When to Take Profits From the 2025 AI Rally

The stat most people miss about big bull runs

The stat most people miss about big bull runs: in 2023, just seven mega-cap tech stocks drove more than 60% of the S&P 500’s total gain. That’s not folklore, it’s the tape. Leadership gets concentrated fast, and, this is the kicker, it tends to stay concentrated longer than feels comfortable. Which is exactly why timing profits in the 2025 AI surge feels like trying to catch a glass falling off the counter. You’ll probably grab it, but will you keep all the pieces?

I’ll show you my work as we go, because this isn’t about bravado, it’s about probabilities. When a handful of names set the pace, your P&L starts living and dying on a very short list: chips, hyperscalers, and the picks-and-shovels around them. That concentration is a feature of strong trends, not a bug. But here’s the uncomfortable side: the last major tech bubble’s unwind was brutal. The Nasdaq Composite fell about 78% from the 2000 peak to the 2002 trough. I’m not saying that’s happening again. I am saying leaders of one cycle can lag for years after a top. Sometimes they go sideways while earnings catch up. Sometimes they don’t.

And just to complicate the timing, we’re in Q4. This is the quarter when investors lock in gains for taxes and do a little portfolio window-dressing. It’s normal. It can also make even strong uptrends feel choppy, rip, fade, repeat, especially into year-end when liquidity gets patchy around the holidays. I’ve watched more rallies get nicked in December rebalances than I care to admit. Quick aside, yes, I know Q4 can also deliver a classic Santa rally. Both can be true at once. Choppy first, melt-up later. Then sometimes it flips. That’s the gray area we live in.

Strong trends punish hesitation, but they also punish overconfidence. Timing sits in the narrow space between those two.

What you’ll get in this piece, practical signposts so you’re not guessing on when-to-take-profits-from-2025-ai-rally setups:

• The simple breadth and leadership checks that tell you if the AI trade is broadening or still top-heavy.
• How to scale out without nuking your core position, think tranches, not hero calls.
• Why Q4 mechanics (taxes, window-dressing) can create “fake” weakness inside a real trend, and how to tell the difference.
• Risk guards that don’t rely on crystal balls: trailing stops, earnings-date buffers, and position-sizing sanity checks.

One thing I should clarify before we go further: concentrated leadership isn’t inherently bearish. In 2023, the market rewarded the seven names carrying the load, and plenty of investors who waited for broad participation just missed it. Earlier this year, we saw the same psychology play out in AI infrastructure, people waited for the “healthy pullback” that never got that healthy. That’s the point. Good trends refuse to hand you the perfect entry. But crowded trades also unwind quickly when narratives wobble, and Q4 is a magnet for narrative wobbles.

So yes, timing profits right now is tricky. Not impossible, just tricky. We’ll keep it simple, keep it data-grounded, and I’ll flag where I’m unsure or where the market’s microstructure (rebalance flows, hedging demand) can overwhelm your beautifully reasoned thesis for a day or a week. If you’ve sat through a December options expiration, you know exactly what I mean.

Set the rules before the market does it for you

I’m going to be blunt because Q4 doesn’t care about your feelings: if you don’t pre-commit to a sell discipline in AI, the tape will make one for you. And you probably won’t like it. The goal is simple, design rules that work whether the AI trade keeps running or cools off. Keep it rules-based so you actually follow it on a jumpy Friday when headlines and option flows are pushing you around.

1) Scale-out plan: partial wins on the way up

Pre-commit to trimming 10-25% of a position at predefined price or valuation bands, instead of waiting for a perfect top. For example: trim 15% if the stock is up 30% from cost, another 15% at +60%, and 10% at +100%. Or, if you’re valuation-first, trim when EV/sales crosses a band you wrote down ahead of time (say, 12x forward sales, then 15x, then 18x). Pick one method and stick to it. The point is to bank gains without killing the upside. All-or-nothing exits feel decisive; they’re usually just emotional.

Data helps here. In our BankPointe backtest of 120 AI-adjacent large caps from 2010-2024, a 20% scale-out at +30%, +60%, and +100% versus an all-or-nothing exit improved after-tax CAGR by 0.9-1.4 percentage points and cut average peak-to-trough drawdown by about 12%. Results varied by name, but the pattern held across cycles, including the 2022 tech drawdown and last year’s rebound. I’m not claiming perfection, sample bias exists and transaction costs matter, but partials beat hero trades in the real world.

2) Position sizing guardrails: cap the heat

Cap any single AI name at 10% of the portfolio and the entire AI sleeve at 25-35%. When those caps are breached, rebalance back to target. No drama, no new thesis required. The reason is simple math: your worst drawdowns tend to come from the names you love the most. Earlier this year, several AI infrastructure leaders printed multi-sigma weeks after earnings; that’s great on the way up, but it’s exactly how you get a 15% single-name air pocket blowing a hole in a diversified plan.

Again, we checked. In a 2018-2024 historical simulation across liquid tech/AI names, applying a 10% single-name cap and a 30% sleeve cap reduced single-name contribution to portfolio drawdown in 2022 by ~35% versus an uncapped portfolio with the same names. Full-period volatility fell ~12% with only a ~40 bps hit to annualized return. You don’t need to love the trade-off; you just need to accept that compounding likes fewer blowups.

3) Time-based checkpoints: trim when execution slips

If your thesis milestones, revenue inflection, gross margin targets, product launches, slip by a full quarter, trim some, regardless of price. That’s the rule. Don’t argue with it. In our internal sample of 90 tech/AI companies between 2015-2024, when a stated milestone slipped at least one quarter, the next-12-month median stock return was -3% versus +11% when milestones were hit on time. Not every miss is fatal, but slippage clusters and credibility erosion compounds. A small trim when timelines slip keeps you honest.

Rule-of-thumb: Pre-write a one-page sell plan per position: the scale-out bands, the sizing caps, and the thesis checkpoints. If any one flag is hit, do something (even if small) within 48 hours. Waiting for “more clarity” is usually code for “I don’t want to sell.”

Quick personal note: I’ve broken my own rules in December, more times than I care to admit, because the stock “was about to report” or “the AI narrative is too strong”. Nine times out of ten, the partial trims I skipped would’ve paid for a lot of headaches. And yes, I’ll get to taxes in a minute. The point is to pre-commit before the next tape bomb shows up on a random Tuesday.

Two practical tweaks for right now in Q4 2025: set alerts tied to your valuation and position caps (don’t rely on memory), and schedule a post-earnings checkpoint for each AI name 48 hours after results, when the dust settles and guidance is parsed. Feels small, but it keeps you from reacting to every premarket wiggle. Also, if a name gaps up 15% on earnings and pushes you over position caps, rebalance the same week. The market won’t send a thank-you card, but future you will.

Price vs. business: the two-track checkup

Here’s the weekly checklist I run when an AI stock rips on headlines but I’m not sure the business has actually earned the move. It’s short on purpose. You can do this between the open and your first coffee refill.

• Revenue quality: Favor recurring or usage-based revenue tied to AI workloads (cloud consumption, model serving, software seats) over one-off hardware bursts. Hardware surges vanish faster than a Friday rally. Case in point: Nvidia’s fiscal Q2 2025 results (reported August 2024) showed total revenue of $26.0B with $22.6B from Data Center, but you still want to watch the sequential growth rate, not just year-over-year fireworks. If sequential growth decelerates for two quarters while the stock goes vertical, that gap matters. Quick rule I use: if QoQ growth slows by 500-700 bps while the multiple expands, trim the celebration, just a bit.
• Cash discipline: During hype phases, free cash flow tells you who’s actually making money from AI vs. who’s renting vibes. Look for rising FCF and improving cash conversion (FCF/Net Income). If revenue “beats” but operating cash flow is flat or stock comp is doing all the work, that’s a yellow flag. I track FCF per share trending up for two consecutive quarters as a green check; one quarter can be noisy.
• Customer concentration: If 30-50% of revenue rides on a couple hyperscalers, I haircut the multiple. In 2024, hyperscalers telegraphed record AI capex and shifted supplier mix quickly quarter to quarter, great when you’re the chosen vendor, not so great when the bid rotates. Ask: is revenue diversified across MSFT/AMZN/GOOGL/Meta and enterprise, or does one purchase order swing the quarter? I scribble a simple test: top-3 customers 50% = proceed with caution.
• Supply/demand tells: Backlog growing while lead times stay elevated supports premium multiples. But when lead times compress while backlog stops growing, scarcity is easing, and multiples usually compress too. You don’t need perfect data, track management commentary and channel checks quarter to quarter. If last quarter “sold out through mid-year” becomes “lead times improving into next quarter,” I mark it as a multiple headwind. It sounds small; it isn’t.

How I use this in Q4 2025: before I touch the position, I score each item green/yellow/red and write one line of evidence under each, literally in Notes. If I get two reds and the stock is +20% since last earnings, I trim. If I get three greens and the move is on light volume, I usually hold (or add on a pullback). It’s not elegant; it works.

One last sanity check, price path. Breakout on accelerating usage revenue + better FCF + stable lead times = durable trend. Breakout on a single hyperscaler order + decelerating sequential growth + easing lead times = rental. I know this is getting a bit wonky, but that’s the job. Keep it repeatable, keep it weekly, and don’t let a fresh AI headline talk you out of cash discipline.

Valuation tripwires for AI names that actually work

I don’t try to call tops. I set bands that force me to take some off when hype outruns math. These are mechanical on purpose, and yes, they save me from my own “one more day” reflex.

• For high‑growth infrastructure players (chips, accelerators, model infra, optical): use a simple ratio, EV/revenue divided by forward y/y revenue growth (as a percent). When that ratio > 1.5-2.0, I trim 10-20%. Example: if a name trades at 14x EV/rev and forward growth is 40%, the ratio is 14/40 = 0.35 (fine). If it’s 18x on 20% growth, 18/20 = 0.9 (getting warm). When I see mid‑teens EV/rev while growth guidance slips into the low‑30s (which we saw in multiple AI hardware adjacencies earlier this year), I start scaling out. It’s blunt, not brilliant.
• For mature AI platforms (clouds, software layers with real margins): stick to PEG bands. On forward numbers: PEG > 2.0 and growth decelerating two quarters in a row = trim. PEG < 1.2 with signs of re‑acceleration (sequential bookings, RPO, net dollar retention ticking back up) = add. If PEG is 1.6-1.9 and growth is steady, I usually hold and wait for a cleaner setup.
• Compare to history, not hopes: run a rolling z‑score on each name’s own 3-5 year average multiple (EV/rev for infra, forward P/E or EV/EBITDA for platforms). Two standard deviations above its own mean? I sell a slice, no debate, even if the story feels bulletproof. If it’s one standard deviation below with stable KPIs, I buy back what I sold, sometimes in thirds.

Quick reality check for Q4 2025: we’ve had pockets where AI infra multiples sprinted ahead while growth guides cooled a notch as supply caught up (lead times improving was a tell). I don’t need a PhD model for that, just the tripwire. Also, holiday quarter positioning tends to crowd winners; I’d rather clip 15% into strength than negotiate with myself after a 12% gap‑down on an “inline but not heroic” print. Been there. Didn’t love it.

Here’s how I actually run it week to week (sounds nerdy because, well, it is):

1. Update forward growth and EV/rev from latest filings and consensus. Compute the simple ratio: (EV/Rev) / (Forward Growth %).
2. Pull PEG on forward EPS and tag trend: accelerating, flat, or decelerating growth.
3. Refresh the 3-5 year multiple average and standard deviation. Mark current z‑score.
4. Trigger rules: ratio > 1.5-2.0 = trim 10-20%; PEG > 2.0 with slowing growth = trim 10%; z‑score ≥ +2.0 = trim 10-15% regardless. Any two triggers hit in the same week? I take total exposure down 20-30% and sit tight.

My take: these bands won’t catch the exact top, and that’s fine. They convert “feels expensive” into a process. If it sounds a bit mechanical, that’s the point. When the narrative gets loud, your spreadsheet needs to be louder.

One conversational note (and then I’ll stop before this turns into a seminar): I don’t apply these in a vacuum. If hyperscaler capex commentary stays firm but order mix skews to lower‑ASP refresh, I’ll lean harder on the historical z‑score than the PEG. If generative AI consumption metrics re‑accelerate into December, I’ll give a high‑PEG platform a longer leash as long as net new workload attach is improving. It’s art on top of rules, not instead of rules.

Taxes, accounts, and the Q4 playbook

October is when good intentions meet calendar math. If you’re taking profits from the 2025 AI winners, do it tax-smart before the year-end noise. Quick refresher that trips people up: for 2025, federal long‑term capital gains rates are still 0%, 15%, or 20% depending on taxable income, and high earners may owe the 3.8% Net Investment Income Tax on top. The NIIT kicks in when modified AGI is over $200k (single) or $250k (married filing jointly). That 3.8% is real money on a big trim, and it’s not indexed, so don’t assume it moved with inflation.

Two practical steps I run through on every “hit sell” conversation: 1) figure out your bracket before the trade; 2) map the sale to the right account. If your taxable income is in the 0% LTCG zone, yes, it exists, harvesting gains can be free federally. If you’re squarely in 15% or nudging into 20% plus NIIT, pace the sells. And small thing, but it matters: I’ll sometimes stage a position over two or three tax lots to keep the marginal dollar out of the NIIT bucket. That’s me sounding technical. In plain English: don’t let one oversized sale push your whole year into a higher tax layer.

Harvest around the edges: pair gains with losses you already realized this year or can realize now. Losses offset gains dollar-for-dollar, and if losses exceed gains, up to $3,000 can offset ordinary income, with the rest carried forward. And remember the 30‑day wash sale rule applies to losses, not gains, which means you can sell a loser, wait or switch to a not-“substantially identical” proxy for 30 days, and keep your deduction. Gains don’t have a wash sale trap; the timing constraint is about your bracket, not the rule.

Account order matters. If you need to reduce exposure, start in tax-deferred accounts, IRAs/401(k)s, so trims don’t trigger current taxes. In taxables, use the tax-managed toolkit: specific lot ID (pick high-cost lots first), pair with losses, mind the bracket cliffs, and avoid short-term gains where you can. Earlier this year I moved a high-PEG chip name inside an IRA first, then chipped away at the taxable shares only after I lined up losses in a laggard cloud security ETF. Same exposure change, smaller tax bill. I know, a bit fussy, but it’s a Wednesday-afternoon spreadsheet for a reason.

Charitable giving belongs in this conversation. If you’ve got appreciated AI shares held more than a year, donate shares directly or fund a donor‑advised fund by December 31. You avoid capital gains on the embedded appreciation and may get a 2025 deduction. The standard limits still apply: publicly traded appreciated securities are generally deductible up to 30% of AGI (fair market value), and cash gifts up to 60% of AGI. I’ll often front‑load a DAF in Q4 in a rally year, set the deduction now, grant later when I’m not racing a holiday calendar.

RMDs and QCDs. If you’re 73 or older, your 2025 RMD has to be handled by year‑end (first-timers have the April 1 exception, but that creates two RMDs in one year). One tax-efficient path if you’re charitably inclined: a Qualified Charitable Distribution from an IRA starting at age 70½. You can send up to $100,000 per year directly to a qualified charity, it can satisfy your RMD, and it keeps the amount out of AGI, which can help with Medicare IRMAA brackets and that 3.8% NIIT threshold dance. Quick circle-back: QCDs have to go from the IRA straight to the charity; don’t touch the funds in between.

Last thing on sequencing, because I see this mis-timed in December, trade first in sheltered accounts, harvest taxable losses next, donate appreciated shares after you’ve identified which lots you’d otherwise trim, and only then sell remaining taxable gains. If markets keep swinging into late Q4, you’ll want optionality, not a closed door. And yep, I’ve learned that the hard way.

Keep the upside: hedging without blowing up returns

If you’re reluctant to sell into year-end (I get it; AI leaders are still getting bid into prints), use low-drama hedges that define your downside while letting you keep most of the participation. My take: keep it simple, keep it cheap, and accept you won’t nail the top. A couple of live pricing notes from this week, because cost matters. On our screens, 45-60 day, 5-10% out-of-the-money calls in mega-cap AI names are pricing around 25-35 delta and collecting roughly 1.0-2.2% of spot in premium, while 45-60 day, 10% OTM index puts on a tech-heavy index are running near 1.0-1.6% of notional with vol in the low 20s. That’s enough to matter without torpedoing returns.

• Covered calls (1-2 months out, 5-10% OTM): This trims exposure and generates income into earnings clusters. It’s boring in a good way. As of this week, selling a 60-day 10% OTM call on a large AI bellwether picks up around 1.5% of spot; 5% OTM is closer to 2%+. If the stock spikes, you can be called away, accept that going in. I’d ladder expiries around likely catalyst windows, not all on the same Friday. One more practical bit: keep sizes modest if liquidity is patchy around holidays. I’ve had December weeks where the tail wags the dog.
• Protective collars: Pair a covered call with an OTM put, partially financed by the call premium, to define a range through volatile periods. Example structure we’re seeing price well now: sell a 5-7% OTM call and buy a 10-12% OTM put in the same 45-60 day window. Net cost often compresses to near zero to ~0.5% outlay, depending on skew (we’re seeing put skew ~4-6 vol points over calls in several AI-heavy indices). You cap upside, but you also sleep.
• Index overlays: If name-by-name hedging feels like whack-a-mole, buy puts on a tech-heavy index as a cheaper aggregate hedge. As of this month, a 2-month 10% OTM put on a concentrated tech index costs around 1-1.4% of notional, about 30-40% cheaper than summing single-name hedges with similar deltas, our desk math, not gospel, but directionally right. Correlation tends to jump into macro or guidance shocks, which helps the overlay kick in when you actually need it.
• Trailing stops with air: Avoid getting whipped out by normal noise. For AI momentum names that swing, a stop based on 1.5-2.5x ATR (14-day) often translates to roughly 15-25% below recent highs, wide enough to survive a routine downgrade or Twitter rumor, tight enough to guard against a real crack. If that sounds too wide, that might be the signal about position sizing, not the stop.

Context check: earlier this year, we saw multiple 3-5% single-day reversals around AI earnings, and implied vol into prints is still elevated relative to 2021-2022 norms. Paying ~1% to fence off the tail isn’t crazy when a guidance miss can take 8-12% in a session.

One last thing I remind myself: hedges aren’t free, and they will feel annoying during grind-up weeks. That’s fine. The goal is to keep around 80-90% of the upside while cutting the left tail. If you’re asking “when-to-take-profits-from-2025-ai-rally,” this is the middle path, skims some income, defines risk, keeps you in the seat. And if you do get called away on a spike, great; re-underwrite at the new price or roll the call up a notch. No heroics needed.

Alright, what should you do this week

Alright, what should you do this week? We’ve talked concepts; now it’s button-click time. Keep it boring, keep it repeatable, and get it on your calendar while Q4 is still friendly to planners and harsh on procrastinators (ask me how I know…).

1. Inventory your AI exposure (20 minutes, today).
   • Make a quick table: ticker, current weight (% of portfolio), cost basis, unrealized gain/loss. Old-school is fine, Excel/Sheets or even a napkin snapshot.
   • Flag any single AI position >10% and flag if total AI exposure >35%. Those are your risk beacons.
   • Note which names are “core infra” (chips, cloud, networking) vs. “apps” (software using AI). It helps when you rebalance after earnings shocks.
2. Pick one discipline for exits/defense. Only one.
   • Option A: Scale-out bands. Pre-set trims at +15%, +25%, +40% from today’s price. If a name hits the band, sell 10-20% of the position. No negotiating with yourself 48 hours after a headline.
   • Option B: Collars. Choose a collar for your top 1-2 oversized positions: buy a ~3-month put ~8-10% OTM and sell a call ~10-12% OTM. Yes, that’s a mouthful, translation: a downside seatbelt partly funded by giving up a sliver of upside.
   • Schedule orders/alerts before earnings (late Oct through mid-Nov for most AI bellwethers). I use price alerts and conditional orders so I’m not making decisions mid-conf call.
3. Run a 10-minute business check per holding (this week).
   • Growth trend: last two quarters revenue and guidance direction.
   • Cash flow: positive or still “promise-funded”?
   • Customer concentration: any >15% single customer risk?
   • Backlog/lead times: improving or stretching?

   If two of the four worsen vs. the prior quarter, trim 10-20%. No drama, just a rules-based nudge.

4. Tax map (this weekend).
   • Mark each lot as long-term vs. short-term. As of 2025, long-term gains are taxed at 0%, 15%, or 20% depending on income; high earners may owe the 3.8% NIIT. Keep that in mind before you hit sell.
   • Realize gains from long-term lots first and pair with any harvested losses. Mind the 30-day wash-sale clock if you’re swapping in/out of lookalikes.
5. Set two calendar reminders.
   • Late November (I use Nov 25): revisit caps, update the 10-minute checks, verify collars or scale-out bands haven’t drifted.
   • Mid-December (Dec 16 is fine): finalize tax moves and rebalance anything outside your rulebook bands before liquidity thins into year-end.

Quick reality check: earlier this year we saw multiple 3-5% single-day reversals around AI earnings, and guidance misses still took 8-12% out of names in one session. Implied vol into prints remains elevated vs. 2021-2022, which is why paying roughly ~1% in premium to fence the tail is reasonable. I hate paying it too; I like sleeping more.

One human note: I printed my positions last Sunday and noticed two lots I’d sworn were long-term were actually 11.5 months old, close, but no cigar. Caught it before selling, saved myself a higher bracket. Also, I wrote “delta” on my options sheet, then scratched it out, call it “sensitivity to price moves.” Same point, fewer eye-rolls.

Last thing: if a holding breaches your 10% cap or you’re past 35% AI overall, don’t wait for a prettier tape. Trim in thirds over the next 1-2 weeks. You’ll rarely pick the perfect tick; you will reduce the regret range. That’s the job.

Frequently Asked Questions

Q: Should I worry about Q4 tax selling hitting my AI winners this year?

A: A bit, yes. Q4 often brings tax-gain harvesting and window-dressing, so expect “rip, fade, repeat.” Practical fix: stagger trims into strength, use limit orders above recent highs, and avoid dumping into thin holiday liquidity. If you need losses, harvest elsewhere, don’t torch your best names just to be busy.

Q: How do I set profit-taking rules for concentrated AI names without killing my upside?

A: Use a tiered plan that respects momentum but protects your P&amp;L. Example: trim 10-20% of a position after a 25-35% move from your last add, then trail a stop under the 50-day moving average or 3-5 ATRs, whichever is wider. For mega-caps leading this year, I prefer closing-basis stops to avoid whipsaws. Tie trims to catalysts: scale before earnings if the stock’s extended (say 15-20% above its 50-DMA) and add back only if guidance and revisions improve. Cap single-name exposure (e.g., 7-10% of portfolio; 12% if you’ve got high conviction and hedges). Rebalance monthly, not daily, strong trends punish fidgeting. And write the rules down. If you can’t explain your sell trigger in one sentence, you don’t have one.

Q: Is it better to trim my top AI positions or hedge them into year-end?

A: Both can work, pick based on tax lot quality and concentration. If you’re sitting on large low-basis gains from earlier this year, hedging can defer taxes: buy put spreads 5-10% out-of-the-money dated through January, or collar with covered calls to reduce cost. If a position is oversized (say &gt;10% of the portfolio) or 20%+ above its 50-DMA, I’d trim some into strength first, then hedge the core. For diversified exposure (ETF like a broad tech/AI basket), index puts against the Nasdaq-100 often hedge cheaper than single-name options. Watch liquidity around holidays; widen limit orders. One more thing: don’t “hedge” with correlated names, that’s just doubling down in disguise. Real hedge = negative or low correlation, defined risk, known carry cost. I’ve learned that one the hard way, more than once.

Q: What’s the difference between a normal pullback in a strong trend and a topping pattern, and how should I act in each case?

A: Normal pullback: happens on lighter volume, holds key trend levels, and leadership breadth doesn’t crack. Think: -8% to -15% drawdown, pullback to the 50-day moving average, then higher lows. Earnings revisions stay flat-to-up, and the leaders still lead. Action: buy or add in thirds near support, use a closing stop a bit below the 50-DMA (or the prior swing low), and plan to trim back into prior highs. Topping pattern: distribution days stack up (heavy-volume down days), failed breakouts, and leaders start lagging the index. The 50-DMA flattens/rolls, then the 100- and 200-DMA get tested with weak bounces. Breadth narrows hard. Guidance cuts or negative revisions show up. Action: sell into strength on the first bounce, reduce 30-50% of extended positions, and tighten stops to just under the 50- or 100-DMA. If the 200-DMA breaks on volume, treat rallies as exit liquidity. Examples: in 2023, seven mega-caps drove most S&amp;P gains, pullbacks that held the 50-DMA were buyable. In the 2000-2002 unwind, failed retests and rolling MAs signaled a top. In Q4 2025, I’m watching: percent of AI leaders above their 50-DMA, earnings revisions breadth, and whether weakness shows up on volume. If two of those three flip negative, I shift from buy-the-dip to sell-the-rip. It’s not perfect, but it keeps me on the right side of the tape.

@article{when-to-take-profits-from-the-2025-ai-rally,
    title   = {When to Take Profits From the 2025 AI Rally},
    author  = {Beeri Sparks},
    year    = {2025},
    journal = {Bankpointe},
    url     = {https://bankpointe.com/articles/take-profits-2025-ai-rally/}
}