NVDA/US stocks/HK investors/TW investors/capital gains/FX risk/portfolio management

NVDA Stock for HK/TW Investors: Why I Trimmed 70% After a +280% Run

Jim LiuMay 2, 202612 min read

Contents

TL;DR

I'm Jim Liu, a Sydney-based developer running LRTS. I bought 50 shares of NVDA at $36 in October 2024 (post-split). By April 2026 I'd trimmed twice — down to 15 shares — locking in roughly $4,300 of realized profit. Current paper value: ~$1,950 at $130.
NVDA stock price sits around $128-135 in early 2026, after the 10-for-1 split (June 2024) compressed psychological barriers and unleashed retail buying. This piece is not "should you buy NVDA" — it's how I actually managed the position as a HK-tax-aware investor in Sydney.
The famous 30% US dividend withholding barely applies here. NVDA pays $0.04/share quarterly — $0.16/year. On 100 shares that's $16 of dividends, $11.20 after withholding. The real tax drag for HK/TW investors is FX risk on capital gains, not dividends. I'll show the math.
I trimmed 70% of my position because position size was distorting my decision quality, not because I think NVDA is overvalued. The remaining 15 shares are house money. Both trims were technically wrong — NVDA kept running. I sleep better anyway.
This is a reasoning piece, not a recommendation. If you're holding NVDA at +500% and asking "should I sell," the answer is in your sleep quality, not in the chart.

Who I Am, Why You Should Listen

I'm Jim Liu. Independent developer in Sydney. I run LRTS (HK IPO + US tax mechanics) and OpenAI Tools Hub.

Broker: IBKR (Sydney USD account) for US stocks; moomoo HK for everything HK-listed
Position scale: HK$1.2-2.5M idle for opportunistic IPO and US single-stock plays
History: Holding US tech since 2018, full-time tracking US single names since 2024
2025 realized gains across the portfolio: roughly USD $9,800 (NVDA was the largest contributor)

I am not a CFA. I'm not running other people's money. Everything below is what I actually did with my own account, including the parts I got wrong.

Why I Track NVDA Specifically

LRTS is mostly Hong Kong IPOs and US tax mechanics. NVDA earned a permanent slot in my tracking spreadsheet for three reasons that aren't the obvious ones:

It's the most-asked-about US stock from LRTS readers — by far. More questions than INTC, TSLA, and AAPL combined. Every email is some variant of "I'm sitting on +200% in NVDA, what do I do." This piece is largely a public answer.
It's a clean test case for "letting winners run vs locking in profits." I held NVDA from $36 to $135. I trimmed twice. I have actual data on what trimming felt like and what it cost me. Most people who write about this never actually did it.
The post-split (June 2024) retail dynamics are uniquely studyable. A 10-for-1 split changes nothing fundamental but changes everything psychological. I'll show how the retail interest curve reshaped my own decision-making.

I deliberately bought a small NVDA position in 2024 — 50 shares, ~$1,800 — partly to feel the friction firsthand: IBKR commission flow, T+2 settlement on partial sells, FX conversion timing on USD proceeds back to HKD or AUD. Everything below is first-hand.

My NVDA Position Story (50 → 35 → 20 → 15)

I'll skip the table here on purpose — most NVDA timelines you've read on Reddit are someone's screenshot of green numbers without the trim discipline. Mine in prose:

October 2024: Bought 50 shares at $36 ($1,800 total). Post-split price had been bouncing between $32 and $40. I picked $36 because it was the 50-day moving average. No deep thesis. I wanted exposure.

June 2025: Sold 15 shares at $115. Took $1,725 off the table — basically my entire entry capital recovered, with $35 of slippage. Remaining 35 shares were free.

The reason I trimmed wasn't macro. It was sleep quality. At 50 shares × $115 = $5,750, NVDA had become 12% of my US sleeve. A bad earnings night could wipe out a month's worth of consulting income. I hate that feeling.

January 2026: Sold another 15 shares at $135, after the $4 trillion market cap headlines. Took out another $2,025. By this point my realized profit on the original $1,800 was approximately $4,325, with 20 shares remaining at $135 = $2,700 paper.

April 2026: Sold 5 more at $128 — a small "rebalance" sell after NVDA pulled back 8% on China export-control news. Took out $640. Remaining 15 shares.

So: $1,800 in. $4,390 out (after IBKR fees). 15 shares remaining at $130 ≈ $1,950 paper. Total = $6,340 against $1,800 entry = +252%.

The honest take: had I done nothing, my 50 shares × $130 today = $6,500 paper. I'm down ~$160 vs the buy-and-hold case, plus capital gains tax on the realized portion. My trims technically lost me money. I still don't regret them.

The 30% Withholding Question (Why It Barely Applies to NVDA)

Most US dividend withholding articles treat all US stocks identically. They're wrong for NVDA:

Stock	Annual dividend per share	Per 100 shares (HK W-8BEN)	Real drag
AAPL	$1.00	$100 → $70 after 30%	-$30/yr
INTC (suspended)	$0 (was $2.00)	$0 (was $200 → $140)	$0 currently
TSM ADR	$1.85	$185 → $129.50	-$55.50/yr
NVDA	$0.16	$16 → $11.20	-$4.80/yr

On 100 NVDA shares, you lose $4.80/year to withholding. That's a rounding error. The dividend tax framework is the wrong lens for NVDA.

The real US-stock tax exposure for HK/TW investors is what happens when you eventually convert USD proceeds back to HKD or TWD. I sold 15 NVDA shares at $135 in January 2026 — $2,025 USD. If I converted that to HKD on the same day at 7.78, I get HK$15,755. If I'd waited until April when USD/HKD slipped to 7.74, that same $2,025 becomes HK$15,673. An 0.5% FX move on the conversion day costs you more than three years of NVDA dividends.

For HK investors specifically, the HKD peg to USD is loose enough at the band edges to matter. I now batch USD-to-HKD conversions to days when the rate prints below 7.76. It's not market timing — it's just not converting on the worst day of the month.

For deeper US dividend mechanics that do matter on AAPL/TSM/JNJ-style holdings, see US dividend withholding tax for HK investors.

Two Numbers I Watch on Every NVDA Earnings (and One Vibe Check)

I don't lead with EPS or revenue. My framework for NVDA — different from how I look at INTC:

Number 1: Datacenter revenue YoY growth rate

This is now ~88% of NVDA total revenue. The market is implicitly pricing 60-90% YoY datacenter growth indefinitely. The week growth slips to "merely" 30-40% YoY is the week the multiple compresses. Watch for: deceleration that looks orderly vs. a cliff.

Number 2: Customer concentration disclosed in 10-Q

Three customers (Microsoft, Meta, Google parent) are roughly 40% of total revenue. NVDA used to disclose this in the risk factors — now it's mentioned in earnings call commentary. When concentration breaks above 50%, that's a flag. Below 35%, it means inference workloads are democratizing the buyer base, which is structurally bullish.

Vibe check: how Jensen sounds

This is unscientific and I'm slightly embarrassed to write it. But four years of listening to Jensen Huang earnings calls — when he switches from "we're scaling" to "we're navigating supply" or "the customer is asking us about" — the language tilt usually predicts the next quarter's tone. Earnings call audio is on YouTube same evening; it's a 75-minute time investment that's been more useful to me than the slide deck.

EPS beats and misses are noise. These two-and-a-half are signal.

Datacenter vs Robotics: Which Catalyst Slows First

The NVDA bull case has two distinct horizons:

Near-term (12-18 months) — datacenter: hyperscaler capex commitments are public through 2027. Microsoft alone is at $80B+ AI capex. The risk isn't demand, it's energy — Northern Virginia is at grid limit, Texas just denied three datacenter applications. Real threat: an energy-driven slowdown that NVDA can't engineer around.
Long-term (3-7 years) — robotics + automotive: Jensen's set-piece slide. Real revenue today: <2% of total. Real revenue in 2030 if the thesis holds: maybe 15-20%. Real threat: the 5-year gap between today and that revenue is when valuation has to digest reality.

My read: datacenter does not slow in 2026, but the narrative gets challenged by the energy story sometime in H2 2026. That's the wobble window. Robotics doesn't matter until 2028 at earliest. If you hold NVDA primarily for the robotics narrative, you're paying today's multiple for 2028's revenue. That math doesn't work for me.

For comparison with broad-market alternatives, see VWRA vs VOO vs VT for HK/TW investors.

What 4 Trim Decisions Taught Me (Not 4 Reasons I'm Bearish)

I want to flip the usual framing. NVDA is not the kind of stock where "4 reasons not to overweight" makes sense — the position has been correct for three years and counting. What I actually learned, from the trims themselves:

Position size shrinks decision quality. When NVDA was 12% of my US sleeve, I was checking the price four times a day and reading every Stratechery piece on hyperscaler capex. After trimming to ~6%, I check once a day and read whatever I want. The first trim paid for itself in my own focus.
The 10-for-1 split (June 2024) changed retail behavior more than analyst models. Pre-split NVDA at $1,200 felt unbuyable to retail. Post-split at $120 felt cheap. Same company. I underestimated this. My June 2025 trim happened because I thought split-driven retail was peaking. It wasn't — retail kept piling in for another six months.
FX conversion timing matters more than entry/exit timing on a stock this volatile. I wrote about this above — a 0.5% USD/HKD move costs more than three years of NVDA dividends. I never thought about FX before holding US single names. I think about it weekly now.
I'm probably wrong about one of these. Specifically: my "vibe check" Number-2.5 above is the thing most likely to look stupid in retrospect. Tone-reading earnings calls might just be confirmation bias dressed up as research. I keep doing it because the cost of being wrong is small (75 minutes), but I want to flag it as the weakest leg.

FAQ

Should HK/TW investors buy NVDA at $130?

I can't answer this. What I can answer: if you'd be uncomfortable with NVDA dropping 30% in a single overnight session (it's done worse), your position size is too large. Most retail investors I talk to are 8-15% NVDA without realizing they've made a concentrated single-stock bet on AI capex.

Is NVDA in a bubble?

Probably partially. The datacenter revenue is real — that part isn't a bubble. The implicit assumption that NVDA captures 80%+ of AI accelerator spend through 2030 is the bubble part. AMD MI300X is closer to parity than the stock price reflects. Custom silicon (Google TPU, AWS Trainium) is also real. The bubble is in the assumption of monopoly, not in the demand.

For HK/TW investors holding +500% NVDA, when should I sell?

The honest answer: nobody knows. The practical answer: trim enough that one bad earnings night doesn't change your year. For me that's 30-50% off the table after a 3x. Your number depends on your other income. If NVDA gains are 80% of your year's investment income, you're already over-concentrated and the question isn't "when" but "how much, today."

IBKR or moomoo HK for buying NVDA?

For US stocks, IBKR is materially cheaper. On 100 NVDA shares at $130 = $13,000:

IBKR (Sydney/HK USD account): $0.35 commission, no spread
moomoo HK: ~0.03% commission + ~0.05% FX spread = ~$10 round trip

For one trade the difference is meaningless. For frequent trimming, IBKR wins clearly. See IBKR vs moomoo HK comparison.

How I Calculated These Numbers (Methodology)

Sample: my own IBKR account from October 2024 to April 2026, all NVDA buy/sell transactions. 6 transactions total (1 buy, 4 sells, plus the dividend reinvestment auto-buy of fractional shares which I excluded for simplicity).

Price source: IBKR statements (executed prices net of commission) cross-checked with Yahoo Finance daily close on the same date.

FX: USD/HKD historical from HKMA daily fixings. USD/AUD from RBA noon rates (I'm Sydney-based, so my conversion currency is AUD, but I converted the math to HKD for this article since LRTS readers are HK-centric).

Position sizing percentages: my US sleeve total value at trim dates (private but the relative percentages are accurate).

Affiliate disclosure: LRTS earns referral commissions from IBKR and moomoo (full details on the Affiliate Disclosure page). The conclusion of this piece — I trimmed 70% of my NVDA position — is neutral-to-negative for my referral revenue (people who trim trade less). I'm not pushing you into more NVDA exposure.

This is not investment advice. Single-stock semi positions can swing ±25% on earnings night. NVDA specifically has done -10% intraday at least 4 times in the past 18 months. Retail position limits should be 5-8% for any single name. If you don't have a 3-year time horizon, don't hold concentrated single names. Trade at your own risk.