Meta’s second quarter reads like two companies sharing one income statement. The first grew ad revenue 27% year over year to $59.4 billion, on total revenue of $60.4 billion, up 28% — a growth rate most businesses at that scale would frame a keynote around. The second grew expenses 55%, to $42 billion, and the itemisation explains why nobody at Meta is framing keynotes around anything right now.
The cost drivers stack up quickly: $1.2 billion in severance following roughly 8,000 layoffs in May; $2.4 billion in legal costs, including youth-related trials that CFO Susan Li warned “may ultimately result in a material loss”; and AI infrastructure spending escalating fast enough that Meta is heading toward debt financing and a one-gigawatt data centre in El Paso built with BlackRock. When a company that prints $59 billion of quarterly ad revenue starts borrowing to fund compute, the scale of the AI bet has left the realm of discretionary spending.
But the figure publishers should actually mark sits in neither the revenue line nor the cost line. In early tests, generative models in Meta’s ads retrieval system lifted clicks 8.3% and conversions 15.7%.
The numbers in this piece
01The number that reprices everyone else
Treat those percentages with appropriate caution — they are Meta’s own early-test figures, disclosed on Meta’s own earnings call, which makes them claims rather than audited results. But take the direction seriously, because the mechanism is straightforward and unforgiving: every point of provable performance a walled garden adds raises the bar that an open-web CPM has to clear to stay in the plan.
The uncomfortable arithmetic works like this. A buyer allocating between Meta and the open web is comparing a channel that hands them a measured conversion lift against a channel that, too often, offers reach, context and assertion. When the measured side improves 15.7% on conversions, the unmeasured side does not stay still in relative terms — it gets 15.7% worse, without changing anything about its actual inventory. “Quality environment” was already a strained argument in performance-budget conversations; against a platform converting generative AI into quarterly-disclosed lift figures, it stops winning arguments on its own.
The expense side is not irrelevant to publishers either, but its relevance is indirect. A Meta carrying 55% cost growth, material legal risk and debt-financed compute needs its ads machine to keep over-delivering — which means the pressure to convert AI investment into demonstrable ad performance is structural, not experimental. The 8.3% and 15.7% are the first numbers from that program, not the last.
02Why this matters for publishers
| The competitive currency is now measured outcomes, not audience narratives | Meta is turning AI investment into performance claims with numbers attached. Open-web sellers answering with reach curves and adjacency language are competing in a currency buyers are ceasing to accept. |
|---|---|
| Performance budgets will consolidate before brand budgets do | The advertisers most sensitive to conversion lift are the first to move spend toward whichever channel proves it. Publishers heavy in performance-adjacent demand — commerce content, lead generation, direct response — feel this soonest. |
| Meta's platform-reported lift figures will be quoted at you | Expect "Meta gives me 15.7% conversion lift" in negotiations this quarter, deployed as if it were an audited cross-channel benchmark. It is neither — it is a self-reported early test — and knowing that distinction is a negotiating asset. |
| The cost drama is a distraction | Severance, lawsuits and gigawatt data centres are Meta's problems, and Meta can afford them. The performance line is the part of this earnings report that lands on your rate card. |
Meta's quarter is being covered as a cost story, and the costs are genuinely spectacular.
03What publishers should do
04The bottom line
Meta’s quarter is being covered as a cost story, and the costs are genuinely spectacular. But the line that reaches into publisher P&Ls is the smallest number in the report: a mid-teens conversion lift from generative models in ads retrieval. The platforms are converting AI capex into performance claims and performance claims into quarterly disclosures, and the bar for open-web spend rises with every one of them. The open web answering with reach and adjectives is how the gap widens; answering with instrumented, provable outcomes is the only version of this contest publishers can win.