Apple sat out the AI spending race - and the market is rewarding it

Information Technology 5 min 80 sources

As Google burns $6bn in cash on the AI buildout and Alphabet's forecast spooks investors, Apple's refusal to pour in the capex made it the Mag7's top performer - while chipmakers and model labs race to make tomorrow's compute cheaper.

Key takeaways

Apple's refusal to spend big on AI infrastructure made it the top Mag7 performer this week, as Alphabet's rising capex and $6bn cash burn worried investors.
The cost of running AI keeps falling - cheaper models from Anthropic, Microsoft and Google, and cheaper chips from Samsung, Nvidia and Etched - which is why holding off can be worth more than committing early.
Governments now treat open AI models as national infrastructure, and the physical AI buildout is straining power grids that ordinary households help pay for.

While its rivals pour tens of billions into artificial-intelligence infrastructure, Apple has mostly refused to join in. In the past week, that looked less like falling behind and more like a plan. Apple shares led the “Magnificent Seven,” the seven big US tech firms that drive the market. Investors treated its light spending as a hedge against everyone else’s heavy bets [57][63].

The company that isn’t building

Google’s parent Alphabet raised its capital-spending forecast again after a strong quarter for its cloud business [19][15]. The Financial Times reported the company burned through $6 billion in cash as that spending climbed [74]. Capital spending, or “capex,” is the money a firm sinks into long-lived assets - here, data centres full of chips. Across Big Tech, that outlay is now large enough to squeeze free cash flow, the money left after a company pays its bills [3].

Investors noticed. Alphabet’s shares fell after earnings despite the beat, a move CNBC called an “ominous read-through” for every mega-cap planning to spend big [70]. IBM cut its yearly revenue-growth forecast, saying customers are diverting budgets into AI infrastructure and away from IBM’s software [33].

Against that backdrop, Apple’s near-absence from the buildout became its selling point. Its chief executive Tim Cook delivered what was reported as his final earnings report [73]. Under scrutiny was Apple’s pricing power: its ability to charge more without losing buyers. Apple still buys AI services rather than building the vast infrastructure to run them itself. This week, the market decided that patience had value.

Why waiting is worth something

The reason is not just thrift. It is that the cost of AI compute is falling fast - so a dollar committed today buys less than a dollar spent later will.

Look at what shipped this week. Anthropic released Claude Opus 5, priced at $5 per million words of input and $25 per million of output [13]. It pitched the model as a cheaper “daily driver”: near-frontier ability at a fraction of the top model’s cost [34]. Microsoft launched in-house models it says cut costs up to 89% versus OpenAI, now running inside Bing, Excel, GitHub Copilot and Azure [9]. Google put out Gemini 3.6 Flash, faster and cheaper, lifting a coding score from 37% to 49% - while its promised flagship 3.5 Pro slipped again [29].

The chips underneath are getting cheaper to run, too. Google is reported to be designing a new chip, nicknamed “Frozen,” to run its models far more efficiently [78][55]. Samsung struck a pact with Broadcom, the custom-chip designer, to cooperate on manufacturing worth more than $200 billion through 2030 [39]. Nvidia signed a $1.5 billion deal with Amkor to expand chip packaging in the US [27]. And Etched, a startup building chips that run only the “transformer” designs behind ChatGPT and Claude, hit a $10.3 billion valuation [2]. That is double its worth seven months ago, on $1 billion of booked orders.

Here is the tension. If compute keeps getting cheaper, whoever waits buys more capability per dollar next year. But the firms building the data centres are the ones making that possible. And if AI becomes central to the phone in your pocket, a company that only rents may pay a premium to catch up. It may even find the door held by a rival. Apple’s patience is a bet that the option to wait is worth more than the head start it forfeits. That bet is not yet settled.

Open-source AI becomes a matter of state

The other big shift this week was governments treating open AI models as national infrastructure. An “open” model is one anyone can download and run, rather than access through a company’s paid service.

At an APEC summit in Chengdu, China, 21 economies including the US and China signed a statement backing open-source AI [4]. It came with a new condition: “strong security assurance” through development and deployment. Days later, Nvidia, Microsoft, Meta and IBM signed a public letter urging US lawmakers to avoid “premature restrictions” on open models. Nvidia’s Jensen Huang warned that curbs would push innovation overseas [6][37].

Behind the diplomacy sits a fear with a technical name: distillation. It means using a big, powerful model to train a smaller cheap one that copies much of its skill. Jeff Dean, Google’s AI lead, has described it as routine engineering [42]. Then China’s Moonshot lab released a model that markets judged competitive with the best American systems [68]. Washington began asking whether distillation lets rivals copy US breakthroughs on the cheap [42]. What was a wonky training trick is now argued over as a security threat.

The bill nobody signed up for

End on the part that is hard to reverse: the power. This week a single fallen power line outside Washington took the grid more than ten minutes to recover from [79]. The cause: more than three gigawatts of data centres cut their draw within about 30 seconds, spiking voltage from Virginia to Chicago. No blackout, but lights flickered across a grid serving 67 million people. One engineer called it “the canary in the coal mine” [79].

The politics followed. President Trump expanded a voluntary “Ratepayer Protection Pledge,” asking utilities, developers and states to keep the data-centre boom from raising household electricity bills [51]. It is voluntary, and it exists because ordinary people are already seeing their bills climb to pay for machines they will never touch.

A data centre, once built, is concrete, steel and a grid connection you cannot un-pour. That permanence is exactly why the option to wait - Apple’s whole wager - has a price on it worth watching.

02 · Lesson · why it matters

The value of not deciding yet

When a choice is costly to reverse and the facts keep changing, keeping it open has real worth - and rushing in quietly spends it.

The company that won by standing still

Everyone else was building. Google, Microsoft, Amazon and the rest have been pouring tens of billions into data centres, chips and power. Apple mostly watched. For a while that read as a company caught flat-footed, missing the biggest shift in tech.

Then, this week, Apple’s shares led the pack. Investors decided its restraint was not a failure of nerve but a stance worth paying for. The company that committed the least to the future everyone was racing toward became, for now, the one the market trusted most.

That is a strange result, and worth sitting with. Doing nothing looked like the smart move.

What a kept-open choice is worth

There is a quiet idea underneath this. When you face a big, hard-to-undo choice, and the facts around it keep changing, the freedom to wait is itself worth something. Not deciding is a decision - sometimes the good one.

Two things give waiting its value. The first is that the commitment is costly to reverse. You cannot un-pour a data centre; the concrete, the steel and the grid connection stay whether the bet pays off or not. The second is that information keeps arriving. Every month, AI chips get cheaper to run and models get cheaper to use. A dollar committed today buys less than a dollar spent next year will.

Put those together and the shape appears. When you can’t take it back, and tomorrow you’ll know more, there is real worth in keeping the choice open until the fog thins. You see the small version everywhere. You don’t buy the winter coat in October when the shop keeps cutting the price and you’re not even sure how cold it’ll get. You wait. The waiting has a value, even though no one hands you a receipt for it.

Waiting is not free

Here is the part that keeps this honest: patience has a price too.

The firm that waits gives up the head start. Say AI turns out to be central to the phone in your pocket. The company that only rents its compute may pay a premium to catch up, or find the door held shut by a rival who built early. The ones sinking billions into data centres are the ones making the cheaper future possible in the first place. Building has real value; so does waiting. Neither is automatically the wise move.

So this is not a story about clever Apple and foolish Google. It is a story about two honest guesses at a future nobody can see. One kept its options open and paid for it in lost ground. The other bought the head start and paid for it in cash and risk. Which was right depends on facts that haven’t arrived yet.

Why patience is so easy to punish

Now the harder layer - the one that isn’t about Apple at all.

The value of a kept-open choice almost never shows up as a number. A balance sheet records the billions a company spends; it has no line for the billions a company wisely did not spend yet. Spending is visible. Restraint is invisible. So the manager who builds looks bold and busy, while the manager who waits looks idle - even on the days waiting is the better call.

That is not a law of nature. It is a choice about what we measure and what we applaud. We built our accounting, our headlines and our instincts to reward the visible move over the invisible one. It poses as plain common sense: of course you judge people by what they do. But it quietly tilts the whole field toward action. It takes an unusual moment, like this week, for patience to get its due.

And the bets reach far past the boardroom. The household seeing its power bill climb is helping pay for a data centre built on a wager that may not pay off. The worker whose company committed early gets cut when the numbers turn. None of them sat in the room where the choice was made, yet all of them are inside its consequences. When a handful of firms place enormous, hard-to-reverse bets, the rest of us are wired into the outcome whether we chose it or not.

What no seat can see

Step back and the humbling thing is plain. Nobody knows which way this breaks. Apple’s patience and Google’s building are both guesses wearing the costume of strategy. In a year, one will look obvious and wise, and we will tell ourselves we could see it coming. We couldn’t. Hindsight manufactures a clarity that no one had at the time.

From any single seat - the chief executive’s, the investor’s, the worker’s, yours - you can see your own move clearly and the world’s answer not at all. The most you can honestly know is whether you kept a costly choice open when the facts were still moving, or spent that freedom before you had to. Everything after that belongs to a future none of us has read.

03 · Lab · your turn

Build or Wait

Rehearse committing early versus keeping a costly, irreversible choice open as the facts arrive.

04 · Hope · carry this

No one can see which way this breaks, and that turns out to be a quiet strength: while some firms build and others wait, whichever future arrives, someone will have been ready for it. People have always chosen well inside a fog they could not lift, and this moment is no different.