Arrowpoint Sells $8.94M XPeng Stake as FastDriveVLA AI Secures AAAI Acceptance
Arrowpoint Investment Partners divested its entire 500,000-share XPeng stake, offloading roughly $8.94 million (8% of its AUM) based on quarterly average prices, potentially signaling profit-taking pressure. XPeng’s FastDriveVLA visual-token pruning framework achieved a 7.5× computational load reduction and earned AAAI 2026 acceptance (17.6% rate), underscoring its AI-driven autonomous driving capabilities.
1. Arrowpoint Exits XPeng Stake
According to a Nov. 13, 2025 SEC filing, Arrowpoint Investment Partners (Singapore) sold its entire 500,000-share position in XPeng Inc., representing an $8.94 million change in value based on quarterly average prices. That stake had accounted for roughly 8% of the fund’s assets under management in the prior quarter and now stands at 0%. Following the exit, Arrowpoint’s top equity holdings include American Equity Investment Life (15.4% of AUM), EchoStar Satellite Services (3.8%), Vanguard Short-Term Corporate Bond ETF (3.4%), Alabama Power (3.4%) and ATA Tactile Systems (2.7%). The sale timing within the quarter is not specified, though XPeng shares peaked at nearly $25 during Q3, suggesting the fund may have locked in significant gains.
2. XPeng–Peking University Research Accepted at AAAI 2026
XPeng and Peking University’s joint paper, “FastDriveVLA: Efficient End-to-End Driving via Plug-and-Play Reconstruction-based Token Pruning,” was accepted by AAAI 2026 out of 23,680 submissions, a 17.6% acceptance rate. The framework reduces onboard computational load by 7.5×—trimming visual tokens from 3,249 to 812—while preserving high planning accuracy on the nuScenes benchmark. FastDriveVLA’s adversarial foreground-background reconstruction strategy mimics human drivers’ focus on critical scene elements, outperforming prior text-visual attention and token-similarity methods. This marks XPeng’s second top-tier AI conference recognition in 2025, following its CVPR WAD presentation in June and November’s unveiling of VLA 2.0, underscoring the company’s full-stack capabilities in advancing scalable Level 4 autonomous driving solutions.