Some short sellers are seeing opportunity in this tech mania. How they're spotting fake AI stocks

Short sellers are targeting AI-related excesses, including Rezolve AI, which Joyce Meng said could fall 60%, and Nvidia, where Culper Research disclosed a bearish position over China-linked revenue risks. The article highlights speculative AI branding, questionable corporate pivots, and the vulnerability of legacy businesses to AI disruption. While largely thematic, the attacks on Nvidia and AI sentiment could pressure individual names and the broader AI trade.

Analysis

The setup is less about one-off frauds and more about a tightening of the financing window for anything that can be plausibly framed as "AI-adjacent." Once the market stops rewarding label changes and starts demanding proof of compute contracts, gross margin durability, and real customer concentration, the weakest balance sheets become forced sellers of equity rather than growth compounders. That creates a reflexive unwind: higher short interest, lower incremental liquidity, and an accelerating premium for companies with audited AI revenue and genuine hyperscaler exposure.

RZLV and BIRD are vulnerable because they sit at the intersection of narrative inflation and low fundamental credibility. If retail momentum fades even modestly, these names can gap down much faster than fundamentals alone would justify because the shareholder base is likely unstable and event-driven rather than long-duration. The second-order effect is that every failed rebrand raises the discount rate on the next speculative "pivot," which should compress valuations across the long tail of nano/micro-cap AI pretenders over the next 1-3 months.

NVDA is a different trade: this is not a growth-decay short, it is an abuse-of-channel-risk short. Even a small leakage of China demand through intermediaries matters because the stock embeds near-perfect supply utilization and policy immunity; the market is paying for clean, compounding earnings, so any evidence of sanctions circumvention can trigger multiple compression before it changes the income statement. The more important catalyst is not whether the allegation is fully correct, but whether it forces investors to model a lower quality revenue mix and a larger geopolitical risk premium.

The consensus is probably underestimating how asymmetric the downside is in names with retail-owned optionality and overestimating how durable the "AI umbrella" remains once the first few stories break. At the same time, the market may be overconfident that Nvidia's China exposure is binary; if the issue is partial and persistent, it can pressure sentiment for quarters without requiring a dramatic earnings miss. That makes this a positioning-driven tape where the best risk/reward is to short the weakest narratives, not the strongest balance sheets, unless evidence emerges that export leakage is measurable and material.

Key Decisions for Investors

• Short RZLV on any strength; target a 4-8 week window. Risk/reward favors a 20-30% downside move on a sentiment break, with the main risk being another retail-driven squeeze if management leans harder into the AI story.
• Short BIRD as a fade of the rebrand trade; use tight risk controls and treat it as a momentum reversal trade over 2-6 weeks. Best case is a 30%+ retracement if volume normalizes and the market stops paying for the pivot narrative.
• Maintain a tactical short NVDA via put spreads or call overwrites into strength, sized smaller than the smaller-cap shorts. This is a lower-conviction geopolitical hedge: if China leakage headlines gain traction, implied multiple compression could matter more than earnings revisions over the next 1-3 months.
• Pair long quality AI enablers / short fake AI basket: long NVDA or a diversified semiconductor basket, short a basket of AI-label microcaps. This isolates narrative decay risk while keeping exposure to the secular capex cycle.
• Avoid chasing any fresh AI rebrand announcements for the next 30-60 days; wait for confirmation in audited customer disclosures and gross margin data before re-rating. The reward/risk on newly branded names is now skewed toward fast mean reversion.