AlphaHedge

11.13 Pre-Market Overview Futures are mostly higher this morning ahead of the PPI report, which comes out at 8:30. - $NVIDIA is slightly higher, rebounding from yesterday's lows. It traded as high as 170. - Some other semiconductor names, notably $Broadcom and $AMD, are seeing some profit taking after a huge week. - Bitcoin is down almost 4%, trading at 69,000. - The Dow Jones is up almost 80 points, the NASDAQ is flattish, $SPY is up 2, and the Qs are flattish. Last Night's Earnings - $Applied Materials ( $AMAT): A beat and a slight raise, but really in line. The stock is down slightly this morning. - $Cisco: Same story. Margins might be a slight issue there, and the stock is down slightly after last night's report. - $Cerebras ( $CBRS): Probably the star of the evening. They reported a very strong quarter and gave great guidance, and it really seems like this company has put in the bottom. We've gotten through the lockup, and this is a stock I'll be focused on during any further pullbacks. It's up 11% this morning, and in after-hours trading last night it got as high as $117. Everyone is looking forward to next week, when we have earnings from $HPE, $Dell and $Deere, and, more importantly, the Anthropic IPO. Analyst Actions - $NVIDIA: Barclays lowered its price target to $210 from $275. - $Cerebras: - Mizuho lowered its price target to $240. They still love the stock. - UBS lowered its target to 155. They have a Buy. - Morgan Stanley reiterates its Overweight rating and 178 price target. - $Nebius ( $NBIS): BofA lowered its price target to 265 from 310, keeping a Buy. - $Applied Materials ( $AMAT), some new targets: - Stifel at 600. - Another firm lowered its target to 520, with a Buy. - Bernstein lowered its target to 600, Outperform. - Mizuho at 510. - Jefferies lowered its target to 580, with a Buy. - $Cisco: Wells Fargo lowered its price target to 130. Citi lowered its target to 120. - $Twist Bioscience: Guggenheim lowered its price target, and the stock remains Buy rated. Canaccord also lowered its target. - $Robinhood: Goldman lowered its price target to 115 from 128. They still have a Buy. Upgrade of the day: $Palo Alto, upgraded to Overweight from Sector Weight at KeyBanc, with a price target of 380. The stock is up quite a bit this morning. Downgrade of the day (though not from a big firm): $Broadcom, downgraded to Neutral from Buy at Rothschild, with the price target cut to 398 from 595. The PPI Matters This morning's PPI does matter. We already had a hot CPI. If this is more hot or inflationary news, you might see a pullback in the markets. Key Levels - $QQQ: If you're a bull, you want to see the 200-day hold, and that's 682. The 21-day is down to 696.60. If we can get through it, we'd launch to the 700–701 area. - $SPY: It's up this morning, and the next upside resistance is 764. That's your 50-day. Today's Focus I'll definitely be focused on $Cerebras, but I'll also pay attention to $AMAT and $Cisco for day-after trades. Those are the three major names. $Cerebras ( $CBRS) The 21-day on $Cerebras is 106, and that would be a nice area to hold. I'd love to build a position there. We'll see how it trades today. Hopefully it puts in a low that we can trade against going forward, because I think $CBRS is going to 120 and eventually 140. $NVIDIA $NVIDIA is rallying, and I think it's going to stop at about 170, where it traded pre-market. If it ever got near 178–180 again, that neckline is a big spot for long-termers to de-risk and at least take some off. I don't think it gets there for a while, and it's probably range bound now. If you saw 150 again, it's probably a buy, like we did yesterday. $AMD $AMD is opening down for the first time this week, I think, in a while. We might see some follow-through to the downside. If we do, and there's some real profit taking, $AMD has room down to the 50-day at 555 and the 100-day at 527. Again, the short side is cute; you really want to be long. If it does correct, this stock is replacing $NVIDIA for a lot of funds and tech money. Still, you want to see a pullback first. At one point this week, I think the stock was up nearly 30%. So $AMD is a short-side focus for today, maybe with puts. If there's follow-through to the downside, on a longer-term basis we want to buy it at 550 or below. I'd really like to see the 100-day and 21-day, in the 521–527 area. $AppLovin ( $APP) $AppLovin is letting the 8-day catch up, and it's finally almost there right now. That's at 377, which means I'm a buyer of $AppLovin in the low 380s. $Amazon $Amazon has a bullish flag. We're flagging and letting the 8-day catch up, and it finally has. The 8-day is 266. You'll notice it keeps going to 273–275 and then pulls back a few dollars. One of these days, I do believe we get through 275, go to 280 and eventually 287. I'm a buyer on any pullback into the 266–267 area. If it ever breaks out above 275, that would be a lotto call idea. $Microsoft $microsoft has a major outage, not just in the United States but worldwide, and the stock is lower because of it. If it stays below the 50-day at 518, I see room down to 504, which is probably where I'd buy it. If it gets back above 518, the 50-day, it's probably a bit healthier again. That's it for this morning.

Jensen Huang wasn’t forced out. He made the decision to step down himself. After all, the pressure was enormous: three House committees have summoned him for a closed-door briefing on November 18, a shareholder class-action lawsuit has already been filed, Republican senators are publicly calling for his resignation, and an independent investigation is underway. Nvidia’s $NVDA real risk was never Jensen’s personal life. It’s the fact that this $4 trillion company has relied on a single brain for 33 years, and that brain has refused to build a backup for itself. Nvidia’s stock price is effectively his hostage. The board has looked around and realized there’s no one else capable of commanding the same confidence. That’s his greatest leverage. The November 17 earnings report will be the first real test. If the new CEO can’t deliver a performance that proves Nvidia can still beat expectations without Jensen, bringing him back could become a serious option for the board. History offers plenty of precedents: Steve Jobs returned to Apple in 1997, Howard Schultz returned to Starbucks in 2008, and Bob Iger returned to Disney in 2022.

Is the value factor dead? I don't think so. My view is that "buy cheap" was never the problem. The problem is how you measure "cheap." Measure it properly and benchmark it correctly, and value still deserves a spot in a portfolio. First, what quant value actually is. It's related to Buffett-style investing: find good companies, wait for a low price, hold for years. But in practice it's a completely different thing. It doesn't study any one company. It bets on a statistical pattern: a group of low-valuation stocks may beat the market average over the long run. The most common mistake is treating it as stock picking. A cheap company doesn't have to go up. Single stocks carry a lot of their own risk. Operating trouble, an industry in decline or bad management calls can all make a cheap stock cheaper. Value is a property of the group. A rough example: calculate valuations for all 5,000+ listed companies, rank them from cheapest to most expensive, and buy the cheapest 30%. That basket may beat the market average. What's true of one stock and what's true of the group are different things. A single cheap stock's result is mostly luck. Hold a few hundred and most of the stock-specific risk cancels out. What's left is the return that comes from cheapness itself. So value has to be held diversified and run by rules. It works on probability, not on a view about any one company. Next: long-short vs long-only. If you pick the wrong benchmark, you'll reach the wrong conclusion. Long-short: rank the whole market by valuation, say price-to-book. Go long the cheapest 30% and short the most expensive 30%. That's the textbook definition. The classic academic HML factor (high book-to-market minus low book-to-market) uses this 30/40/30 split. It also sorts by size to cut down on small-cap noise. Put that long-short portfolio next to the S&P 500 and it trails badly. That's where the "value is dead" takes come from. But that's the wrong comparison. The S&P 500 is a long-only strategy that earns mostly from the market going up. In a long-short portfolio, the long side and the short side roughly cancel out market moves, so market risk is mostly removed. It doesn't earn the market's rise. It earns the gap between cheap and expensive stocks. So long-short value should be measured against the risk-free rate, meaning US Treasury yields. On that basis, long-short value has done somewhat better than the risk-free rate. After 2010, people said value had been "dead for over a decade." Measured long-short against the risk-free rate, it still has an edge. That edge just keeps getting smaller. Long-only: buy only the cheap stocks, no shorts. This takes on market risk, so comparing it with the S&P 500 is fair. But don't compare returns alone. Look at return and risk together, which means the Sharpe ratio. Sharpe = (portfolio return − risk-free rate) ÷ the volatility (standard deviation) of portfolio returns. It tells you how much excess return you get for each unit of volatility. If two portfolios return about the same, the less volatile one has the higher Sharpe. It's also easier to hold and easier to replicate. The S&P 500 is a very good portfolio and a very good strategy in its own right. A long-only value portfolio is only well built if its Sharpe beats the S&P 500's. If its Sharpe is lower, that way of building it isn't good. The original value factor used price-to-book. More precisely, it used the inverse: book equity divided by market cap, called "book-to-market." A higher number means cheaper and more worth buying. A lower number means more expensive and less worth buying. This comes from Fama and French's 1992 work on the cross-section of stock returns. They found that two simple measures, size and book-to-market, explain a large share of the differences in stock returns. The three-factor model was built on that. Why use the inverse? There's a practical reason. When book equity is negative or close to zero, P/B becomes huge or meaningless. Flip it and the numbers are continuous, so the ranking is more stable. P/E works the same way. A loss-making company's P/E means nothing, but earnings yield (EPS ÷ price) is much easier to work with. Markets have changed a lot since then, and many people think the original method no longer works. So several top firms have reworked it. Here are three main approaches. AQR: a combined score from several valuation measures. There's a lot of detail, but one point matters most: P/B alone isn't enough. Use P/E, P/B and other measures together. Calculate them for every company in the market, average them, see which companies are cheaper or pricier overall, and build the factor from that. The logic is simple. Every valuation measure has a blind spot. P/B is unfair to companies with few physical assets. P/E gets distorted when profits swing. P/S can't see margins. Using several together dilutes the noise from any one of them. In practice the measures sit on very different scales, so you usually convert them to ranks or z-scores before averaging. Avantis: add a quality screen and strip out goodwill. Avantis spun out of Dimensional. Dimensional is closely tied to the people who came up with the value factor: Fama and French sit on its board. Dimensional's value factor is very pure. It has always used P/B and nothing else. Avantis argues you need a quality check on top of the traditional factor. A cheap stock isn't always truly cheap. It may be a value trap, where the price is low because the company really is that bad. So they screen for quality using fundamentals. In their view, the stocks worth buying are decent companies that look cheap because the market sold them off unfairly. Quality is usually measured by profitability (gross margin, ROE), how stable earnings are, and how healthy the balance sheet is. The idea is to take the cheap stocks and remove the ones that are cheap for good reason. Avantis also says goodwill should be removed when building the value factor. The US has seen a huge amount of M&A over the past few decades, and buyers usually pay a premium. Say company B is worth 10 billion and company A pays 12 billion for it. The extra 2 billion is the premium, and it goes on the books as goodwill. The bigger the premium, the bigger book equity looks. The problem is that goodwill can't be turned into cash. Receivables get collected. Inventory gets sold. Goodwill turns into nothing. After so many deals, US balance sheets carry a lot of it. So Avantis subtracts goodwill from shareholders' equity before calculating P/B. The effect is easy to see. Companies that make lots of acquisitions have book equity inflated by goodwill, so their P/B looks low and they seem "cheap." Remove the goodwill and they may not be so cheap after all. And when goodwill is written down, the hit goes straight to equity and earnings. That's a common source of risk for acquisition-heavy companies. Research Affiliates: add intangible assets back in. Research Affiliates is another factor-investing firm, but its fix goes the opposite way from Avantis. It looks at assets the books leave out: intangibles. Under US accounting rules, a lot of R&D spending is expensed right away and never becomes an asset. As companies spend more and more on R&D, writing all of it off makes less sense. The technology, patents and brands that R&D creates are really assets that pay off for years. So Research Affiliates adds the expensed R&D back to assets before deciding whether a company is expensive. The issue behind this is that traditional P/B naturally favors asset-heavy industries. Banks, steel and energy have lots of assets on the books. Much of the value in tech and pharma isn't on the balance sheet at all. Without an adjustment, a "value portfolio" easily turns into an "old-economy portfolio" and misses many asset-light companies that really are undervalued. These three are just examples. What they have in common: the flaw in the original factor isn't the idea of buying cheap. It's that "cheap" isn't measured accurately enough. The build I prefer: several valuation measures, a light quality screen, and EV/EBITDA. It's mostly a mix of AQR and Avantis, with one more layer for debt. Layer 1, from AQR: a combined valuation score using P/E, P/B and P/S. Layer 2, from Avantis: a quality screen, but a very light one. Here's why. If the quality bar is too high, what's left looks more and more like a quality portfolio, and you lose the source of return that makes value work. A light screen just keeps the most obvious junk out. Layer 3, my own idea: screen out companies with debt problems. Some companies look very cheap but carry huge debt. I don't want them. They're especially likely to blow up. The tool for this is EV/EBITDA, added alongside the combined valuation score. EV is enterprise value: roughly market cap plus interest-bearing debt, minus cash. It's the price of buying the whole company, both its stock and its debt. EBITDA is earnings before interest, taxes, depreciation and amortization. It shows how much cash the core business generates, and it isn't affected by how the company is financed or how it depreciates assets. P/E and P/B only look at the shareholder side. If a company borrows heavily and its market cap gets crushed, those measures say it's "cheap." EV counts the debt, so a heavily indebted company shows a much higher EV/EBITDA and stops looking cheap. Ranking on it pushes the companies that only look cheap because of borrowing down the list. Built this way, the results are much better. I compared this method with the other firms' methods on long-short Sharpe. It isn't the best, but it's in a decent range. Long-only is where it really stands out. Combining P/E, P/B and P/S, adding EV/EBITDA, and applying a light quality screen gives a higher Sharpe than the other traditional ways of building a value factor. It isn't a quick win and it isn't exciting. But it's a value factor build that is actually in use. Where people go wrong with value: • Testing the factor on single stocks. One cheap stock that doesn't go up proves nothing. Judge the factor by how a whole portfolio does over a long period. • Using the wrong benchmark. Compare long-short against the risk-free rate and long-only against the index. Compare Sharpe ratios, not just returns. • Underestimating how long the drawdowns last. Value can trail for many years in a row; the stretch after 2010 is the example. If you can't sit through it and quit halfway, you never get the long-run edge. • Ignoring sector concentration. With P/B alone, the portfolio piles into asset-heavy sectors like financials and energy. Then you're betting on sectors, not on cheapness. • Ignoring trading costs. Value portfolios need regular rebalancing. Rebalance too often and turnover and market-impact costs eat an excess return that was small to begin with. This isn't buy or sell advice. It's the framework I use to check any value product or a strategy I build myself. Here's what I'm watching: • How valuation is measured. Is it P/B alone, or a combined score across several measures? Does it deal with goodwill and R&D, the items that distort book value? • How debt is handled. Is there a debt-aware measure like EV/EBITDA that keeps out companies that only look cheap because they borrowed? • How strong the quality screen is. Is there something to guard against value traps? Is the screen so heavy that the value factor has turned into a quality factor? • How it's evaluated. Long-short against the risk-free rate, long-only against an index like the S&P 500, with Sharpe as the final test. If a value strategy passes all four, I'm willing to stick with it through the lean years. If it fails them, the problem is almost always in how it defines "cheap," not in the idea of buying cheap.

Buying the Dip in U.S. Stocks? Watch the 200-Day Moving Average First Nasdaq has bounced hard over the last couple of days, and everyone is asking again whether it's time to buy the dip. My view: only buy dips in a bull market, and stay away while the index is sitting right on its 200-day moving average. Guessing whether this particular dip is the right one is the wrong question. The useful question is what a good dip-buy in US stocks actually looks like, and whether you can write it down as rules you can follow and backtest. You can. The whole framework is built around one line: the 200-day. Where things stand this round: • Nasdaq broke below its 200-day and is now back near it • Some indexes are still below their 200-day • The S&P 500 never broke its 200-day at all Those differences matter. As you'll see, they decide how the strategy plays out on each index. Why the 200-day? It's the average of the last 200 closes. A US trading year has roughly 250 days, so the 200-day covers most of a year of price action. It tracks the medium-to-long-term trend, and short-term swings barely move it. It's also the market's consensus bull/bear line. Fund managers, the financial press and technical traders all watch it. An index crossing above or below it changes how some of that money behaves, and that shared belief makes the line matter even more. What really makes it matter is the long-run data. Above and below the 200-day, Nasdaq and the S&P behave like two different markets. Returns, 1971 to 2026: • Nasdaq above the 200-day: 15.9% annualized. Below: -1.7% • S&P above: 9.8% annualized. Below: 4.6% Volatility, same period: • Nasdaq above: 15.8% annualized. Below: 28.4%, nearly double • S&P above: 13.2%. Below: 24.9% In one sentence: below the 200-day, returns are lower and volatility is roughly twice as high. Put those together and the Sharpe ratio gap is huge. Sharpe is the return above the risk-free rate divided by volatility. A smaller numerator over a bigger denominator gives a much worse ratio. For the same unit of risk, you get paid far less below the line. Here's the paradox of the 200-day: near it, playing offense is right, and so is playing defense. The case for offense: since 2008, the 200-day has been very strong support for US stocks. For both Nasdaq and the S&P, most moves down to the line or below it ended in a deep V back up. The case for defense: go back to the 1970s, and exiting on a break of the 200-day was the right call for dodging the real disasters. The 2000 dot-com crash, the 2007–2008 subprime crisis, even the 2022 decline. So the same price action means completely different things depending on your risk appetite and strategy. If avoiding big drawdowns comes first, you start defending at the 200-day. A short-term trader gets excited there, because the market is about to enter a high-volatility zone, and over the last 15 to 20 years that zone has usually been strong support. The key point: the 200-day doesn't give you the answer. It tells you the market has reached the point where opinion is most divided. From there I set three principles first, then turn them into rules. Principle 1: Avoid the area right around the 200-day. Since both offense and defense make sense there, the worst outcome is the index chopping back and forth across the line. Above today, below tomorrow, above again the day after. Anyone trading the crossover gets whipsawed again and again. Trading is hard, and both the win rate and the payoff are poor. The opportunities come in two states: • The index is running steadily above the 200-day. Fundamentals haven't changed, but sentiment or some outside shock sets off a sharp drop. Drops driven by emotion tend to recover fast. • The index is far below the 200-day. Over the last two decades, US stocks stretched too far below the line have tended to snap back in a deep V. In short: trade it when it's steadily above, trade it when it's sold off far below, and never trade it while it's hugging the line. Principle 2: Only time the market in a bull market. In bull markets, technical indicators work fairly well and sharp drops recover quickly. Once a bear market is confirmed, timing gets brutal. Bear markets bring unexpected big drops that blunt or break almost every indicator. Blunting means the indicator flashed oversold long ago, but price keeps falling. Oversold turns into more oversold, and anyone buying on the signal gets trapped again and again. In a bear market, cheap tends to get cheaper. How do I tell a bull market from a bear market? Simple. If the 200-day is still rising, it's a bull market. If the 200-day has started turning down, it's a bear market. Principle 3: Set a take-profit and a stop-loss. A short-term strategy without exit rules isn't a strategy. Entry rule 1: the emotional flush in a bull market. This is for an index running steadily above its 200-day that suddenly drops on sentiment. Three conditions, all of which must be met: • The close is above the 200-day. This confirms we're above the line. • The 200-day is higher than it was 20 trading days ago. You can also say the 200-day slope is positive; it means the same thing. 20 days is about a month, so this confirms the line kept rising over the past month. This is the no-timing-in-bear-markets filter. • The 5-day decline is more than twice the average 5-day move. The second condition screens out bear markets. September 2000 through 2003, 2007–2009 and 2022 were obvious bear phases. Comparing the 200-day to its level 20 days earlier, or checking its slope, picks them out, and the strategy stays out during those periods. The third condition spots emotional selling. Sentiment-driven drops have one thing in common: volatility suddenly jumps. So I just check whether the short-term drop is clearly outside the normal range. Why these parameters: • 5 days because it's one week. Don't over-tune the parameters, or you'll overfit. • 2x because two standard deviations is the usual line between likely and unlikely events. Bollinger Bands default to 2 standard deviations too. How to calculate it: from daily closes, compute the rolling 5-day % change. Over a long historical window, take the standard deviation of those 5-day changes. That's one common measure of the average 5-day move. If the latest 5-day drop is more than twice that, the condition is met. Assuming a roughly normal distribution, moves beyond 2 standard deviations happen only about 5% of the time, and less often on the downside alone. So a trigger means a truly unusual sell-off. Bollinger Bands work on the same idea. The middle band is the 20-day average, the outer bands sit 2 standard deviations of 20-day volatility above and below it, and touching the lower band is often read as short-term oversold. The difference is that in a one-way sell-off, price can ride the lower band all the way down. That's exactly why you have to rule out bear markets first, using the direction of the 200-day. Entry rule 2: a deep drop below the line while the 200-day is still rising. This covers the other state: the index has fallen well below its 200-day, but the 200-day itself is still rising. The basis is that the gap from the line tends to revert to the mean. Gap = (close − 200-day) / 200-day, as a percentage. For both Nasdaq and the S&P 500, this gap clearly reverts. It doesn't keep moving in one direction; at extremes it gets pulled back. So when the index is below the line and the line is still rising, once the drop reaches a threshold you can bet on a deep-V bounce. Two conditions: • The index is 10% or more below its 200-day • The 200-day is higher than it was 20 trading days ago, or its slope is positive. Same as rule 1: no timing in bear markets. Why 10%? Again, no tuning. Pick a sensible round number. To be more aggressive, use 8%. To be more conservative, use 12%, 13% or even 15%. What matters is not tweaking parameters down to the decimal point just to make the backtest look better. Why I care so much about overfitting: parameters found by trial and error on historical data may just happen to fit those past episodes and stop working in the future. The simpler and more intuitive the parameters (a week, a month, 2 standard deviations, 10%), the more likely the strategy is capturing a real market pattern rather than a fluke in the data. The test is easy: move the parameters up and down a little. If the results barely change, the strategy is robust. If a small change breaks it, it's overfit. Exits. Exit rules depend on the person, because they come down to take-profit and stop-loss. Take-profit depends on how much you want from a trade. Stop-loss depends on how much risk you can take: some people can't stand a 10% loss, others are fine with 30%. Here's one example set, just to show the logic. Exit as soon as any one of these is met: • Take-profit: a close back above the 5-day moving average ends the trade. Or stop once the index itself is up 2% or more, or 5% or more. Using options adds more leverage. • Stop-loss: for example 15%. Adjust it to your situation. • Time: after 20 trading days the trade ends, win or lose. The time stop is the one people ignore. Never mix short-term and long-term strategies. Buy-and-hold in US stocks already wins often: buy the S&P or Nasdaq on a random day and you come out ahead about 53% to 54% of the time. Turning a losing short-term trade into a long-term hold mixes two different logics, and that doesn't hold up professionally. It's a short-term trade, so it lasts a month at most. At 20 trading days, you close it. The backtest. With entries and exits defined, you can backtest. The results are strong: a win rate of roughly 70% or more, over 80% on the S&P, and a payoff ratio of 1.5 or more. That's excellent. Win rate and payoff ratio are the two basic measures for judging any trading strategy. Win rate is winning trades divided by all trades. 80% means about eight winners and two losers out of ten. Payoff ratio is the average win divided by the average loss. If you make a little each time you win and lose a lot when you lose, the payoff is low and the trade isn't worth it. A payoff above 1 means your wins more than cover your losses. Together they decide whether a strategy makes money. Expected value per trade is roughly win rate × average win − loss rate × average loss. With a 70% win rate and a 1.5 payoff, counting the average loss as one unit: 0.7 × 1.5 − 0.3 × 1 = 0.75 units. On average, every trade is clearly positive. Buffett often says investing should be about doing things with high certainty. Short-term timing should be the same. A high win rate with a low payoff means one big loss wipes out many small wins. A low win rate with a high payoff means sitting through long losing streaks, which is very hard to stick with. In short-term timing neither works. You want both the win rate and the payoff to be decent. Two preconditions to understand before you use this. First, it depends on how US markets trade. US stocks trade almost 24 hours: the regular session, pre-market, after-hours and an overnight session. Prices move almost continuously, so take-profits and stops fill close to their set levels and overnight gaps rarely blow them up. Without that, the rules work much worse. Second, it only works on the S&P 500 and Nasdaq. On gold, oil or some other emerging-market stock indexes it basically doesn't work. Backtest it yourself if you're curious. The reason is simple. US broad indexes rest on corporate earnings that rise over time, and lots of money buys on pullbacks. That's why bull-market sell-offs recover and big drops below the line snap back. Commodities follow supply-and-demand cycles and don't necessarily rise over the long run. Some emerging-market indexes move sideways for years, so the 200-day keeps changing direction and the rules' basic assumption breaks down. Also: backtests are history, not promises. The bear filter relies on the 200-day turning down, and moving averages lag by nature. Early in a bear market the line may still be rising, and buy signals there are where the strategy is weakest. That's exactly why the stop-loss and time exit exist. Did this round trigger anything? Nasdaq didn't. It broke the 200-day and came back, and the whole move stayed around the line. That's the zone this framework sees as the most dangerous and uncertain. Principle 1 says no trade. The S&P triggered both a buy and a sell. The day it dropped without breaking its 200-day met rule 1's sell-off condition. Then it closed back above its 5-day average, the take-profit hit, and that was one complete trade. So when people say they called the Nasdaq dip right this time, by this logic it was often luck. Buying on gut feel works this time and loses next time. A good short-term timing strategy is a set of measurable rules you can actually follow over the long run. In real trading the mental pressure is huge, and fear and greed change your judgment at the worst moments. Without clear rules you can't resist that over time. Rely on gut feel as little as possible. My checklist before any dip-buy: • Direction: is the 200-day above where it was 20 trading days ago, or is its slope positive? If it has turned down, it's a bear market. No timing. • Position: if the index is chopping back and forth across the 200-day, no trade. If it's steadily above, wait for a 5-day drop of more than twice the average 5-day move. If it's below the line, wait for a gap of about 10% (8% aggressive, 12% to 15% conservative). • Set exits before you enter: take profit on a close back above the 5-day average (or the index up 2% to 5%), set a stop (15% in the example), and hold for 20 trading days at most. • Use it only on the S&P 500 and Nasdaq. Don't apply it to gold, oil or other emerging markets, and don't turn losing short-term trades into long-term holds. What I'm watching: whether Nasdaq moves clearly away from its 200-day instead of chopping around it, whether the 200-day lines keep rising versus 20 trading days earlier, and whether any index falls far enough below its line to approach a 10% gap. Until one of those setups shows up, I stay out.

I still believe Jensen Huang is highly unlikely to step down as CEO, or at least I certainly hope he won’t. Because we cannot let these racists win! There is something deeply suspicious about how the media is handling this situation. Tonight, The Wall Street Journal published a so-called exclusive report titled “Nvidia Prepares to Oust Jensen Huang.” It is pure clickbait. Rather than simply reporting the news, the article appears to be taking a position, shaping public opinion, and attempting to push Huang out of office by creating a fait accompli. Anyone with even a basic understanding of corporate governance knows that a board of directors cannot make such a consequential decision overnight. The article itself states, citing people familiar with the matter, that “the board plans to meet as early as Wednesday to decide Huang’s future.” If the board hasn’t even met to make a decision, why does the headline make it sound as though the outcome has already been determined? That isn’t reporting the facts. That’s manufacturing a narrative. This time, Huang is being attacked from both sides of the media’s political spectrum. On the left, The Economist published what I consider a racist article titled “Jensen Huang Must Resign.” On the right, The Wall Street Journal ran the headline “Nvidia Prepares to Oust Jensen Huang.” Fox News even suggested that, given Huang’s appearance, he should have anticipated that a Chinese national approaching him might be a spy. Meanwhile, politicians from both parties are piling on. Personal attacks, racism, and character assassination. It is absolutely disgusting. Think about it. If Jensen Huang had visited Jeffrey Epstein’s island, there would be right-wing politicians rushing to defend him. If he were Black, there would be left-wing politicians standing up for him. But today, he is being targeted for one reason: he is a successful Chinese American.

Despite October’s U.S. CPI coming in hotter than expected across the board, with markets even pricing in the possibility of a 50-basis-point rate hike, the S&P 500 and QQQ still held firmly above 7,500 and 670 yesterday. This resilience, despite such negative news, suggests that the market remains determined to push higher. Today, AI giant Anthropic is launching a historic IPO and private placement process, targeting a staggering valuation of $1.3–$1.5 trillion, while also expanding its computing capacity leasing agreements with Microsoft and Google by more than $10 billion. Meanwhile, CoreWeave ($CRWV) delivered a massive earnings beat after yesterday’s close and significantly raised its guidance, providing a powerful confidence boost to the AI infrastructure trade. I took advantage of yesterday’s pullback to buy SanDisk ($SNDK). I expect memory stocks, optical communication chips, AI power infrastructure, and AI hardware to stage a strong rebound today. There should still be opportunities to buy on dips.

October CPI came in slightly above expectations across the board. Combined with strong employment data, markets are now pricing in roughly a 90% chance of a Fed rate hike in December, with growing expectations for at least two hikes by mid-2027. Meanwhile, Trump said there is no need for large-scale military action for now, easing US-Iran tensions and reducing geopolitical risk premiums. My trading strategy: Yesterday, I advised against chasing the gap-up rally and suggested waiting for today’s pullback to buy. If $QQQ holds above $670 during the first 30 minutes of trading, I believe it’s a good opportunity to start buying high-conviction names. For AI stocks, I’m focusing on companies with strong fundamentals and direct exposure to AI capital spending. Capital will likely continue flowing toward industry leaders with real competitive advantages, particularly advanced chip manufacturing, semiconductor equipment, and core AI infrastructure players like $TSM. The key: Buy quality, avoid hype. Stay away from speculative AI stocks without meaningful revenue to support their valuations.

I just took a closer look at Jensen Huang’s letter, and buried in it was a line that had absolutely nothing to do with his marriage: “I love our country and our company, and I am committed to protecting U.S. national security.” That’s a declaration of political loyalty. Just ahead of NVIDIA’s earnings, Huang is trying to reposition himself from a “suspicious conduit” to a “patriotic businessman.” And that interpretation makes perfect sense. Come January, Elizabeth Warren is set to take over the Senate Banking Committee, while Chris Coons takes the helm of the Foreign Relations Committee. With subpoena power in hand, Democrats will have a whole list of targets: Elon Musk, Mark Zuckerberg, Jeff Bezos, and Jensen Huang among them. Huang admitting to infidelity isn’t simply about a sex scandal. It’s about sacrificing a pawn to save the king. I’d bet he’s already had private conversations with Trump. With Democrats having won the midterms, the AI industry is now squarely in their crosshairs. And Huang, as a Chinese-American who controls NVIDIA, makes an almost perfect political target. Rather than let Democrats destroy him with accusations of leaking sensitive technology to China, he may have chosen to admit to a personal moral failing first, hoping to neutralize the far more dangerous national-security narrative. The Democrats’ real game is to turn their oversight powers into political leverage throughout 2027 and 2028. Warren isn’t just going after Huang. She’s targeting an entire network of Wall Street institutions, Big Tech giants, and the Trump family. Huang is merely the first name to surface, not some specially selected centerpiece of the 2028 election strategy. This is the beginning of a new regulatory cycle, one driven by the political spoils of the midterm elections. And it’s only an appetizer for what Democrats have planned for 2028. The real reason Huang is being targeted isn’t infidelity, corporate misconduct, or even national security. It’s that he backed the wrong side. Huang positioned himself closer to the center of political power than almost anyone else in Silicon Valley. And in Washington, proximity to power is itself a form of risk exposure.

How I'll know the panic is actually fading. My go-to advanced signal is the VIX futures term structure: the curve you get by lining up VIX futures prices across expiry months. Normally the back months trade above the front months. That's contango. The market feels calm right now and sees more uncertainty further out, so it pays more for longer-dated protection. In a panic, traders rush into the front month, its price jumps above the back months, and the curve inverts. That means the market is pricing the risk right now. A lot of charts color contango yellow (sentiment back to normal) and inversion purple (panic still on). Right now we're deep purple. Deep inversion. The flip from purple to yellow, from inversion back to contango, is the signal the panic is receding. It usually comes later than the VIX peak, but it's a much better confirmation that the storm has passed. I'd pair it with two flow signals: • CTA selling has run its course. • Hedge funds start actively buying individual tech names again. My checklist for the next panic Nobody can predict the market, but every panic is a chance to calibrate. Next time, I'll go through these in order: • Is sentiment at an extreme? Are VIX, VVIX and VXN all printing rare single-day spikes at the same time? • Is positioning flushed? Have hedge funds de-levered enough, have CTAs finished their expected selling, and is retail still buying the dip with leverage? • Is the panic fading? Has the VIX futures curve gone from deep inversion back to contango? • Is new money stepping in? Are hedge funds going long tech names again? The first two tell you whether it has fallen enough. The last two tell you whether it can go up. If you jump in on the first one alone, you may be right on direction and still sit through another round of volatility. Where we stand: the first box is clearly checked. The second is only partly done. Hedge funds have cut hard, but CTAs still have selling to do and retail is still levered up. What I'm watching: • Whether the roughly $28 billion of expected CTA S&P futures selling actually gets done over the coming week. • The VIX futures curve turning from deep purple back to yellow. • Hedge funds moving from cutting risk to actively buying tech names. • Whether retail's record leveraged-ETF dip buying gets washed out. Until those line up, I think the direction is right but the timing is early.

How I'll know the panic is actually fading. My go-to advanced signal is the VIX futures term structure: the curve you get by lining up VIX futures prices across expiry months. Normally the back months trade above the front months. That's contango. The market feels calm right now and sees more uncertainty further out, so it pays more for longer-dated protection. In a panic, traders rush into the front month, its price jumps above the back months, and the curve inverts. That means the market is pricing the risk right now. A lot of charts color contango yellow (sentiment back to normal) and inversion purple (panic still on). Right now we're deep purple. Deep inversion. The flip from purple to yellow, from inversion back to contango, is the signal the panic is receding. It usually comes later than the VIX peak, but it's a much better confirmation that the storm has passed. I'd pair it with two flow signals: • CTA selling has run its course. • Hedge funds start actively buying individual tech names again. My checklist for the next panic Nobody can predict the market, but every panic is a chance to calibrate. Next time, I'll go through these in order: • Is sentiment at an extreme? Are VIX, VVIX and VXN all printing rare single-day spikes at the same time? • Is positioning flushed? Have hedge funds de-levered enough, have CTAs finished their expected selling, and is retail still buying the dip with leverage? • Is the panic fading? Has the VIX futures curve gone from deep inversion back to contango? • Is new money stepping in? Are hedge funds going long tech names again? The first two tell you whether it has fallen enough. The last two tell you whether it can go up. If you jump in on the first one alone, you may be right on direction and still sit through another round of volatility. Where we stand: the first box is clearly checked. The second is only partly done. Hedge funds have cut hard, but CTAs still have selling to do and retail is still levered up. What I'm watching: • Whether the roughly $28 billion of expected CTA S&P futures selling actually gets done over the coming week. • The VIX futures curve turning from deep purple back to yellow. • Hedge funds moving from cutting risk to actively buying tech names. • Whether retail's record leveraged-ETF dip buying gets washed out. Until those line up, I think the direction is right but the timing is early.

$IREN Jumps 10%+ After Hours! AI Cloud Revenue +183% QoQ, Backlog Hits $20.3B (Deep Dive) This report landed at peak fear, and it fits one label: growth is showing up, delivery is slipping. The growth engine has started, but delivery and funding are both a step behind. AI cloud revenue came in at $200M (+183% QoQ). Total revenue was $228M (+66% QoQ). Backlog rose to $20.3B (+22% QoQ). Quarterly capex was about $4.89B. The target of more than $4B in annualized revenue run-rate (ARR) has been pushed to January 2027, February at the latest. The stock took a roller-coaster ride. During the regular session on Nov 5 (ET), Bloomberg reported that Microsoft is tightening oversight of neocloud data center projects and named cabling problems at IREN's Texas campus. Add a broad tech selloff that day, and $IREN closed down 16.95% at $18.03, after touching a 52-week low of $17.155 intraday. After the print it traded at $19.95 at 19:59 ET, up 10.65% from the close. The after-hours bounce means the worst case didn't happen. It does not mean the good news has been delivered. Contracts and pricing are intact. The real tests for the next few months are the delays, the losses and the $6B funding gap. ■ 1. Financial Snapshot • Total revenue: $228M vs. $137M last quarter, +66% QoQ. Down about 5% YoY from $240M, which was mostly bitcoin mining revenue • AI cloud revenue: $200M vs. $71M last quarter, +183% QoQ. That splits into $135M of AI cloud services and $65M of AI cloud leasing • Bitcoin mining revenue: $28M, vs. $67M last quarter and $233M a year ago. Miners are being retired faster • Gross margin before D&A: 80%, up from 76%. AI cloud segment gross margin was 86%, flat QoQ • Adjusted EBITDA: $66M at a 29% margin, vs. $19M at a 14% margin last quarter • GAAP net income: a $263M loss. Last quarter's loss was $684M, including large miner impairments. A year ago IREN earned $385M, mainly from $665M of unrealized gains on financial instruments • Diluted EPS: -$0.66 vs. $1.08 a year ago • Operating cash flow: +$529M, but $645M of that came from higher deferred revenue tied to customer prepayments • Free cash flow: about -$4.36B ($529M of operating cash flow minus about $4.89B of combined spending on PP&E, compute hardware and intangibles) The loss is worth breaking apart. D&A was $148M, stock comp $106M, unrealized losses on financial instruments $28M, and a fair-value decline on held-for-sale assets $21M. Together these non-cash items total about $300M, more than the entire net loss. That does not make the loss fake. GPU depreciation is a real economic cost, and SG&A jumped from $128M last quarter to $217M, partly because Mirantis staff are now consolidated after the acquisition. The balance sheet is levering up fast: Unrestricted cash ended the quarter at $3.52B, down from $5.90B at end-June, plus $2.26B of restricted cash. Total debt is about $10.2B: $6.66B of converts, $1.54B of USPP senior notes and $1.28B of GPU financing. Net debt on the company's definition is $4.43B, up from just $216M last quarter. The good news: most of the converts carry low coupons and mature in 2032 or later, so there's no maturity wall to refinance in the next two years. Customer prepayments have become a major funding source: The deferred revenue balance is $2.54B, up from $1.84B at end-June. On one newly signed five-year contract, the customer's prepayment covers about 90% of the related GPU capex. The share count keeps growing, with no buybacks and no dividend: IREN issued 12.27M shares via the ATM this quarter for $541M in net proceeds. Basic weighted shares were 399M vs. 271M a year ago, up about 47%. Common shares outstanding rose from 407M on Sep 30 to 424M on Oct 30. Next-quarter guide (the key to this print): Management gave no traditional revenue range. It only said the December quarter will recognize one quarter of the roughly $1B operating ARR reached at end-September. That's about $250M of revenue, roughly +10% QoQ. Most of the new capacity from Horizon 2–4 and Mackenzie comes online in late December or early 2027, so it adds very little this quarter. Cash opex is expected to rise $50–60M QoQ. On that math, December-quarter adjusted EBITDA will most likely fall QoQ. The real revenue step-up has to wait for January–March 2027. ■ 2. Results vs. Expectations • Revenue: $228M vs. Zacks consensus of $219.95M, a beat of about 3.7%. Against the $230.45M compiled, a miss of about 1.1%. Broadly in line • Adjusted EBITDA: $66M vs. about $64M expected (, a slight beat of about 3% • EPS: -$0.66 vs. Zacks consensus of -$0.49. The loss was $0.17 wider than expected, clearly worse • Timing of the >$4B ARR target: pushed from "end of 2026" to "January 2027, end of February at the latest," short of prior guidance • December-quarter revenue: about $250M. No reliable, dated single-quarter consensus could be found, so there's no line-by-line comparison. But as of Nov 4, S&P Global's FY2027 (ending June 2027) revenue consensus was $2.81B, which requires about $2.3B from the second half of the fiscal year That last point is the overlooked risk. Even if the full $4B ARR is live by end-February, the two quarters covering January–June will struggle to reach $2.3B combined. The press release itself flags that ARR is an operating metric and that recognized revenue "may be significantly lower." FY2027 revenue consensus will most likely have to come down. Key context: why "in line" carried so much weight this time. Going into the print, IREN was already priced for the worst case. FY26 Q4, reported Aug 27, showed revenue of $137M vs. $157M expected, plus a $684M loss on miner impairments, and the stock fell about 7% after hours. In October the whole neocloud group crashed. Microsoft's capex came in below expectations, Gartner warned that most neoclouds will struggle to survive independently, and the market started worrying about rising funding costs across the sector. IREN dropped about 42% in the month. Two days before earnings, B. Riley cut its target from $91 to $64 and Bernstein halved its target from $100 to $55. The stock fell 11.24% on Nov 4 and another 16.95% on Nov 5, a drawdown of about 77% from the 52-week high of $76.87. From there, as long as the delivery delay didn't turn into a contract cancellation, the market was going to exhale. That's the logic behind the after-hours bounce. ■ 3. Why the Stock Jumped 10%+ After Hours 1. The Microsoft contract is intact, and the delay is only 1–2 months (the core driver) The daytime selloff was trading on Bloomberg's report of "Microsoft tightening oversight, with possible delays of weeks or months." The earnings release was more specific. The delay comes from the customer asking to change the network configuration to support frontier-model training at 200MW scale. That is a design change outside the original contract scope, and the extra capex is "not material." The project timeline is clearer too: Horizon 2 GPUs are racked and in commissioning, all Horizon 3 GPUs are on site and the data hall is energized, and Horizon 4 GPUs arrive within weeks. The Microsoft contract, worth $9.6B, is still the largest piece of the backlog. A 1–2 month delay is far milder than the "Microsoft may walk" scenario priced in during the day. 2. Pricing is rising, and the demand evidence is hard Annualized revenue per MW is up about 11% on three-year contracts and about 12% on five-year contracts. A five-year contract signed after the quarter runs at about $22M per IT MW, and short-term contracts exceed $40M per MW. Older H100s were re-contracted for multiple years at twice the original price. With the market worried about a compute glut, rising prices are the most direct rebuttal on demand. 3. Backlog growth is accelerating, and the customer mix is deliberately tiered Backlog is $20.3B, +22% QoQ, and about two-thirds of it comes from investment-grade customers, including Microsoft and NVIDIA. After the quarter, IREN also expanded its contract with a "leading frontier AI lab." The other third is weaker-credit customers, and the company hedges that risk by demanding higher prepayments. This speaks directly to the worry that the customer base is concentrated in Microsoft. 4. Pulling back capex to buy funding certainty Remaining FY2027 capex was cut to about $15B. About $9B of that is already covered by cash, committed GPU financing and customer prepayments, which leaves about $6B still to secure. Management was explicit that the pullback reflects the funding environment, not weaker demand, and the 0.8GW capacity target for end-2027 stands. For a stock that crashed on funding fears, swapping "expand at any cost" for "build at the pace returns allow" is the stance the market wanted to see. ■ 4. Business Breakdown: Where the Growth Is Coming From Already proven by the numbers: • AI cloud is now clearly the core business. Its share of revenue jumped from about 52% last quarter to about 88%, and segment gross margin held at 86%, so GPU cloud unit economics have not worsened at scale • Delivery capability: Horizon 1 took about 14 months from groundbreaking to a validated GB300 cluster and earned NVIDIA's GB300 NVL72 Exemplar Cloud certification. Management compares that with a typical industry build cycle of about 18 months • Operating ARR doubled in one quarter to about $1B, and operating capacity rose 127% QoQ to about 0.1GW • Mining exit: miner retirement should be largely complete by the end of the December quarter, and the bulk of miner impairments was booked last quarter Still only early signals: • $4B ARR: contracts and funding are in place, but Horizon 2–4, Mackenzie and other projects all have to come online within January–February. Any vendor slip pushes it out further • Sweetwater 2GW campus: Sweetwater 1 (1.4GW) is connected to the grid. It has been conditionally included, together with Sweetwater 2, in ERCOT's Batch Zero baseload process, but the conditions weren't disclosed. Texas data center reviews and a possible moratorium are also listed as risks • The 0.8GW target, >5GW of secured power and >35GW development pipeline: the company itself admits many early-stage sites will never get interconnection • Growing data center assets from $6B today to $10B by June 2027, about $25B in total assets once GPUs are included, assumes the $6B funding gap gets filled ■ 5. Earnings Call: What Really Mattered Management framed the delay as the customer asking for more, not as an execution failure, but the third-party account doesn't fully match. Management attributed the delay to Microsoft's requested network redesign and vendor timing. Bloomberg's report said cabling problems disrupted the construction schedule. Add April's generator fire and the hardware delivery delays disclosed in August, and this is the third delivery-related setback in a year. Management did not directly address Bloomberg's report that Microsoft is internally debating whether to keep signing new neocloud contracts. That question decides whether IREN wins follow-on Microsoft orders, and right now there is no evidence either way. The narrative has moved from "scale" to "deliverable, financeable, profitable." Management said on its own that investor questions have shifted from how big the contracts are to whether IREN can build them, fund them and make money on them. Its answer is three-layer vertical integration (land and power, data centers, AI cloud), with the emphasis that IREN "doesn't lease other people's data centers and doesn't carry lease liabilities on the balance sheet." The company has raised about $19B over the past 12 months, including about $6.5B of GPU financing since May at roughly 6%–9.5%. Last quarter the message was "full-speed expansion, $25–30B of 2027 capex." This quarter is a clear shift to defense. Unit returns are the bulls' strongest argument: GPU payback is about 18 months, and faster with higher prepayments. Put that next to $22M-per-MW five-year pricing and H100 renewals at double the price, and management's message is clear: price increases are offsetting higher GPU costs. The company has not locked in all of next year's GPU purchases, to limit price risk. Vera Rubin capacity arrives later in 2027 and will be priced off recent contracts. But "18-month payback" is management's own estimate. No contract-level IRR, interest cost or depreciation life was given, so outsiders can't verify it. The $6B gap is meant to come from private and infrastructure capital, with no timeline. Management cited channels beyond public markets, including private markets and infrastructure funds, and noted capex can flex up or down. It did not say how much would come from common equity. With the stock down from an average ATM price of about $44 to below $20, the dilution cost of equity funding has more than doubled. That is the core of the bear case. Management's stance: "We can wait and accelerate when markets improve." That cuts both ways. It concedes the current environment doesn't support aggressive expansion, and it also implies the FY2028 growth curve may be flatter than sell-side models. ■ 6. What Wall Street Is Saying Per StockAnalysis (S&P Global data, as of Nov 4, 2026), the 20 covering analysts have a consensus "Buy" rating, with an average price target of $75.42, a median of $76.50 and a range of $37–139. Against the Nov 5 close of $18.03, the average target implies about 318% upside. Even the lowest target is more than double the current price, so targets are badly lagging the stock, and a wave of cuts over the next few weeks is likely. • Bernstein: Outperform, PT $100 → $55 (11/4), citing uncertainty over new Texas interconnection approvals and rising borrowing costs for AI infrastructure • B. Riley: Buy, PT $91 → $64 (10/30) • H.C. Wainwright: Buy reiterated, PT $74 (10/28) • Rothschild & Co Redburn: initiated at Neutral, PT $40 (9/21) • Northland: initiated at Buy, PT $99 (9/18) At the time of writing, no post-earnings rating or target changes (from the Nov 5 after-hours session onward) had turned up. The bull case: 1. Two-thirds of the $20.3B backlog is investment-grade, and the $4B ARR is contracted and funded. A 1–2 month slip doesn't change the end state 2. Rising prices, high prepayments, ~18-month GPU payback, and AI cloud gross margin steady at 86% 3. Owned land and power (>5GW secured) get scarcer as interconnection approvals tighten, a point Bernstein also made 4. Valuation has been crushed to about 3x EV/target ARR, so any delivery could trigger a re-rating The bear / cautious case: 1. Microsoft is tightening its neocloud partnerships. The largest customer's appetite at the margin is falling, which puts follow-on orders in doubt 2. A $6B funding gap plus a collapsed stock makes equity dilution far more expensive, and net debt jumped from $216M to $4.43B in one quarter 3. Repeated delivery setbacks (generator fire, hardware delays, cabling problems, design changes) are eroding execution credibility 4. The GAAP loss was worse than expected, December-quarter EBITDA may fall QoQ, and FY2027 revenue consensus has room to come down ■ 7. Valuation & Price Scenarios Current price: $18.03 (Nov 5 close), $19.95 after hours (19:59 ET, Nov 5). On 424M shares as of Oct 30, the closing market cap is about $7.65B. Add $4.43B of net debt and enterprise value is about $12.1B. Where valuation sits: • EV / target ARR ($4B): about 3.0x • EV / FY2027 consensus revenue ($2.81B): about 4.3x • EV / FY2028 consensus revenue ($7.41B): about 1.6x. But FY2028 revenue first requires about $15B of capex, and by then debt and share count will both be much higher, so this multiple flatters the stock • Vs. its own history: at about $50.6 in late April, market cap was about $16.8B. Today it's less than half that • Vs. peers: per Zacks' pre-earnings preview, IREN traded at 2.02x forward 12-month P/S vs. 2.58x for the industry, 4.26x for Applied Digital and 8.51x for TeraWulf. After the Nov 4–5 drop, it's even lower now GAAP EPS is still negative, so P/E doesn't apply. The scenarios below use annualized EBITDA at full $4B ARR × a multiple − net debt. The EBITDA assumption: AI cloud gross margin of about 86%, minus about $700M of annualized cash opex after next quarter, gives roughly a 65% EBITDA margin on $4B ARR. The base case uses 60% for cushion. Because IREN still has to keep borrowing and issuing stock, the base case assumes net debt rises to about $7B on about 450M shares. All of this is subjective modeling, not consensus. • Bear case ($8–12): Microsoft delays further or scales back, effective ARR is only $3.2B by mid-2027, 55% EBITDA margin (about $1.76B), 6–7x EV/EBITDA, $7B net debt, 450M shares. With this much leverage on the equity, a small EBITDA shortfall gets amplified in the share price • Base case ($22–32): full $4B ARR live by end-February, 60% EBITDA margin ($2.4B), 7–9x EV/EBITDA, $7B net debt, 450M shares. That's roughly +22% to +78% from the close • Bull case ($43–57): Mackenzie and Sweetwater 1 contribute early, ARR reaches $5B before end-2027, 65% margin ($3.25B), 9–11x EV/EBITDA, net debt up to $9B, 470M shares. That lands near Bernstein's cut target of $55 One caveat: EBITDA is not cash flow to shareholders. Depreciation on billions of dollars of GPUs and 6%–9.5% interest on GPU financing will eat a lot of the profit, so be careful putting high EV/EBITDA multiples on this. ■ 8. Bottom Line & What to Watch The verdict: This report proves IREN's AI cloud business works on its own terms: revenue +183% QoQ, 86% gross margin, rising prices, backlog +22%. It also exposes the two weakest links of a capital-heavy, highly levered neocloud that leans on one big customer: the delivery schedule and the funding pace. The 10.65% after-hours bounce, after a drawdown of about 77%, is the market repricing for "no worse news." Still unproven: whether the $4B ARR goes fully live by end-February and turns into revenue; whether Microsoft signs new orders; whether the $6B gap can be filled without heavy, cheap equity issuance; and whether, after heavy depreciation and interest, the business can ever produce positive GAAP profit and free cash flow. What to watch next: 1. Horizon 2–4 delivery announcements are the first hard checkpoint. Whether IREN announces full delivery to Microsoft and the start of billing by end-January decides whether the ARR target slips a second time. 2. The structure and cost of the $6B raise. Watch the terms of data center asset financing, private or infrastructure fund deals, and how much the ATM gets used. Heavy issuance around $20 would sharply dilute per-share value. 3. Order flow from Microsoft and new customers. Watch whether the expanded frontier AI lab contract becomes a new large order that can offset Microsoft pulling back at the margin. 4. December-quarter EBITDA and cash opex. After a $50–60M rise in cash opex, can the EBITDA margin stay above 20%? 5. Texas interconnection and regulatory policy. The conditions of ERCOT Batch Zero, Texas data center reviews and a possible moratorium bear directly on the value of Sweetwater, the biggest asset. Conclusion: Short term, $IREN is still a high-beta neocloud sentiment stock. Bloomberg reports, Nasdaq swings and peer headlines will move it more than fundamentals, and the after-hours bounce may not hold. Medium to long term, if the $4B ARR lands on the new timeline and funding comes mostly from debt and prepayments, the current ~3x EV/target ARR leaves a lot of room for a re-rating. Investors who can stomach high volatility and believe AI compute will stay supply-constrained can wait for the Horizon deliveries and the funding plan, then build in tranches. There's no need to chase the bounce during a news vacuum. For existing holders, hitting the ARR target in January–February is the test for staying in. Any one of these breaks the thesis and means stepping aside: the ARR target slipping a second time to after March, Microsoft cutting the existing contract, or a large equity raise at a steep discount.