Aubrey Capital Flags AI Capex Risks; Trimmed Exposure, Eyes India Rotation

Why a Fund Manager Is Trimming AI Stocks and Watching India

Mark Martyrossian, a director at Aubrey Capital Management, is questioning the sustainability of the artificial intelligence infrastructure boom and has begun adjusting his funds’ positioning. While the hyperscale cloud providers continue to ramp up spending—Alphabet’s most recent results showed another increase—Martyrossian sees the Oracle downgrade as a “red flag.” It signals, he argues, that funding may shift from cash-flow financing to debt, a move that could erode the appetite for ever-larger capex commitments.

On the supply side, consensus still points to shortages of memory and other components lasting until at least the second half of 2027. However, the narrative could be disrupted from China. Last week’s release of Moonshot AI’s Kimi 3 agentic model has, in Martyrossian’s view, the hallmarks of another “DeepSeek moment”—it raises doubts about whether paying top dollar for the most advanced chips is money well spent. Chinese chipmakers are also a factor: the recent IPOs of CXMT (whose shares surged 460% on their first day) and YTMC underscore domestic ambitions, even though one analyst estimates China’s capacity would need to quadruple just to meet internal demand, and incumbents SK Hynix and Samsung still dominate the most sophisticated memory.

Against this backdrop, Martyrossian is watching India. The country has been a loser from the AI trade, with foreign institutional investors (FIIs) pulling out $27bn as of mid-July—the lowest net buying since 2016. But India’s economy grew more than 7% in each of the last two quarters, and equities are now considerably cheaper than their September 2024 peak. Net buying of $1.8bn in the first two weeks of August may be the first flicker of a reversal. Aubrey has added modestly to its Indian exposure, while still maintaining a decent position in AI names.

Unpacking the AI Capex Conundrum and China’s Supply Ambitions

The Oracle Warning and the Financing Question

The downgrade of a major enterprise technology company is rarely just a single-stock event. For Martyrossian, it casts a shadow over the assumption that AI capex will keep rising effortlessly. When financing migrates from operating cash flows to debt, corporate treasuries—especially in a high-rate environment—become more selective. The implied risk is that a break in the capex trend would quickly ripple through the semiconductor supply chain and data-centre operators whose valuations have been built on unbroken growth. Crucially, Alphabet’s latest capex increase shows the immediate trend is still intact, but the fund manager’s alert is that the margin for error is shrinking.

China’s Two-Pronged AI Supply Threat

Martyrossian sees a dual push from China. First, large-language models that achieve competitive results with less advanced hardware would challenge the premium pricing that NVIDIA commands on its most powerful GPUs. The Kimi 3 agentic model is the latest example of a possible step-change in capability, echoing the surprise that erupted when DeepSeek demonstrated high performance without cutting-edge chips. Second, on the memory front, CXMT’s explosive IPO and YTMC’s emergence signal a domestic industry that, despite the massive capacity gap identified by analysts, is attracting capital and state backing. While US export controls restrict access to the most advanced tools, they also supercharge domestic self-sufficiency efforts. The interplay means the current consensus on shortages until late 2027 must now be weighed against a supply-side wildcard that could compress the cycle sooner—or alter the competitive landscape permanently.

India as a Risk-Off Rotation Play

India’s stock market has suffered from a powerful gravitational pull towards the US AI boom: $27bn of foreign outflows meant net FII buying hit a multi-year low. Yet the fundamentals present a stark contrast. GDP growth above 7% for two consecutive quarters is among the fastest for any major economy, and valuations have corrected significantly from the September 2024 peak. Martyrossian’s thesis is straightforward: a fatigue in the AI trade would likely trigger a reversal of those flows. The early August data point of $1.8bn of net buying is thin, but it aligns with a broader pattern of late-cycle repositioning. A fluctuating oil price could still temper enthusiasm, but the ground is being prepared for a rotation.

What the Rotation Signal Means for Investor Portfolios

Martyrossian’s commentary points to several concrete signals investors can watch:

• The next round of quarterly results from Alphabet, Microsoft and Meta will show whether hyperscaler capex is plateauing. Any cut to forward spending guidance would directly validate the downturn risk and could mark a broader sector turn.
• A competitive Chinese AI model that performs on par with Western systems while using less advanced hardware would be a critical catalyst—it would question the demand trajectory for NVIDIA’s highest-margin products and could shift capital toward Chinese memory and logic chipmakers.
• The memory supply story remains a tug-of-war between dominant incumbents (Hynix, Samsung) and the emerging Chinese capacity that, if scaled, would compress pricing power. Keep an eye on capacity announcements from CXMT and YTMC, which are the leading indicators of whether the supply shortage ends earlier than expected.
• India’s valuation discount and accelerating GDP make it a candidate for a sharp rebound if the AI trade unwinds. The $1.8bn of net buying in early August is a small but early datapoint; a sustained fortnight of inflows above $2bn would be a stronger signal that the rotation is gaining traction.