Meta Raises 2026 AI Capex to $145B — And the Stock Drops 7%

Meta reported Q1 2026 results on April 29 with revenue of $56.3 billion (up 33% year-over-year) and a quietly buried guidance change: full-year capital expenditures were raised to **$ 125–145 billion**, up from a prior range of $115–135 billion. The stock fell roughly 6–7% in after-hours trading. The earnings were not the problem. The capex was.

This is the third upward revision Meta has issued for 2026 capex in twelve months. At the new midpoint of $135 billion, Meta will spend almost exactly 2x what it spent in 2025 on AI infrastructure — and the market’s question is no longer whether the spending is real, it is whether anything on the revenue side will catch up.

The Spending Decomposition

CFO Susan Li attributed the revision to two factors:

Higher component pricing — primarily NVIDIA Blackwell-class GPUs, where supply remains tight and pricing has not softened despite a year of buildout
Additional data center costs to support future-year capacity — meaning Meta is now pre-funding capacity for 2027 and beyond

Q1 alone showed $19.84 billion in capex against a$ 27.57 billion consensus — meaning the run rate is actually below the new annual guidance. The implication is that the back half of the year ramps hard. That is the period where data centers Meta has been signing leases on for the past 18 months come online and start pulling power.

Total Q1 expenses ballooned 35% year-over-year to $33.4 billion, with infrastructure and compensation being the two drivers. Headcount is rising too — Meta’s AI organization under Alexandr Wang’s Superintelligence Labs has been hiring aggressively since the restructuring announcement, and the cost shows up here.

Why the Stock Reacted

Meta beat on revenue and beat on earnings. The 6-7% drop is therefore not a result quarter — it is a guidance quarter. Investors are repricing one specific question: at what point does AI capex stop being treated as “buying future revenue” and start being treated as “burning cash”?

The bear case is straightforward. Meta’s stated rationale for the spend is that AI makes ad targeting better, which compounds Reels and Stories engagement, which compounds ad revenue. The numbers are plausible — recommendation quality demonstrably moves the needle on time-on-platform — but the question is the multiplier. Does $20 billion in incremental compute drive$ 5 billion in incremental ad revenue, or $50 billion? The honest answer is nobody knows, and the market is no longer willing to extend the benefit of the doubt indefinitely.

The bull case is that this is what the front edge of the next platform shift looks like. The companies that overbuilt for the cloud in 2008-2012 were the ones that owned cloud by 2020. The companies that overbuild AI infrastructure in 2025-2027 are the ones that own AI by 2030. Zuckerberg has been explicit that he would rather be early and over-spend than late and under-built.

Which case is right depends on whether Muse Spark and successors produce the user-facing value Meta is betting on — and on that front, the field has gotten more crowded fast.

How This Compares Across Big Tech

The Q1 2026 earnings cycle has now produced a clear picture of the industry-wide capex spree:

Meta: $125–145B for 2026
Microsoft: ~$100B+ (driven by Azure and OpenAI partnership infrastructure)
Alphabet: $95B+ (now ramping with TPU v8 deployment)
Amazon: $100B+ (AWS, Trainium, Anthropic capacity commitments)

The combined Q1 capex across the four hyperscalers is in the neighborhood of $700 billion for 2026 — a number that did not exist as a category two years ago. This is the scale at which AI infrastructure has stopped being a line item and started being a macro phenomenon. Power grid planners, semiconductor supply chains, and commercial real estate markets are all being repriced around it.

The flip side is concentration risk. NVIDIA’s revenue mix is so dependent on this cohort that any softening — a Meta delay, a Microsoft pause, an Alphabet shift toward in-house silicon — would compress NVIDIA’s growth rate by tens of billions overnight. The whole bull thesis on AI infrastructure rests on these four companies continuing to underwrite the buildout at this pace through 2027.

The Anthropic Comparison

It is worth contrasting Meta’s spending against Anthropic’s, since Anthropic just secured 3.5 gigawatts of compute from Google and Broadcom. Anthropic is at a $30 billion revenue run rate, growing fast, and is funding multi-gigawatt deployments through a combination of equity, debt, and revenue. Meta is at hundreds of billions in revenue, spending more in absolute terms, but the ROI uncertainty is larger because the revenue lift from AI is indirect — better ads, not paid product.

The two strategies have different risk profiles. Anthropic’s spend is matched to a customer that exists. Meta’s spend is matched to a customer that will exist only if the recommendation flywheel works as projected. Both are reasonable bets. They are not the same bet.

What This Means

For builders, the relevant takeaway is that compute supply is going to remain expensive and constrained for the foreseeable future. Pricing on frontier model APIs is not going down on the back of a sudden glut — it is going down because of Chinese open-weight pressure and architectural efficiency gains, not because GPUs got cheaper.

For investors, the framing has shifted. The first phase of the AI capex bubble was “will they actually spend?” — answered with a clear yes. The second phase is now “will the spending pay back, and on what timeline?” — and the market has decided it wants quarterly evidence, not annual narratives.

Meta will get the next chance to make that case in July. Until then, every NVIDIA earnings beat is a partial proxy.