“The smartest people, theoretically, are always in the room. But you always have to make a business judgment as to where and when you spend the incremental dollar and how you do it. And I think if you’re under spending, you’re going to find yourself playing defense very aggressively.” That observation, from CNBC analyst and investor Josh Baruch, cuts to the center of the most important capital allocation debate in markets right now: in the AI era, does talent still determine winners, or does the decision of where to deploy capital?
The proof is already in the financials of the two companies most central to the AI infrastructure build-out.
Why Capex Discipline Now Determines Competitive Position
Baruch’s point is grounded in a real pattern: companies that sat on their hands during major technology transitions did not just fall behind, they were forced to play catch-up at far higher cost. The AI infrastructure cycle follows the same logic. The question for investors is not whether to spend, but whether the spending is directed at the right things at the right time.
Meta is the clearest case study. As Baruch noted, Meta “got tattooed for capital spending” around 2022 when its Reality Labs ambitions drew skepticism and the stock fell sharply. Management reined in spending, refocused, and then reinvested aggressively in AI-driven advertising and recommendation systems. The result was a business transformation that rewarded shareholders who stayed patient.
That pattern is now repeating at a far larger scale. Meta’s 2026 capital expenditure guidance is $115 to $135 billion, up from $69.7 billion in full-year 2025, itself an 87% increase over the prior year. The stated purpose is to support Meta Superintelligence Labs and core business infrastructure. The size of this commitment raises a legitimate question: is this smart capex, or the undisciplined spending Baruch warns against?
The operating results provide an early answer. Revenue for Q4 2025 reached $59.9 billion, up 24% year over year, with ad impressions growing 18% and average price per ad up 6%. The underlying ad business is not just growing, it is accelerating, and management attributes meaningful gains directly to AI-driven improvements in recommendation systems and ad targeting.
The Real Cost of Smart Capex: Margin Compression That May Be Temporary
Baruch’s framework does not say spend freely. It says under-spending is the greater risk. Meta’s financials show the genuine short-term cost of aggressive infrastructure investment. The short-term cost of this investment is visible in the income statement. Operating margin compressed from 48% to 41% year over year in Q4 2025, as total costs grew 40% against revenue growth of 24%. Management has signaled it expects this to reverse as infrastructure investments mature and generate returns.
For an investor evaluating whether this is smart or reckless spending, the key signal is what management expects next. Despite the infrastructure ramp, Meta’s management projects that 2026 operating income will exceed 2025 levels. That forecast implies the capex is building capacity that generates returns faster than it erodes margins, which is exactly what smart capital allocation looks like in practice.
Zuckerberg’s framing of the productivity angle is worth noting. “We’re starting to see projects that used to require big teams now be accomplished by a single very talented person,” he said on the Q4 earnings call. The company reported a 30% increase in output per engineer since the beginning of 2025, with power users of AI coding tools showing an 80% year-over-year output increase. This is the mechanism Baruch is describing: capital invested in AI infrastructure multiplies the productivity of the talent already in the room.
NVIDIA: What Happens When You Are on the Right Side of Others’ Capex Decisions
If Meta illustrates the investor thesis from the spending side, NVIDIA illustrates it from the supply side. Every dollar of AI infrastructure capex committed by Meta, Amazon, Google, and Microsoft flows through NVIDIA’s data center business. NVIDIA’s data center revenue for Q4 FY2026 reached $62.3 billion, up 75% year over year, with data center networking revenue up 263%.
The business model contrast with Meta is instructive. NVIDIA generated $34.9 billion in free cash flow in Q4 alone, with FY2026 operating cash flow of $102.7 billion. NVIDIA’s own capital expenditures are a fraction of what its customers spend, because the company designs chips rather than building the factories that run them. The customers absorb the capex risk; NVIDIA captures the margin.
Jensen Huang described the current environment on NVIDIA’s Q4 earnings call: “Enterprise adoption of agents is skyrocketing. Our customers are racing to invest in AI compute, the factories powering the AI industrial revolution and their future growth.” Those commitments are substantial: NVIDIA holds $95.2 billion in total supply commitments and $27 billion in multi-year cloud service agreements. That locked-in future demand gives NVIDIA unusual revenue visibility in a business that could otherwise be cyclical. The scale of these forward commitments signals that hyperscalers are not hedging their AI bets; they are doubling down.
Top Gaining Stocks
Top Losing Stocks