Retail traders favour bullion in week-one of AI-driven prediction platform
UpsideOnly, an AI-powered prediction and paper-trading platform operated by USO Labs Ltd., a subsidiary of NASDAQ-listed Perpetuals.com Ltd., recorded robust early engagement following its May 19, 2026 launch. The platform said it attracted 24,568 active users from a pre-registration pool of more than 80,000, generating $2.38 billion in cumulative simulated trading volume and 101,511 fills across 26 instruments in its first seven days.
The distribution of activity surprised the company and may be notable for market observers: precious metals led volume with $757.2 million in XAU/USD, followed closely by bitcoin at $709.5 million. Precious metals in total accounted for about 36.7% of reported platform volume, outperforming many major crypto pairs combined.
How the platform works
UpsideOnly allows users to submit simulated trades across crypto, equities, commodities and FX. According to Perpetuals, those user predictions are analysed by proprietary AI models, which are used as a signal layer. When the algorithm identifies sufficiently strong signals, Perpetuals says it will deploy its own capital to execute live trades and share upside with eligible participants when those trades are profitable.
The model aims to separate the expression of market conviction from the risk of real-money trading, effectively turning collective retail predictions into an information input for an institutional trading book. That architecture attracted users from around the world in week one, with activity reported from 185 countries and notable early participation in markets including Indonesia, the Philippines, India, Kenya, Nigeria and Ghana.
Week-one mix and user metrics
Beyond the bullion-versus-bitcoin headline, UpsideOnly provided a week-one breakdown showing broad participation across asset classes. Highlights included high turnover in other crypto majors such as SOL and ETH, equity interest concentrated in large US names, and activity in FX and commodity futures. Per-user figures were elevated: average prediction volume per active user was reported at roughly $96,700 for the week, with an average fill size of about $23,400.
Those per-user averages are unusually large compared with typical retail account sizes, underscoring the cautionary point that paper-trade volumes can materially exceed what users would risk with real capital. Paper trading often incentivises risk-taking and may not map directly to future live-flow behaviour.
What the numbers mean for markets and firms
UpsideOnly’s early data points are relevant to several industry trends. First, if retail traders do indeed broaden their focus beyond crypto and single-name equities to macro assets such as gold and FX, that could alter the informational content of retail order flow and the way sell-side and buy-side firms monitor retail sentiment.
Second, the platform represents a new variant of the signal-to-capital model: collective predictions feed AI analytics that in turn inform a proprietary trading book. For firms considering similar architectures, the appeal is straightforward: aggregate diverse views without exposing retail users to principal loss, while monetising high-quality signals. But execution matters. The economic value of aggregated retail predictions depends on signal stability, the absence of gaming or wash activity, and the timing and transparency of how a firm translates signals into live trades.
Risks, regulatory questions and limitations
Prediction markets and gamified trading products have drawn regulatory attention in recent years over concerns about investor protection, misleading design and potential for harm to unsophisticated users. UpsideOnly’s paper-trading environment sidesteps immediate capital risk for users, but it raises adjacent questions: how are signals vetted, what safeguards prevent manipulation, and how transparent will the sponsor be about which user-generated signals triggered a live trade?
Moreover, paper or simulated trading is prone to behaviour that differs materially from real-money markets. Traders often take larger or more speculative positions when not risking capital, which inflates reported volumes and can produce misleading impressions about demand. Regulators and market participants will likely scrutinise whether aggregated simulated activity serves as a reliable input for live trading and whether users are adequately informed about the limitations and incentives embedded in the model.
Industry context and next steps
Perpetuals’ leadership framed the week-one result as an early signal that retail investors, when offered a broader asset menu, may behave more like macro traders than crypto-only speculators. That observation, if borne out over time, could influence product design at retail platforms and how institutional desks interpret retail flows.
For market-watchers, the key questions will be how persistent the asset mix is, whether paper-driven signals translate into profitable live trades after transaction costs and slippage, and how Perpetuals governs the process of converting simulated predictions into executed positions. As the platform matures, transparency about eligibility criteria for users to share upside, the firm’s execution quality, and anti-abuse measures will determine whether UpsideOnly’s model can scale without raising regulatory flags.
Perpetuals has made senior executives available for comment. For now, the first-week figures offer an early data point in the evolving landscape of AI-driven prop trading that leverages retail-sourced signals, but they are not definitive evidence that simulated conviction will remain an accurate predictor of live-market outcomes.
Methodology note: All platform metrics reported here are based on disclosures provided by UpsideOnly/Perpetuals.com for the platform’s first week of operation.
