By March 2027, A Big Tech Giant Will Cut AI Capex

There is a point where even Wall Street stops clapping for "gobs of money." My call: before March 31, 2027, at least one of Apple, Microsoft, Alphabet, Amazon, or Nvidia will pull a visible $20 billion or more out of its AI capex plans and blame it, at least in part, on investor nerves about profit and free cash flow.

The consensus story says the opposite. The supercycle is ordained, infrastructure is the constraint, and any number under a couple hundred billion is "under investing in the future of intelligence." Morgan Stanley has literally told clients that spending hundreds of billions is fine because everyone else is doing it and cloud is growing fast.

When the sell side is quoting your peer group as your capital budgeting framework, you are closer to a turn than to a new normal.

The Bet, Stated Cleanly

Here is the scorable version.

By March 31, 2027, at least one of the Big 5 will:

• Reduce or push out at least $20 billion of previously stated, numeric AI related capex plans for the 2025–2027 period, compared with its own prior guidance, and
• Explicitly connect that move to concerns from investors or the board about profitability, free cash flow, or balance sheet capacity, not just to "timing" or "supply chain."

If they quietly smooth spend a little or hide it in a new line item, that does not count. This is a call for a public walk back that you could circle on an earnings slide.

The Arms Race Is Real, The Math Is Uglier

Start with the headline numbers. Roughly for 2026 alone:

Alphabet: about $195–205 billion in capital investment. Amazon: about $220 billion. Microsoft: $175–190 billion. Meta, as a benchmark, wants $130–145 billion. Nvidia and SK Group are talking about more than $500 billion of AI infrastructure. The street thinks total cloud capex could top $1.2 trillion around 2027.

It is not imaginary spend either. Microsoft just put $41 billion out the door in one quarter, up more than 70 percent year on year. Free cash flow fell 23 percent to $19.6 billion. Alphabet hiked its guidance, the stock sold off even as EPS tripled expectations, and Morgan Stanley rushed in with a reassuring "that is OK."

On the other side of the ledger you have very real demand. AWS revenue is up about 37 percent to more than $42 billion a quarter. Amazon says its AI and custom chip businesses are both north of $25 billion run rate with triple digit growth. Google Cloud is growing more than 80 percent year on year. Azure crossed $100 billion in revenue with growth in the 40s. Everyone has a backlog that sounds like a sovereign debt market. Microsoft alone cites $678 billion of commercial cloud backlog and $329.1 billion of future data center lease commitments.

This is the dream setup for an arms race: huge visible demand, visible shortages in compute and power, and executives who know what happens to the CEO who "under invests in the next platform." The only thing missing is a brake pedal.

Where The Brake Pedal Comes From

The brake does not come from moral restraint. It comes from numbers that stop going up in the right column of a quarterly deck.

Today investors are grading on a curve. Amazon boosts 2026 capex from about $200 billion to roughly $220 billion, says it still cannot meet demand, and the stock rips because AWS and AI are growing like weeds. Microsoft’s free cash flow takes a 23 percent hit and no one panics because the backlog looks bottomless and Copilot has 30 million paid users. Alphabet gets whacked for a quarter when it lifts capex guidance, then wins forgiveness when analysts decide the spend is "rational."

That tolerance is conditional. It lasts exactly as long as three fragile things hold at once:

First, AI revenue growth stays spectacular. Triple digit AI run rates at Amazon, 80 plus percent at Google Cloud, 40 plus percent at Azure. Any visible deceleration while capex still ramps and the mood flips from "you are buying the future" to "you are paying for someone else’s margin."

Second, the market stays happy to fund this. Ultra high equity valuations and easy debt make it painless to build twenty data centers before breakfast. A couple of inflation scares, a credit wobble, or an ugly tech selloff, and boards will remember that they also promised buybacks and dividends.

Third, the backlog actually converts. CoreWeave is the stress test here. It has nearly $100 billion in backlog, including around $21 billion from Meta, revenues more than doubling, and yet the stock is trading at a steep discount to prior highs because capex and interest expense are beating realized cash. That is the movie the hyperscalers are betting they never star in.

Break any of those three, and the same analysts who are blessing "gobs of money" will rediscover the phrase "capital discipline" in about six hours.

Who Blinks First?

You do not need a failure of the AI story. You just need a wobble at one firm whose balance sheet and investor patience are a little less bulletproof.

Alphabet already has the clearest tells. It is pushing toward $200 billion in annual capex, it got an immediate market slap the last time it raised guidance, and it does not have Apple level cash hoards or Microsoft level enterprise lock in. If Google Cloud growth normalizes and Gemini pricing power is less heroic than the pitch deck, it is the most obvious candidate for a "we are phasing spend to align with returns" slide that quietly takes $20–30 billion out of the 2026–2027 plan.

Amazon has the most convincing AI economics so far, plus a cultural allergy to talking about margins until forced, but even it is saying that $220 billion will not be enough. If rates back up or retail has a bad year, shareholders will eventually ask whether the company is a store with a cloud or a GPU hedge fund that also sells paper towels.

Microsoft and Apple have the deepest cushions. They can hide a lot inside general "cloud" and "platform" capex, and they know how to move spend off balance sheet with leases and partners. That makes them less likely to be first to a clean, countable cut. Nvidia is trickier, since much of its "capex" is really a mix of foundry commitments, joint ventures, and customer funded builds, but if data center growth slows against that $500 billion plus infrastructure narrative, even it can be pushed into a more modest ramp.

The key point: someone will decide they would rather miss a bit of theoretical AI demand than explain to investors why free cash flow is down again while buybacks are on pause. That is when a multi hundred billion capex promise becomes a $20 plus billion capex apology.

How This Trade Loses

There is a world where I am wrong. In that world, AI demand remains so violently strong that these companies can both spend the $1.2 trillion and still grow free cash flow, because pricing power and efficiency gains outrun the build. Backlogs swell, not shrink. Rates stay tame. Regulators nibble around the edges but do not slow a single data center.

In that version, management will trim and reclassify at the margins, but no one will put a big red line through a guidance slide and blame shareholders. The arms race rolls right through 2027 and the only cuts are in analyst estimates that were too low.

I am betting instead on something more boring and more common: grand capex narratives meeting quarterly math. These companies have told the world exactly how much they plan to spend on AI. They have made the capex wall visible. Now we get to watch who hits it hard enough to leave a trace.

When they do, the press release will talk about "phasing," "disciplined allocation," and "aligning investment with proven demand." The translation will be shorter: your GPU dreams are now subject to the dividend policy.