emphasizing board visibility and operational guardrails.
The European Union's AI Act entered a new enforcement phase on August 2, with transparency obligations for general-purpose AI providers now active. High-risk applications under Annex III face compliance deadlines in December 2027, with embedded high-risk systems covered by August 2028, according to European Commission guidance updated this summer.
A June systematic review of machine-learning trading strategies concluded that many academic papers showing strong predictive accuracy fail to translate into consistent net profitability once real-world execution costs are factored. Transaction costs, market impact, and capacity constraints erode theoretical alpha. The review didn't name Prodigy Research, but the timing is notable.
What Comes Next
Cboe plans to launch near-24x5 U.S. equities trading on its EDGX exchange in December, pending final SEC approval and operational readiness, the company said in May. Early-hours trading average daily volume surged 1,203% from 2022 to the first quarter of 2025. The expansion will shift microstructure dynamics and cross-time-zone liquidity patterns for algorithmic strategies, potentially creating new opportunities and new risks.
Two Sigma wrote in a January outlook that "organizational capabilities matter as much as model quality" in AI-driven investment management. The firm emphasized research discipline, engineering excellence, and collaboration over claims of breakthrough model performance alone. It was a subtle dig at the hype cycle, perhaps, or just a statement of the obvious.
Independent verification remains the missing piece. Huuush concluded its August 11 analysis by noting that without administrator-verified returns, disclosed methodology, or transparent capacity and leverage figures, investors cannot assess whether Prodigy Research's claims represent genuine alpha or statistical noise amplified by short time horizons and undisclosed risk. Until that changes, the 140% figure is just a number on a website, repeated in press coverage but backed by nothing an investor can independently examine. In finance, that's usually where the story ends badly.
