When AI chip stocks fall, I ask two questions before I look at a single headline. Where is the price sitting on the chart? And the last time this stock fell hard, how fast did it get half of it back?
The second question is the one I lean on most. When a stock wins back half its drop in a few weeks, my read is that buyers were waiting underneath it. When it takes months and still can't, that tells me something too.
Memory stocks are having a rough week. On October 13, Micron fell about 4% after China's CXMT claimed a breakthrough in memory chip production, which raised worries about more competition. Citi called the selloff overdone the next day: it doesn't expect CXMT's new design in mass production before 2028 and puts CXMT 4 generations behind the big memory makers. As of about 1 pm ET today (October 15), the Nasdaq 100 ETF ($QQQ) was down 0.7% on the day and about 5% below its 52-week high, while the S&P 500 ETF ($SPY) was flat (IBKR quotes). Micron traded as low as $858 this morning, almost 32% under its June high. So this is a good week to show how I read these charts.
What happened this summer
Between late June and July 29, 3 of the most heavily traded AI hardware stocks fell hard (IBKR daily prices, intraday highs and lows):
Micron ($MU): $1,255 on June 25 to $737.88 on July 29. Down 41%.
Sandisk ($SNDK): $2,354 on June 22 to $998.19 on July 29. Down 58%.
Marvell ($MRVL): $329.88 on June 18 to $162.90 on July 29. Down 51%.
Then look at the way back. Micron and Sandisk both closed above the halfway point of their drop on August 17. That's 13 trading days after the low. Marvell did it on August 20, 16 trading days after.
A 41% to 58% drop is brutal. Getting half of it back in about 3 weeks tells me, as my reading of the price action, that a lot of money was waiting for a lower price. I want to know which names have shown that before the next drop. I weigh it when I decide how much room to give the next pullback, knowing it can't tell me how far that pullback goes.
Broadcom ($AVGO) is the contrast. It fell 32%, from $495 on June 3 to $335.81 on September 16. That took 72 trading days. A slow leak, with one bounce to $432.73 on August 10 that faded. And through yesterday's close it hadn't gotten back even 38% of the drop. Its best bounce since the September low topped out at $380.84 on October 6, and this morning it traded at $334.61, below the September low.
Broadcom has company news of its own in the mix. Anthropic's IPO filing shows Broadcom agreed to lend it up to $42 billion, through a convertible note, to help fund a lease of Google's TPU computing capacity (Reuters, October 1). A chart can't tell you which piece of news it's reacting to. But same sector, same summer, very different behavior: that's the clue I'm reading.
Lines 1 to 4: the 20, 50, 100 and 200-day EMAs
A moving average is the average closing price over a set number of days, redrawn every day. On a chart it shows up as a smooth line running through the price.
I use exponential moving averages (EMAs). An EMA gives more weight to the most recent days, so it turns faster than a plain average when the price changes direction. I run 4 of them:
20-day: about a month. The short-term trend. Strong stocks tend to ride it.
50-day: about 2.5 months. The medium-term trend.
100-day: about 5 months. Where I start paying close attention in a real pullback.
200-day: about 10 months. The long-term trend.
In a drawdown, I'm asking which of those lines the stock has lost and which ones it's still holding:
Above the 20 and 50. Strong uptrend. Dips are usually shallow.
Lost the 20 and 50, holding the 100 or 200. A pullback inside a long-term uptrend. This is the zone I watch most closely.
Below all 4. The long-term trend is broken or being tested. I wait for repair before I do anything.
Here's where the names I watch closed yesterday, October 14 (my calculation on IBKR daily closes):
Stock | Close | 20 EMA | 50 EMA | 100 EMA | 200 EMA |
|---|---|---|---|---|---|
Micron ($MU) | $914.13 | $1,016.17 | $989.45 | $914.42 | $752.38 |
Sandisk ($SNDK) | $1,403.34 | $1,620.82 | $1,611.70 | $1,516.18 | $1,243.98 |
NVIDIA ($NVDA) | $221.30 | $227.78 | $222.60 | $215.75 | $204.44 |
AMD ($AMD) | $570.82 | $595.25 | $555.60 | $505.89 | $424.42 |
Marvell ($MRVL) | $245.28 | $260.73 | $245.60 | $225.79 | $190.38 |
Broadcom ($AVGO) | $346.25 | $358.22 | $365.52 | $371.49 | $363.45 |
How I read that table:
Micron closed 29 cents under its 100 EMA. This morning it dipped to $858, well through it, and was back around $909 by 1 pm ET. That 100 EMA is the line I'm watching it fight over.
Sandisk has lost the 20, 50 and 100. It's still about 13% above its 200 EMA.
NVIDIA and Marvell are sitting right on their 50 EMAs, with the 100 below them as the next line.
AMD lost its 20 EMA this week but closed yesterday above its 50.
Broadcom is below all 4, and the lines themselves are tangled together. Its 50 EMA is $2.07 above its 200 EMA and has slid lower for most of the past 2 weeks. When the 50 crosses under the 200, traders call it a death cross. The name is dramatic, and on its own it's a lagging signal. But it's one more thing that has to repair before I'd read Broadcom's chart the way I read Micron's.
Line 5: Fibonacci retracements
This one sounds fancier than it is.
Take a stock's last big move up, from a clear low to a clear high. Fibonacci retracement levels are just percentages of that move, measured back down from the high. The ones that matter to me are 50%, 61.8% and 78.6%.
Why those odd numbers? 61.8% and 78.6% come from the Fibonacci math sequence. 50% is just the halfway point, which traders watch anyway. My reading is that these levels work as often as they do because so many traders and algorithms watch the same ones and put orders there. That's also why I only use them on stocks that trade billions of dollars a day. That more traders watching makes a level hold more often is a bet I can't prove.
How I use them:
50%: first area of interest.
61.8% to 78.6%: the zone where I actually start building a position, in pieces.
Below 78.6%: the move up is mostly undone, and I want to understand why before I do anything.
Here's Micron's swing from the July 29 low ($737.88) to the September 25 high ($1,108.72):
50%: $923.30
61.8%: $879.54
78.6%: $817.24
Micron closed at $914.13 yesterday, just under the 50% line. This morning it dipped to $858, inside the 61.8% to 78.6% zone, and was back around $909 by 1 pm ET.
Sandisk's swing ran from $998.19 (July 29) to $1,909.48 (September 22). Its 61.8% level is $1,346.30, and its 78.6% level is $1,193.21. This morning it touched $1,276, and around 1 pm ET it was trading near $1,346, right on the 61.8% line.
Sandisk is also the cleanest example of an EMA doing its job. Its July 29 low of $998.19 landed within about a dollar of its 200 EMA ($999.44 that day, my calculation). From there it ran 91% in less than 2 months. Its 200 EMA is now about $1,244, the next line below today's price.
The spot I pay the most attention to is where a Fibonacci level and an EMA land on the same price. Both come from the same price history, so an overlap is a coincidence worth noting, nothing more. When 2 different ways of measuring land on the same number, I write that number down. Three of those showed up this week (EMAs as of the October 14 close):
Micron: 50% level $923.30, 100 EMA $914.42.
NVIDIA: 50% level $216.62 (June 29 low to October 6 high), 100 EMA $215.75.
Marvell: 78.6% level $222.16 (September 1 low to October 6 high), 100 EMA $225.79. Today's low was $227.31.
How to draw these in TradingView
I use TradingView for all of this. The free plan allows 2 indicators per chart (TradingView pricing page, October 15), and the setup below uses exactly 2.
Add the EMAs
Open a chart and set the timeframe to 1D (daily) in the top bar.
Click Indicators in the top toolbar and search "EMA 20/50/100/200."
Pick the community script by drsweets. One click puts all 4 lines on the chart, each in its own color.
Click the star next to it so it sits in your Favorites from then on.
On a paid plan, you can use TradingView's built-in tool instead: search "EMA," pick Moving Average Exponential, and add it 4 times with the Length set to 20, 50, 100 and 200. That uses 4 indicator slots, which the free plan doesn't have.
Add the Fibonacci retracement
I let TradingView draw it for me.
Click Indicators again and search "Auto fib retracement." It's a built-in TradingView tool. Star it too.
It finds the last big swing on its own and draws the levels across it. The 0.5, 0.618 and 0.786 lines are your 50%, 61.8% and 78.6% levels.
Check that it grabbed the swing you meant. It goes for the most recent one, which is sometimes a small wiggle. If it did, open its settings and raise the Depth number until it lands on the bigger move.
Check that the 0 and 1 lines sit on the extremes: the tip of the lowest wick and the tip of the highest wick. Those are the prices that actually traded, and every level is measured from them. If the auto version lands a little off, draw it by hand.
To draw it by hand, pick Fib Retracement from the Fibonacci group in the left toolbar. Turn on Magnet in the same toolbar so your clicks snap to the wick tips. Click the lowest wick first, then the highest. If the 0.786 line ends up near the top, open the drawing's settings and tick Reverse.
One habit that saves me: I check the swing it picked, and once I'm happy with it, I leave it alone. If you keep adjusting the swing until the lines match where the price is, you're drawing what you want to see.
My drawdown checklist
When an AI infrastructure name I follow falls, this is the order I go in.
Is the business still fine? Revenue growth and guidance first. If the business broke, none of the chart work matters. For Micron, the latest read is from September 30: quarterly revenue of $54.23 billion, up from $11.32 billion a year earlier, and next-quarter guidance of $61.5 billion, plus or minus $1.5 billion (Micron fiscal Q4 2026 results). The open questions are the CXMT news and labor: Reuters reported on October 15 that a union representing Micron workers in Taiwan has authorization to strike over bonuses, with no strike date set. (My September 10 issue walked through how a stock can fall 40% while sales grow.)
Which EMAs has it lost, and which is it holding? The 20 and 50 going is normal in a pullback. The 100 and 200 are the ones I care about.
Where are the Fibonacci levels on the last big swing? And does any of them line up with an EMA?
How fast did it recover last time? I look at the last real drawdown and count the trading days until it closed back above the halfway point. Weeks is what I want to see. Months tells me buyers aren't in a hurry.
Is it liquid? Over the last 50 sessions, Micron and NVIDIA each averaged about $26 billion of shares traded a day (my math on consolidated daily volume and closing prices, through October 14). I keep this to names that trade billions a day. It's where I trust the levels most.
Pieces, never all at once. I split an entry into 2 or 3 parts: a small one at the 50% level, more in the 61.8% to 78.6% zone, and the rest only if it holds the 200 EMA. If a stock never gets that low, I own less of it than I wanted, and I'm fine with that.
A Fibonacci level can fail, and a moving average can get sliced through on one bad headline. I use the lines to write a plan while I'm calm, so a red day doesn't get to make the decision for me.
Stay disciplined,
Koh
Make Your Own Alpha is live. Today's issue covered how I time an entry. The book covers how I decide which companies earn a place in the portfolio in the first place, with the frameworks I use before a chart ever comes into it. Get your copy here.
Positions as of publication: Micron ($MU), Sandisk ($SNDK), AMD ($AMD), Marvell ($MRVL), NVIDIA ($NVDA).
Disclaimer: Nothing in this email, the Reset and Invest newsletter, the Make Your Own Alpha book, or any course or digital product from Starshine Media LLC constitutes investment advice or a recommendation to buy or sell any security. Numbers and observations are as of publication. I may hold positions in companies discussed. Always do your own research and consult a licensed financial advisor before making investment decisions.