Back to index

Do Not Buy The AI Basket; Buy The Names Whose Capex Gets Paid For

Part 1: The Market Is About To Ask The Right Question

The consensus has not missed the fact that AI demand exists. That was last year's question. The question tonight is harsher and more useful: which companies can spend hundreds of billions on AI infrastructure and make the market believe the spending is self-funding?

That is why this earnings cluster matters. Microsoft, Alphabet, Meta, and Amazon are not just four earnings reports. They are the market's cleanest real-time audit of the AI capex cycle. The temptation is to call it "AI good" or "AI bad" by tomorrow morning. That will be too crude. The better split is productive capex versus open-ended capex.

The completed topic-research rounds around msft-goog-meta-amzn-earnings-apr29 sharpened the frame. Round 001 started with a broad capex credibility test. Round 002 fixed the weak part: capex does not mean the same thing for all four companies. Round 003 turned the setup into gates. Round 004 took scenario odds. Round 005 reduced tomorrow to execution rules.

My stance is direct: the best pre-print expressions are Microsoft and Amazon, in that order for quality, with Amazon having the cleaner trading setup if AWS validates. Alphabet is repair-only. Meta is an avoid unless the call proves ad AI return on investment and spend discipline. QQQ is a thermometer, not the trade.

This is not fence-sitting. It is expression selection.

Microsoft has the cleanest business-quality setup. Azure and Copilot give investors an understandable path from AI infrastructure spend to enterprise revenue. Satya Nadella can explain demand; Amy Hood has to prove it is not eating the model. The stock, however, is not a free long. The recursive TA says MSFT is flat until 424.95 accepts and follows through 425.78-425.82; below 423.68, it turns defensive toward 421.79. That chart state does not contradict the quality argument. It says the market has not yet paid for it in a clean way.

Amazon has the best trading asymmetry if AWS does its job. AWS growth and margin are a more direct capex-payback test than almost anything else tonight. The tape is also better than GOOG, META, or MSFT, but it is not clean. AMZN held 261.03, yet remains capped until 264.15 repairs and 265.91 confirms. The memo stance is: buy AMZN only if AWS quality and the tape agree. If AWS is strong but 264.15 fails, the market is telling you the spend story is not enough.

Alphabet is the trap candidate. It can absolutely rally if Google Cloud margin and Search monetization are credible. But it enters as the weakest tape: 349.04 and 349.90 rejected, with 344.28 the downside support. That means GOOG does not get the benefit of the doubt. It must repair 349.90; a merely decent print is not enough.

Meta is the most binary but not the best trade. It has the clearest ad-monetization question and the clearest risk of open-ended AI spending. The adversary caught the key weakness: 668.68 is too close to the current setup to carry the thesis by itself. A break there can be mechanical. I need either a disciplined capex/ad ROI story that reclaims 670.84 and 672.77, or I leave it alone. If spend sounds open-ended and 663.87 fails, the downside confirms.

The historical analogs are not decorative. hyperscaler-capex-credibility-test is the right direct analog: the market rewards capex when it is tied to measurable demand, and punishes it when it looks like faith-based spending. The meta_capex_disappointment_2022 analog remains the warning: revenue growth can be overwhelmed by investor rejection of open-ended spend. The aws_growth_deceleration_2022 analog is the cloud warning: AWS, Azure, and GCP growth rates can dominate otherwise acceptable consolidated numbers.

The adversary's best objection is worth preserving: capex can be bullish for suppliers and bearish for spenders at the same time. A capex raise is not bullish by itself. A capex raise with backlog, RPO, cloud margin, ad ROI, and free-cash-flow discipline is bullish. A capex raise with vague "investing for the future" language is not visionary. It is multiple compression waiting for a catalyst.

So the stance is:

Part 2: The Trade Is Capex That Pays Rent

Everyone is watching EPS. EPS is not the trade.

The trade is capex that pays rent.

If Microsoft says Azure demand is strong, Copilot is monetizing, cloud gross margin is not falling apart, and Amy Hood can guide without sounding like she is apologizing for the data-center bill, MSFT is the quality long. But the trigger is not the press-release adjective. It is 424.95. Above that, with 425.78-425.82 follow-through, the market is accepting the story. Below 423.68, the story may still be good, but the stock is not ready.

If Amazon prints an AWS reacceleration with margin support and a clean Q2 guide, AMZN is the better trade. It has the right proof point and the better relative tape. But again, the tape has to pay. Hold 261.03, reclaim 264.15, push through 265.91. That is the clean path. Lose 261.03, and I do not care how good the AWS paragraph sounds. The market is rejecting it.

Alphabet is not my long unless it changes character. A weak tape needs a strong repair. The stock has to reclaim and hold 349.90, and the reason has to be real: Google Cloud margin, Search monetization, YouTube strength, or capex explained by profitable demand. If it rejects 349.04/349.90, it stays the weakest name. If 344.28 breaks, stop pretending the print was misunderstood.

Meta can squeeze. I am not denying that. The problem is that the good version and bad version both start with "we are spending a lot on AI." The difference is whether ad tools are visibly paying for it. If Susan Li gives numbers and the stock reclaims 670.84 then 672.77, fine, there is a trade. If Zuckerberg sells a dream and the stock loses 668.68 then 663.87, that is not bravery. That is the market rejecting the budget.

My probability map is still:

The most important thing tomorrow is not the opening gap. The opening gap is usually the most expensive way to learn the least. QQQ and mega-cap ETF flow can move everything together first. I want the first pullback, the call digestion, the midday relative-strength check, and then the close.

The staged playbook is:

The clearest bullish market outcome is not "all four are green." It is MSFT and/or AMZN holding repair levels while GOOG and META are judged more selectively. That says the market is paying for productive capex, not blindly buying the AI basket.

The clearest bearish outcome is not "one company misses." It is two or more companies beating headlines while supports fail because management cannot explain the spending. That is the capex-fade scenario, and it is the one most people will underestimate because it will not look bearish in the first headline.

The memo's trade stance is therefore simple:

Run risk through MSFT/AMZN only after confirmation. Treat GOOG as a repair trade. Treat META as an ROI proof trade, not an AI story. Use QQQ as the beta filter. If the market buys productive capex, participate. If it buys slogans, fade the enthusiasm.

Tripwire Sheet

Related Research