AI Shorts Spread Beyond Chips

Securities-lending data shows AI short activity widening across software, robotics and infrastructure, but bearish conviction is becoming more selective.

Artificial intelligence has moved from a concentrated investment story centred on a handful of US technology leaders into a global economic theme spanning semiconductors, servers, storage, data centres, software, robotics and power infrastructure. Securities-lending data suggests that short sellers are following the same path, but not through uniform positioning against AI. Instead, investors are increasingly testing where AI-related investment can translate into sustainable returns and where valuations, margins, funding needs or capacity plans may be more vulnerable.

This broadening reflects the growing economic footprint of AI, as set out in S&P Global Market Intelligence’s analysis of AI-related investment, exports and manufacturing activity[1]. That analysis indicates that AI-related investment is supporting US business spending, while demand for semiconductors, storage, servers and technology equipment is strengthening exports across Asia. Economies central to AI development, including the US, mainland China, Japan, Taiwan, South Korea and the Netherlands, have outperformed the global manufacturing benchmark since late 2025. Technology equipment also recorded the fastest expansion in output and new orders of any global sector in June.

That widening economic footprint is now visible in securities lending data, where revenue generation is appearing across less obvious parts of the AI value chain. During the first half of 2026, high-revenue generating positions included application-layer businesses such as SoundHound AI (SOUN), which generated $47.8 million from an average loan balance of $1.39 billion, and smaller specialists such as Brand Engagement Network (BNAI), where limited lendable supply contributed to elevated average fees. Exposure also extended into autonomous driving and robotics through Kodiak AI (KDK) and Serve Robotics (SERV), Asian hardware through Horizon Robotics (9660 HK), specialist processors through Cerebras Systems (CBRS), quantum computing through Quantum Computing Inc (QUBT), and data-center infrastructure through Applied Digital (APLD).

By August, this dispersion was even more visible among the leading revenue-generating securities. Z.Ai Co Ltd (2513 HK) produced $47.7 million of lending revenue from a $1.11 billion loan balance, while MiniMax Group (0100 HK) and Horizon Robotics (9660 HK) generated $2.93 million and $3.33 million respectively. Beyond software, the revenue list included cloud infrastructure names such as Nebius Group (NBIS) and CoreWeave (CRWV), specialist processors such as Cerebras Systems (CBRS), autonomous-vehicle companies including Pony AI (PONY) and WeRide (WRD), and quantum-computing exposures such as Horizon Quantum Holdings (HQ), IQM Quantum ADR (IQMX) and Quantum Computing Inc (QUBT).

Asian hardware is a particularly important channel for this activity. In August, Taiwanese server manufacturers Quanta Computer Inc (2382 TT) and Wiwynn Corporation (6669 TT) generated approximately $11.4 million and $11.1 million of lending revenue. The list also included Gigabyte Technology Co Ltd (2376 TT), WT Microelectronics Co Ltd (3036 TT), Realtek Semiconductor Corp (2379 TT), Hanmi Semiconductor Co Ltd (042700 KS), GlobalWafers Co Ltd (6488 TWO) and Jusung Engineering Co Ltd (036930 KQ). This extends short positioning from chip designers into server assembly, memory, testing, wafer supply, semiconductor equipment and electronic components.

That geographical expansion mirrors the underlying economy. AI-related exports are supporting manufacturing activity across Asia Pacific, but the strength of that growth is not without risk. The same S&P Global Market Intelligence report referenced previously, notes that recent semiconductor export gains may be driven more by prices than volumes, while shortages, increased freight costs and longer delivery times remain potential constraints. These conditions create opportunities for dispersion trades: investors can remain positive on structural AI demand while potentially positioning against suppliers whose valuations, margins or capacity plans appear vulnerable to normalising prices or slower unit growth.

That wider stock-level activity should not be confused with a broad acceleration in bearish conviction. The more important signal is dispersion: short sellers appear to be extending their focus across the AI ecosystem even as aggregate short-interest measures have weakened. North American technology hardware short interest, measured as the percentage of market capitalization on loan, fell from approximately 0.73% at the start of January to 0.28% by 11 September. Software and services declined from around 0.84% to 0.69%, despite rising temporarily above 1.5% in June. Asian semiconductor short interest also ended at approximately 0.65%, below the 0.69% recorded at the beginning of the year and well below the levels seen during parts of 2025.

Fee movements tell a similar story. Average lending fees declined across several previously high-cost AI-related names during August, even where loan balances remained meaningful. SoundHound AI (SOUN), Datavault AI (DVLT) and MiniMax Group (0100 HK) all saw borrow costs ease from elevated first-half year levels. Lower fees do not remove the significance of those positions, but they do suggest that borrow pressure has become less concentrated in some of the earlier high-profile trades.

The resulting picture is therefore one of greater breadth but lower uniformity. Short sellers are no longer looking only at prominent AI software companies or leading chip designers. They are examining where value is captured across the entire investment cycle, from advanced processors and memory to servers, networking, data-centre capacity, robotics and application software. They are also differentiating between liquid incumbents, where fees remain low, and emerging companies where scarce supply can still produce exceptionally high borrowing costs.

This selectivity may increase as the AI cycle matures. The S&P Global Market Intelligence report states that only 46% of enterprise AI initiatives launched over the past year are considered on track to achieve a positive return within 12 months, while just 37% are live and delivering value. Against that backdrop, securities-lending activity indicates neither a wholesale rejection of AI nor an indiscriminate short squeeze. It points instead to a more developed market in which investors are testing which companies can turn investment, capacity and technological promise into sustainable returns. In securities lending, the AI trade is no longer simply about being long or short the theme; it is becoming a test of which parts of the ecosystem can justify the capital now being committed to them.