Mirae Asset Global Investments Co., Ltd.

The “biggest buys” list reads like a who’s who of the AI supply chain, not of end-user apps. Mirae is shifting incremental dollars from broad tech winners into the most operationally leveraged beneficiaries of AI compute demand.

• GOOG: Even after trimming GOOGL, they surge the non-voting GOOG line by +19088.4%, adding about $706.2M. That’s a structural preference for Alphabet at the share-class level, not a pullback from the company’s AI monetization story.
• KLAC: KLA explodes higher by +1116.2% in shares, roughly $506.5M more, despite being about -61.3% versus their average cost. They are averaging down into process control, effectively saying yield management is non‑negotiable in the next node race.
• TSM: Taiwan Semi jumps +805.2% in shares, adding about $459.9M. This is a direct bet that wafer capacity and advanced packaging are the scarce choke points in AI compute.
• ASML: A +6097.9% increase, about $340.6M more, states the obvious: if EUV is the bottleneck, own the bottleneck.
• AMD, NVDA, MU: Larger adds to AMD (+32.9%), Nvidia (+8.0%), and Micron (+16.7%) extend the theme from tools into memory and accelerators.
• BRK.B: A $208.4M ramp in Berkshire suggests they want an internally diversified, cash-gushing anchor alongside this concentrated AI stack.

Layered on top is a big lift in SOXX (+185.7%), plus fresh size in ARM and incremental AMAT, LRCX, and ADI. Together, this is a full-system expression: logic, memory, foundry, tools, and the index wrapper around them.

If adds are all about AI plumbing, the sells tell you what Mirae no longer needs to own in size. The pattern is clear: trim liquid, broadly owned winners and lower‑conviction growth to fund much sharper factor and supply-chain bets.

• GOOGL: A -29.2% cut and roughly -$636.0M out of GOOGL looks like classic position management. They’re up 92.7% versus cost, and they rotate exposure into GOOG plus higher-operating-leverage hardware instead of walking away from Alphabet entirely.
• IVV and IEMG: IVV is down -17.0% (about -$51.6M), IEMG -13.5% (about -$32.8M). These are pure liquidity reservoirs — broad beta trimmed to pay for semis and Berkshire without blowing up single-name risk in tech.
• QCOM: Qualcomm sees an -8.2% reduction (about -$31.7M) even though it’s modestly profitable at +10.4% vs cost. Relative to Nvidia, AMD, and the tool chain, handset‑centric RF and SoCs simply offer less torque to the AI datacenter buildout.
• PLTR and NFLX: Palantir is cut -12.1% (about -$25.4M), Netflix -6.4% (about -$17.7M). Those trims look like pruning more speculative or fully‑valued growth to keep the AI bet focused on compute and manufacturing rather than software narratives.

The message: when everything is working, Mirae sells what everyone else can own — mega‑cap platforms and index beta — to double down on what fewer investors have the stomach to concentrate in: cyclical, capital-intensive AI infrastructure.

On the surface, sector weights barely move: Technology ticks up to 68.85% from 68.25%, while Consumer Discretionary, Health Care, and Finance each slip by a few basis points. Under the hood, the book is getting more binary.

Within Technology, the center of gravity is racing toward semiconductors and equipment. Nvidia, Micron, Broadcom, AMD, TSM, ASML, KLAC, LRCX, AMAT, ADI, TXN, Intel, Marvell, ARM, and Seagate dominate, complemented by the SOXX ETF. Software heft remains via Microsoft, Alphabet, Meta, and Palantir, but incremental capital is going into fabs, tools, and memory rather than ad platforms or consumer apps.

Outside tech, they quietly reinforce a barbell of defensives. Adds in Walmart, Costco, Home Depot, PepsiCo, Coca‑Cola, and Procter & Gamble (despite a flat mark-to-cost) in consumer, plus larger stakes in Eli Lilly, Amgen, Merck, UnitedHealth, and Johnson & Johnson in health care, give the portfolio durable earnings streams that can coexist with a semi cycle.

Unclassified exposure — effectively ETF wrappers and Berkshire — edges up to 7.13% from 6.8%. Berkshire’s jump, mixed with VOO, SPYM, and SOXX shifts, creates a buffer that dilutes single‑name drawdown risk without sacrificing the AI theme. Energy (Chevron), basic materials (Linde), telecom (Cisco, Verizon), and industrials (Tesla, Rocket Lab) stay small but rising, adding just enough cyclical and idiosyncratic exposure to keep the book from being purely a semi factor bet.

This 13F paints a manager convinced that the AI infrastructure trade is still under-owned, even after a powerful run. By pushing capital into the most cyclical, operationally leveraged pieces of the stack — lithography, process control, foundry capacity, memory — Mirae is saying the bottleneck economics will trump fears of a classic semi down-cycle.

The swap inside Alphabet from GOOGL into GOOG, alongside growing Microsoft, Amazon, Apple, and Meta positions, suggests they still want core platform exposure, but as a funding source it now ranks below the chips that actually power those platforms. The modest but broad-based trimming of ETFs indicates they will keep harvesting beta to underwrite more concentrated, high‑active‑share bets.

On the risk side, they seem keenly aware of how much cyclicality they’ve bought. The simultaneous build‑out of Berkshire, consumer staples, big pharma, and managed-care is not accidental; it’s a conscious effort to bolt a quality, cash‑flow floor under an otherwise high‑octane book.

Looking forward, expect more of the same: incremental dollars rotating toward whichever pieces of the AI buildout look most constrained — whether that’s high‑NA EUV, HBM memory, or advanced packaging — with ETFs and lower‑leverage growth serving as the cash machine. Unless the AI capex story structurally breaks, Mirae’s filing argues they’d rather be early and concentrated in the plumbing than diversified across the apps.