How Nvidia's record earnings show up in your CPMs

When Nvidia earnings reset the AI-chip-demand conversation, the useful advertising question is not whether GPUs are important. They are. The useful question is which reported platform line item has already moved. In Meta’s Q2 2026 report, the checkable number is average price per ad: up 12%, alongside ad revenue of $59.4 billion, impressions up 14%, capex of $31.1 billion, operating cash flow of $31.86 billion, and free cash flow of just $784 million. That means capex consumed roughly 98% of operating cash flow in the quarter, while the auction price line was already higher.[1]

That is the cleaner bridge from Nvidia’s record chip demand to advertising platforms: not a one-hop claim that “Nvidia raised your CPMs,” but a funding chain. Platforms spend heavily on AI infrastructure. Their cash-flow cushion narrows. The ad business remains the monetization engine. Then buyers see the only auction number the platform actually reports: price per ad. The boundary still matters, and the site’s claim-check on Nvidia and ad costs has the right verdict: indirect, lagged, and partially offset.

The Meta bridge is the one worth carrying into a budget meeting

Meta is the sharpest case because there is no large cloud revenue line to make the infrastructure story harder to read. Alphabet can point to Google Cloud. Amazon can point to AWS. Microsoft can point to Azure. Meta’s AI spending is funded, overwhelmingly, by an advertising business sitting in plain sight.

The reported Q2 2026 sequence is uncomfortable for a media plan built on last quarter’s efficiency: ad revenue rose faster than impressions, the average price per ad rose double digits, and almost all operating cash flow was absorbed by capital expenditure.[1] That does not prove that a specific Nvidia GPU purchase caused a specific CPM increase. It does show that Meta’s ad platform is monetizing through a period when AI infrastructure spending is using nearly the whole operating-cash-flow envelope.

The free-cash-flow detail is easy to misuse. Meta’s free cash flow fell roughly 91% year over year, while net income fell 14%; those are different movements with different causes.[2] The free-cash-flow compression is the capex story. The net income decline reflects other expense dynamics as well, including legal and severance items in the reconciliation cited by DigitalApplied.[2] For auction planning, free cash flow is the more direct signal of how much infrastructure spending is eating into cash generation. Net income is not a substitute.

Average price per ad is not the same thing as your account-level CPM. It blends format, geography, placement, campaign mix, advertiser density, and Meta’s own auction mechanics. Still, it is a company-reported price line, not a vendor case study. If a paid-social forecast has to explain why Meta CPA assumptions may not repeat, that 12% reported price increase belongs near the top of the page.

Where Nvidia belongs in the read

Nvidia is the upstream pressure source, not the auction ledger. Its verified Q1 FY2027 record showed revenue of $81.6 billion, up 85% year over year, and Data Center revenue of $75.2 billion, up 92% year over year. Nvidia also guided Q2 FY2027 revenue to $91 billion, plus or minus 2%, with no China Data Center compute revenue assumed in that outlook.[3]

As of August 26, 2026, Nvidia’s Q2 FY2027 result should be checked against the site’s Nvidia earnings tracker. Pre-release consensus context was roughly $92 billion in revenue and about $2.09 in EPS, but those figures are analyst expectations, not the verified result.[4][5]

That distinction changes the planning conversation. Nvidia revenue tells you that demand for accelerated compute is real and large. It does not tell you how much of Meta’s Q4 CPM will be attributable to GPUs, model training, Advantage+ ranking improvements, advertiser competition, creative fatigue, holiday demand, or mix shift. The auditable advertising evidence starts when the platform reports capex, cash flow, ad revenue, impressions, and price per ad.

The wider capex race supports the pressure story, but it does not replace Meta’s evidence

The broader spending race is not a Meta-only phenomenon. CNBC reported in February 2026 that Google, Microsoft, Meta, and Amazon were expected to spend roughly $700 billion on AI-related capex in 2026, while also noting Amazon’s roughly $200 billion 2026 capex guidance and analyst projections for negative free cash flow at Amazon from Morgan Stanley and Bank of America.[6] Futurum separately framed 2026 AI capex as a $660 billion to $690 billion infrastructure sprint across five companies including Oracle.[7]

Those are not interchangeable numbers. They differ by company set, date, and definition. They are useful for scale, not for a direct CPM formula. If a forecast says “hyperscalers are spending hundreds of billions,” it should still separate reported platform results from analyst-estimated capex aggregates.

Alphabet is a good comparison because it shows that ad platforms with cloud businesses are also raising infrastructure spending. In Q2 2026, Alphabet capex was reported at $44.9 billion, roughly twice the prior-year level, and its FY2026 capex guide was raised to $195 billion to $205 billion; Search and other ads revenue was $63.3 billion, up 17%.[8] That supports the broader AI-infrastructure race, but it is a less clean auction-cost read than Meta because Alphabet has multiple monetization paths for the same infrastructure base.

The same caution applies to Microsoft and Amazon. Their infrastructure spending can show up through cloud pricing, internal workload economics, AI subscriptions, retail media, search ads, and marketplace tools. For ad buyers, that means the capex race matters, but Meta’s reported price-per-ad line remains the more usable H2 2026 signal.

How the cost can pass through without appearing as a GPU surcharge

Advertising platforms do not need to add an “AI chip fee” to campaigns for infrastructure cost to reach buyers. The pass-through can be quieter: auction design, ranking systems, automated campaign defaults, measurement products, principal-media packages, minimum commitments, or a higher tolerance for price inflation when advertiser demand is strong.

One concrete pass-through clue came from Digiday’s July 2026 reporting on principal media deals. In one reported buyer example, a CMO was offered absorbed AI costs in exchange for 70% of the budget running through principal inventory.[9] That is not industry-wide proof. It is a single reported commercial structure. But it is a useful reminder that AI infrastructure costs can be routed into media allocation terms rather than itemized as token billing.

The same logic can sit inside platform buying products. A platform can improve model quality and claim better conversion matching, creative generation, or campaign optimization. Some of that may be real. Some of it may offset higher media prices if conversion rates rise enough. But a buyer still has to defend the media plan against the bill that actually clears: spend, CPM, CPC, CPA, ROAS, and margin after fees. Vendor lift claims belong in the test log, not in the base forecast without verification.

How to re-forecast H2 2026

For H2 2026, the practical move is to stop treating AI capex as a narrative risk and start treating platform-reported price movement as a forecast input. Meta’s 12% average-price-per-ad increase is not a command to raise every CPM assumption by 12%. It is the anchor that keeps the forecast from pretending the auction was flat.

A defensible re-forecast should separate three layers that often get blended after earnings week:

• Platform-reported price pressure: Meta’s average price per ad rose 12% in Q2 2026. Carry that into the CPM discussion before debating AI performance claims.[1]
• Cash-flow pressure: Meta’s Q2 capex of $31.1 billion consumed nearly all of its $31.86 billion operating cash flow, and FY2026 capex guidance sits at $130 billion to $145 billion.[1]
• Performance offset: better ranking, creative, and measurement may improve conversion efficiency, but that has to be proven in account data rather than imported from a platform announcement.

For a Meta-heavy plan, that means a flat H2 CPM assumption now needs an explicit defense. Maybe a category has weaker competition. Maybe creative refreshes are outperforming. Maybe conversion rate gains are absorbing higher auction prices. Those are valid account-level arguments. “AI makes ads more efficient” is not enough on its own when Meta has already reported higher average price per ad and a much tighter post-capex cash position.

For Google and Amazon plans, the adjustment should be more cautious. The infrastructure spending is real, but the disclosure path is less direct. Look for reported ad revenue growth, segment-level capex where available, cloud cross-subsidy effects, and auction-price indicators inside your own accounts. The stronger the platform’s non-ad monetization path, the less clean the bridge from chip spending to ad prices.

For principal-media or managed-service deals, the question is different: not only “did CPM rise?” but “where did the AI cost get embedded?” It may sit in inventory commitments, margin spreads, optimization fees, measurement bundles, or budget-allocation requirements. That is where the Digiday example is useful: not as a market average, but as a reminder to inspect the commercial wrapper around the media.

What not to claim after Nvidia earnings

Do not write a forecast note that says Nvidia revenue rose, therefore Meta CPMs must rise by a fixed amount. The available evidence does not support that precision. The chain has delays, offsets, and competing explanations. Seasonal demand, advertiser density, category mix, placement mix, model quality, conversion-rate changes, and platform pricing choices all sit between chip demand and the number in an ad account.

Also do not let net income do the job of free cash flow. If the question is whether infrastructure spending is tightening the cash model behind the ad platform, operating cash flow, capex, and free cash flow are the relevant bridge. Net income can move for reasons that do not describe infrastructure funding pressure.

And do not treat price per ad as a perfect CPM proxy. It is a platform-level average. A retail advertiser in Advantage+ Shopping, a B2B lead-gen account, and an app-install buyer can all experience different auction pressure. The platform has disclosed a double-digit increase in the price line while its AI capex is absorbing almost all operating cash flow; that does not mean every account gets Meta’s 12%.

The forecasting rule is simple enough to defend: anchor H2 2026 assumptions on platform-reported capex, free cash flow, and price per ad; use Nvidia’s record demand as upstream context; and label any direct chip-capex-to-CPM bridge as analyst reasoning, not proven causation.

References

1. Meta Reports Second Quarter 2026 Results — Meta Investor Relations, 2026
2. Meta Q2 2026 Earnings: Ad Strength, Capex Selloff — DigitalApplied, 2026
3. NVIDIA Announces Financial Results for First Quarter Fiscal 2027 — NVIDIA Newsroom, 2026
4. Nvidia earnings live updates and commentary, August 2026 — Kiplinger, 2026
5. NVIDIA Earnings Date and Reports 2026-8-26 — MarketBeat, 2026
6. Google, Microsoft, Meta, Amazon AI cash — CNBC, 2026-02-06
7. AI Capex 2026: The $690B Infrastructure Sprint — Futurum Group, 2026
8. Alphabet Q2 2026 Earnings: Search Ads, AI Overviews — DigitalApplied, 2026
9. How AI costs are quietly reshaping principal media deals — Digiday, 2026-07