META · Q2 2026 · reported 2026-07-29 · revenue $60.80bn
Assessment
Q2 2026 keeps Meta’s AI money in spend and adds a named vendor bill: the 10-Q now lists third-party AI token costs among the causes of higher marketing and sales and research and development, with no amount (ledger ballpark ). Capital expenditures were , nearly all of operating cash flow of , leaving free cash flow of ; the company issued of notes and the CFO says outside capital supplements cash flow, so the build is no longer paid from advertising cash flow alone. Leases not yet commenced reached and commitments .
A second off-balance-sheet campus followed after the quarter, as a subsequent event: an El Paso venture with BlackRock, with residual value exposure of about expected at closing. Depreciation ( ballpark) and third-party cloud () are named causes of cost growth with no AI amount from management, sized because the filing ties the increase in infrastructure investment to AI. AI talent, named inside technical hires, and interest, tied to long-horizon initiatives especially AI infrastructure, have nothing that separates or bounds AI's part and are left unsized, as is the capital level, named for AI efforts and the core business together.
Revenue credited to AI still comes through the ads system: management gives larger measured lifts from LLM and GEM ranking models, more conversions on Facebook among them, but no dollar part, and the filing explains the price and impression growth by several causes together, so both ad channels are read described, unsized and relabelled. New priced products (Meta One and a model API) launched at about the time of the call with no revenue, business agents have pricing in outline only, and the advertiser support assistant goes unmentioned. The severance of for the May reduction is not attributed to AI.
Sized channels against the income statement, Q2 2026
6 of 17 channels sized
Each blue mark is one channel's dollars for the quarter; a bar is the range of an estimate. Grey marks are the company's reported lines. The distance between them is the point: how large the AI channel is next to the line it sits in.
Revenuereported line
$60.80bn
Total costs and expensesreported line
$42.03bn
Research and developmentreported line
$21.66bn
Cost of revenuereported line
$11.33bn
General and administrativereported line
$5.61bn
Marketing and salesreported line
$3.43bn
Third-party cloud capacity bought for AIspend · our inference
Depreciation of the AI infrastructure buildspend · our inference
Off-balance-sheet data center ventures with residual value guarantees (RVG)spend · our inference
Third-party AI token costsspend · our inference
Engineering and product development work done with AI tools and agentscost displaced · our inference
AI glasses salesrevenue in · our inference
$10mn$100mn$1bn$10bn$100bn
AI channel, dollars for the quarter low to high of an estimate reported lineLog scale: each gridline is ten times the one before.
New money and old money, Q2 2026
4 new10 expanded3 relabelled
Each channel is tagged once for whether its money existed before language models, from the company's annual report and call at the start of the period. A bar splits one flow's sized dollars by that tag. The incremental total is the part that would not be there without the models: a new channel counts in full, an expanded one only for what AI added, a relabelled one at zero.
Flow, sized total
Split by novelty
Incremental total
Paid for AI$1.96bn to $7.36bn sized
new $158mn to $1.32bnexpanded $1.80bn to $6.05bn
Incremental total $1.46bn to $7.36bnpoint $2.27bn$1.86bn in 1 channel has no traced baseline
Cost displaced by AI$29mn to $897mn sized
expanded $29mn to $897mn
Incremental total $29mn to $897mnpoint $242mn
Revenue arriving through AI$129mn to $302mn sized
new not sizedexpanded not sizedrelabelled $129mn to $302mn
Incremental total $0
The sized total counts every channel the company credits to AI, including relabelled money that existed before language models and the ledger's own estimates for it. The incremental total counts relabelled channels at zero. A flow is split by layer where its dollars sit at more than one: end use, compute sold to builders, and hardware. The same dollar can be a buyer's spend, a cloud's revenue and a chipmaker's revenue, so the layers are never added together.
Paid for AI7 channels · $1.96bn to $7.36bn sized · $1.46bn to $7.36bn incremental · 3 not sized
other
Capital expenditures on servers, data centers and network infrastructure for the AI build, including principal payments on finance leases
Not sized
The filing gives the capital spending as support for its AI efforts and core business together (claim c2); the statement that infrastructure investments increased in connection with AI initiatives speaks to the increase, not the level, and no phrase bounds AI's share of the level. Nothing separates AI's part (the methodology rule for capital named together with AI). Capital expenditures, , are quoted as the ceiling.
described, no sizedisclosure: quantified· motive: exploratory· before LLMs: expanded
Capital expenditures including finance lease principal were , about of operating cash flow of , leaving free cash flow of ; the company issued of notes in May, and the CFO says outside capital supplements cash flow (claim c5), so funding is mixed. The year is now to . Leases not yet commenced rose to , with more signed in July, and sits in escrow under infrastructure purchase agreements. The level is named for AI efforts and the core business together and nothing separates AI's part, so the channel is left unsized; capital expenditures are the ceiling. It is capital and would be left out of totals in any case.
Evidence: 13 quotes, 11 figures, 2 confounds, 2 from before coverage
Spending on servers (GPUs from outside chip suppliers, the company’s own custom silicon developed with Broadcom, AMD chips), data centers and network infrastructure, which management reports including principal payments on finance leases and says supports its AI efforts and core business. Neither the filing nor the call gives an AI share; at this company ranking and recommendation models for the core ads business are also AI workloads, so the AI share is read wide. Capitalized: traced and left out of flow totals; the income-statement cost is read on the depreciation, cloud and interest channels. The accelerator purchases are other ledger companies’ revenue (NVDA); the same dollar is recorded there by that company’s own filing, not netted here. Layer end-use: the filings put the build first to the company’s own products and model training; from Q2 it also serves the paid model API, which is the compute layer and is read on its own channel, unsized.
Why this motive
The tells still conflict: the CEO says a substantial amount of the compute trains models for the lab (claim c3), the rest serves products including the measured ads gains; the less durable reading is taken.
Before LLMs: expanded
Capital spending on servers, data centers and network infrastructure predates the build: in fiscal 2024 and in its first quarter, with the anchor plan for the next year already tied to the core business and generative AI. The baseline for the incremental total is that quarter, : the whole of the anchor quarter’s capital spending, not an AI part of it, set against an AI share of the current quarter’s. The size is the whole of an activity that existed before.
“In particular, we expect our AI initiatives will require increased investment in infrastructure and headcount.”
“We anticipate making capital expenditures of approximately $60 billion to $65 billion in 2025 to support our core business and generative AI efforts.”
Capital expenditures including principal payments on finance leases · 2026-CQ2
Capital expenditures planned for 2026, low end of range · 2026-01-01..2026-12-31
Capital expenditures planned for 2026, high end of range · 2026-01-01..2026-12-31
Cash flow from operating activities · 2026-CQ2
Free cash flow · 2026-CQ2
Capital expenditures as a share of operating cash flow · 2026-CQ2
Senior unsecured notes issued in May 2026 · 2026-CQ2
Long-term debt at quarter end · as-of 2026-06-30
Operating and finance leases not yet commenced, data centers, colocations and network infrastructure · as-of 2026-06-30
Data center leases entered in July 2026, commencing in 2027 and 2028 · as-of 2026-07-31
Money market funds held in escrow under multi-year infrastructure purchase agreements · as-of 2026-06-30
What else could explain it
line composition: The same servers run the ads and recommendation models and the frontier models; no AI share is given.
other: Part of the year’s spending is higher component pricing rather than more capacity.
Quotes
“Capital expenditures, including principal payments on finance leases, were $31.1 billion, driven by investments in servers, data centers, and network infrastructure. Free cash flow was $784 million.”
“In terms of the different opportunities and how we think about the compute, overall, a substantial amount of the compute goes towards training models to be a leading lab, and I think that's an important investment.”
“The rest of it goes towards a set of different products and revenue opportunities, which spans from optimizing and improving our core business, to building new consumer products that we're releasing soon to the API, to the business agents work, to the developer tools work on the roadmap that I alluded to, and also the opportunity to sell compute directly.”
“In funding these infrastructure investments, the strength of our balance sheet gives us the ability to attract capital from a wide range of markets to supplement the cash flow generated by our business.”
“One is we are today and expect to be in the sort of foreseeable future demand-constrained. That really includes our core business too, where we still have numerous ROI-positive places that we would put compute toward if we had it.”
“In particular, we have significantly increased our infrastructure investments in connection with our AI initiatives, including third-party cloud capacity arrangements and investments in servers, data centers, and network infrastructure, and expect our investments to continue to increase.”
“Obviously, our strong operating cash flow certainly has put us in a position of strength as it pertains to funding our infrastructure build-out. We've also been evolving our capital structure in recent years to include a greater mix of debt as we work to bring down our cost of capital.”
“In connection with escrow requirements under certain multi-year infrastructure purchase agreements, $10.80 billion of money market funds was reclassified as restricted cash equivalents as of June 30, 2026.”
“In July 2026, we entered into additional data center leases with lease obligations of approximately $68 billion, which are expected to commence in 2027 and 2028, with lease terms of 18 to 20 years.”
“In addition, we have been making strategic investments in areas like our internal custom silicon effort, which will provide long-term strategic flexibility and supply chain leverage.”
“We currently anticipate that our available funds and cash flow from operations and financing activities will be sufficient to meet our operational cash needs and fund our cash commitments for investing and financing activities, including investments in infrastructure and AI initiatives, as well as any return of capital to stockholders over the next 12 months and thereafter for the foreseeable future.”
Q1 2026quantified · described, no size · exploratory
Q2 2026quantified · described, no size · exploratory
cost of-revenue
Depreciation of the AI infrastructure build
2.1% to 3.4% of the quarter’s revenue
Incremental total: counts in full.
Matched line moved : the whole line, not this channel.
our inferencedisclosure: direction only· motive: exploratory· before LLMs: expanded
Depreciation was against , of which servers and network assets ; the CFO again names it first among the causes of infrastructure cost growth (claim c10). The size, , is the ledger’s.
Evidence: 4 quotes, 3 figures, 2 confounds, 3 from before coverage
Depreciation of the servers, network assets and data centers bought for the AI build, most of it in cost of revenue and part in research and development as infrastructure cost. The filing reports depreciation for the whole company and names higher depreciation as a cause of infrastructure cost growth; the AI part is assumed. Data center operating costs (power and operations) are named beside it and are not sized here. Layer end-use: the filings put the build first to the company’s own products and model training; from Q2 it also serves the paid model API, which is the compute layer and is read on its own channel, unsized.
Why this motive
Carried with the capital channel (claims c10, c3).
Before LLMs: expanded
Depreciation was in fiscal 2024, about a quarter, of which servers and network assets ; the anchor already explained cost of revenue growth by data center and infrastructure costs, mostly depreciation. The size is the depreciation above that quarterly level, times an assumed AI share. The size is the change AI made, not the whole line.
“In particular, we expect our AI initiatives will require increased investment in infrastructure and headcount.”
“Cost of revenue in 2024 increased $4.20 billion, or 16%, compared to 2023. The increase was primarily due to higher operational expenses related to our data centers and technical infrastructure, mostly from higher depreciation expense.”
“The increases were primarily due to higher employee compensation, infrastructure expenses related to our data centers, technical infrastructure, and third-party cloud services, and third-party AI token costs.”
“The increases were primarily due to higher infrastructure expenses related to our data centers, technical infrastructure, and third-party cloud services.”
“In particular, we have significantly increased our infrastructure investments in connection with our AI initiatives, including third-party cloud capacity arrangements and investments in servers, data centers, and network infrastructure, and expect our investments to continue to increase.”
Q1 2026direction only · our inference · exploratory · $1.11bn to $1.76bn
Q2 2026direction only · our inference · exploratory · $1.31bn to $2.07bn
cost of-revenue
Third-party cloud capacity bought for AI
0.81% to 6.5% of the quarter’s revenue
Incremental total: counts at zero at the point and in full at the high end, with no traced baseline.
Matched line moved : the whole line, not this channel.
our inferencedisclosure: direction only· motive: exploratory· before LLMs: expanded
Third-party cloud is named in the cost of revenue and research and development explanations (claims c11, c12); commitments, mostly cloud capacity and servers, rose to . Asked why the company buys capacity while being offered a premium for its own, the CEO says there is not enough compute for the demand (claim c14). The size, , is the ledger’s level: an assumed prior-year cloud cost plus a share of the rise in Family of Apps other costs after legal charges, depreciation and the token estimate. CoreWeave (CRWV) names Meta as a significant customer in its own Q1 2026 10-Q; that commitment is cited in the estimate and is CoreWeave’s revenue on its own ledger.
Evidence: 8 quotes, 5 figures, 2 confounds, 5 from before coverage
Cloud capacity rented from outside providers under multi-year deals, which the filing lists among the infrastructure investments made in connection with the company’s AI initiatives and names as a cause of higher cost of revenue and research and development. Meta’s own filings do not name the providers. CoreWeave (CRWV, on this ledger) names Meta in its Q1 2026 10-Q as a significant customer under a March 2026 order form running to 2032, with the commitment it states cited in this channel’s estimate; those dollars are CoreWeave’s revenue on its own ledger and are not netted here. Non-cancelable commitments, mostly cloud capacity and servers, and a contingent obligation to purchase more cloud capacity are traced as metrics. Third-party AI token costs are a separate channel and are taken out of this channel’s base. Layer end-use: the filings put the build first to the company’s own products and model training; from Q2 it also serves the paid model API, which is the compute layer and is read on its own channel, unsized.
Why this motive
Capacity bought while the company is compute-constrained, largely for training (claims c3, c7).
Before LLMs: expanded
The fiscal 2024 annual report never mentions cloud capacity bought from third parties: infrastructure was the company’s own data centers, inside cost of revenue ( for the year) and research and development. By the 2025-CQ4 reference quarter cloud spend was a named cause of infrastructure cost growth (anchor claim meta-anchor-c28) and year-end commitments of were mostly third-party cloud capacity and servers (anchor claim meta-anchor-c30), so cloud predates coverage at a level no source gives. Renting compute could exist without LLMs and the line that holds it existed, so the tie-break gives expanded, sized as a level with no traced quarter before AI. The size is the whole of an activity that existed before.
“In particular, we expect our AI initiatives will require increased investment in infrastructure and headcount.”
“Cost of revenue in 2024 increased $4.20 billion, or 16%, compared to 2023. The increase was primarily due to higher operational expenses related to our data centers and technical infrastructure, mostly from higher depreciation expense.”
“We have $131.05 billion of non-cancelable contractual commitments as of December 31, 2025. These commitments are mostly related to third-party cloud capacity arrangements and our continued investments in servers and network infrastructure, data centers, and consumer hardware products in Reality Labs.”
Non-cancelable contractual commitments, mostly third-party cloud capacity, servers, network infrastructure and data centers · as-of 2026-06-30
Contingent obligations to purchase cloud capacity over five years · as-of 2026-06-30
Family of Apps other costs and expenses (infrastructure, professional services, partner arrangements, marketing, facilities, legal; negative: a cost) · 2026-CQ2
Family of Apps other costs and expenses, prior-year quarter (negative: a cost) · 2025-CQ2
Charges related to legal proceedings in the quarter · 2026-CQ2
Reported line it is matched to
Family of Apps other costs and expenses rose over the prior-year quarter (shown as costs).
2026-CQ2: 2025-CQ2:
What else could explain it
line composition: Family of Apps other costs also include data center operating costs, professional services and marketing; token costs are taken out at the ledger’s estimate.
one time item: The legal charges of are taken out of the base.
Quotes
“In particular, we have significantly increased our infrastructure investments in connection with our AI initiatives, including third-party cloud capacity arrangements and investments in servers, data centers, and network infrastructure, and expect our investments to continue to increase.”
“The increases were primarily due to higher employee compensation, infrastructure expenses related to our data centers, technical infrastructure, and third-party cloud services, and third-party AI token costs.”
“The increases were primarily due to higher infrastructure expenses related to our data centers, technical infrastructure, and third-party cloud services.”
“Look, the high-level observation is that there's just nowhere near enough compute for all the demand. That is why we see that basically, we are getting a large number of offers for the compute that we have, we have a lot of internal uses that we think are going to be quite valuable.”
“One is we are today and expect to be in the sort of foreseeable future demand-constrained. That really includes our core business too, where we still have numerous ROI-positive places that we would put compute toward if we had it.”
“In terms of the different opportunities and how we think about the compute, overall, a substantial amount of the compute goes towards training models to be a leading lab, and I think that's an important investment.”
“The increase in costs and expenses was primarily due to increases in employee compensation, including severance expenses; infrastructure expenses related to our data centers, technical infrastructure, and third-party cloud services; legal-related costs; and third-party AI token costs.”
Q1 2026direction only · our inference · exploratory · $381mn to $3.17bn
Q2 2026direction only · our inference · exploratory · $492mn to $3.98bn
vendor bill
Third-party AI token costs
0.26% to 2.2% of the quarter’s revenue
Incremental total: counts in full.
Matched line moved : the whole line, not this channel.
our inferencedisclosure: direction only· motive: exploratory· before LLMs: new
The 10-Q now names third-party AI token costs as a primary cause of the marketing and sales rise of and as a cause of the research and development rise of (claims c13, c11); the CFO lists them among the drivers of expense growth (claim c9). No amount, provider or use is given. The size, , is the ledger’s: shares of the two rises after an assumed part of the severance of , with the prior-year bill taken as negligible because the 2025-CQ4 reference quarter names no token cost.
Evidence: 4 quotes, 4 figures, 2 confounds
Payments for tokens from outside model providers, which the Q2 10-Q names as a primary cause of higher marketing and sales and as a cause of higher research and development; in Q1 the CEO said product teams had prototyped on other companies’ APIs. The providers and the uses are not named; a token bill could not exist without LLMs.
Why this motive
Named as a cause of cost growth with no product or result attributed to it (claims c9, c13).
Before LLMs: new
The anchor shows no payment to model providers: research and development () and marketing and sales () in fiscal 2024 were explained by compensation, infrastructure and restructuring. The 2025-CQ4 reference quarter’s call, release and annual report name no token cost either; the Q2 2026 10-Q is the first source to name one. A token bill is new money by the first test.
Figures
Marketing and sales · 2026-CQ2
Marketing and sales, prior-year quarter · 2025-CQ2
Research and development, increase over the prior-year quarter · 2026-CQ2
Severance expenses for the May 2026 headcount reduction, inside employee compensation · 2026-CQ2
Reported line it is matched to
Marketing and sales rose over the prior-year quarter.
2026-CQ2: 2025-CQ2:
What else could explain it
one time item: Severance for the May reduction is the other named cause of the marketing and sales rise; the filing does not split it by line, so it is netted out at assumed shares.
line composition: Research and development holds compensation, infrastructure and cloud services as well.
Quotes
“Year-over-year growth was primarily driven by increases in employee compensation, infrastructure costs, legal-related costs, and third-party AI token costs.”
“The increases were primarily due to higher employee compensation, infrastructure expenses related to our data centers, technical infrastructure, and third-party cloud services, and third-party AI token costs.”
“The increase in costs and expenses was primarily due to increases in employee compensation, including severance expenses; infrastructure expenses related to our data centers, technical infrastructure, and third-party cloud services; legal-related costs; and third-party AI token costs.”
Q2 2026direction only · our inference · exploratory · $158mn to $1.32bn
research
Compensation growth from technical hires, particularly AI talent
Not sized
AI talent is named inside a wider cause: compensation growth was driven by technical hires, particularly AI talent (claim c15), with no rate, share or ranking of AI's part, and the 10-Q explains the rise mainly by share-based compensation. Nothing separates AI's part (the methodology rule for AI named beside another cause, which holds at the upstream companies). The rise in employee compensation, , is quoted as the ceiling.
Matched line moved : the whole line, not this channel.
described, no sizedisclosure: direction only· motive: exploratory· before LLMs: expanded
Excluding severance of , the CFO again attributes compensation growth to technical hires, particularly AI talent (claim c15); employee compensation was against , up excluding severance. AI talent is named inside a wider cause, technical hires, with no rate or share of its own, so the channel is left unsized; the rise in compensation is the ceiling.
Evidence: 4 quotes, 5 figures, 3 confounds, 3 from before coverage
The part of employee compensation growth that the CFO attributes to technical hires added over the past year, particularly AI talent, and that the filing describes as investment in headcount, including specialized technical personnel, to develop and train AI models. Most of it sits in research and development, where the rise is mainly share-based compensation. Severance from the May 2026 headcount reduction is taken out of the base.
Why this motive
Carried: hiring to build models and a lab whose products are not yet monetized (claim c15).
Before LLMs: expanded
At the anchor headcount was and growth in engineering and technical headcount already drove research and development compensation, with the filing expecting AI to require more headcount. The size is a share of the compensation rise over the prior year, so it is sized as an increment. The size is the change AI made, not the whole line.
“In particular, we expect our AI initiatives will require increased investment in infrastructure and headcount.”
“The higher employee compensation was mainly from a 13% growth in employee headcount from 2023 to 2024 in engineering and other technical functions supporting our continued investment in our family of products and Reality Labs.”
Employee compensation, total (negative: a cost) · 2026-CQ2
Employee compensation, total, prior-year quarter (negative: a cost) · 2025-CQ2
Severance expenses for the May 2026 headcount reduction, inside employee compensation · 2026-CQ2
Headcount at quarter end, including employees impacted by the reduction · as-of 2026-06-30
Employee compensation, increase over the prior-year quarter, excluding severance · 2026-CQ2
Reported line it is matched to
Employee compensation rose over the prior-year quarter, excluding severance.
2026-CQ2: 2025-CQ2:
What else could explain it
mix shift: Share-based compensation is the main named cause of the rise.
one time item: Severance for the May reduction is taken out of the base.
other: Technical hires outside AI, in monetization and infrastructure, are part of the same named cause; the AI part is not split.
Quotes
“Excluding the previously mentioned severance expense, growth in employee compensation was driven by technical hires we've added over the past year, particularly AI talent.”
“The higher employee compensation was mainly from increases in share-based compensation expense and severance expenses during the three and six months ended June 30, 2026.”
“The increases were primarily due to higher employee compensation, infrastructure expenses related to our data centers, technical infrastructure, and third-party cloud services, and third-party AI token costs.”
“We ended Q2 with over 75,000 employees down 3% from Q1. This total includes approximately 8,000 employees impacted by the May 2026 headcount reduction.”
Q1 2026direction only · described, no size · exploratory
Q2 2026direction only · described, no size · exploratory
other
Off-balance-sheet data center ventures with residual value guarantees (RVG)
0.72% to 1.1% of the quarter’s revenue; capital spending on the build, not added into totals (its depreciation is)
our inferencedisclosure: quantified· motive: exploratory· before LLMs: expanded
The Louisiana investment’s carrying value rose to , with maximum exposure of . The size, , is the ledger’s reading of the quarter’s contribution from the change in carrying value (); the filing attributes no dollar to AI, so the strength is inferred, and it is capital, left out of totals. As a subsequent event, in July 2026 the company agreed exclusivity on an El Paso campus in which it would own ; at closing, expected in Q3, it expects to contribute about of held-for-sale assets, receive about back and provide residual value guarantees with a maximum aggregate exposure of about (claims c19, c62), and the CEO names BlackRock as the partner (claim c20). Those amounts belong to the quarter of closing. Funding is mixed: in Louisiana the parties fund pro rata shares and the company holds a minority interest; the El Paso terms beyond the interest await definitive agreements.
Evidence: 5 quotes, 7 figures, 1 confound, 1 from before coverage
Data center campuses co-developed through ventures the company does not consolidate: a Louisiana campus entered in October 2025, which the filing says is to meet future infrastructure capacity needs as AI markets and technologies develop, and from July 2026 an exclusivity agreement for an El Paso campus that management announced with BlackRock. The company owns a minority membership interest, funds its pro rata share of the development, will lease the buildings and provides residual value guarantees (RVG) on them. The parts on the other side are the partner, which funds its own pro rata share of the development costs, and the venture as landlord. The quarter’s money is the company’s contribution, read from the change in the equity investment’s carrying value; it is capital, traced and left out of flow totals, and the lease payments begin only in 2029. Layer end-use: the filings put the build first to the company’s own products and model training; from Q2 it also serves the paid model API, which is the compute layer and is read on its own channel, unsized.
Why this motive
Capacity for future needs as AI markets develop, structured for flexibility (claims c18, c21).
Before LLMs: expanded
At the anchor data center capacity was already contracted ahead of use: leases not yet commenced, mostly data centers, were . No venture or residual value guarantee appears in the fiscal 2024 annual report. A financing structure for capacity the anchor already contracted is expanded, sized as a level with no traced quarter before AI. The size is the whole of an activity that existed before.
“As of December 31, 2024, we have additional operating and finance leases, that have not yet commenced, with total lease obligations of approximately $34.12 billion, mostly for data centers, network infrastructure, and colocations.”
Change in the carrying value of the venture investment in the quarter (contributions net of any share of results) · 2026-CQ2
Carrying value of the equity investment in the Louisiana data center venture · as-of 2026-06-30
Maximum exposure to loss related to the Louisiana venture · as-of 2026-06-30
Membership interest to be held in the El Paso venture · as-of 2026-07-31
Held-for-sale data center assets the company expects to contribute to the El Paso venture at closing (July 2026 agreement, a subsequent event) · as-of 2026-07-31
One-time distribution the company expects to receive from the El Paso venture at closing (July 2026 agreement, a subsequent event) · as-of 2026-07-31
Maximum residual value exposure (RVG) the company expects to provide on the El Paso venture leases at closing (July 2026 agreement, a subsequent event) · as-of 2026-07-31
What else could explain it
other: The carrying value moves with the company’s share of the venture’s results as well as with contributions.
Quotes
“This Venture provides strategic optionality and flexibility, which we expect will enable us to effectively meet future infrastructure capacity needs as AI markets and technologies develop.”
“In July 2026, we entered into an exclusivity agreement to co-develop a data center campus in El Paso, Texas, through a venture in which we would hold a 20% membership interest.”
“Yesterday, as part of our Meta Compute effort, we announced a new strategic venture with BlackRock to develop a new 1 GW data center in El Paso, Texas.”
“Our announcement with BlackRock yesterday is an example of the partnerships that we can structure to complement our approach to building infrastructure capacity.”
Interest on debt raised for the AI infrastructure build
Not sized
The CFO ties long-duration capital to initiatives with long time horizons, especially AI infrastructure projects (claim c22), and the 10-Q explains the rise by higher debt balances (claim c24); 'especially' names AI as the main case of a wider group and does not bound its part of the debt. The build's own costs carry an AI share only where the company bounds AI's part (the methodology rule for the build's own costs). Interest expense, , is quoted as the ceiling.
described, no sizedisclosure: direction only· motive: exploratory· before LLMs: expanded
Interest expense was against (shown as expenses), up on higher debt balances (claim c24); the company issued of notes in May, bringing long-term debt to , and the CFO ties long-duration capital to long-horizon initiatives, especially AI infrastructure projects (claim c22). That names AI first among several uses and bounds nothing, so the channel is left unsized; interest expense is the ceiling. The interest is paid from operating cash flow; the principal it carries funds the build.
Evidence: 5 quotes, 4 figures, 2 confounds, 1 from before coverage
Interest on the senior unsecured notes, below operating income. The filing attributes the rise in interest expense to higher long-term debt balances and names financing among the sources for investments in infrastructure and AI initiatives; in Q2 the CFO ties long-duration debt to AI infrastructure projects. The AI share of the debt is assumed. Layer end-use: the filings put the build first to the company’s own products and model training; from Q2 it also serves the paid model API, which is the compute layer and is read on its own channel, unsized.
Why this motive
Debt raised for long-horizon AI infrastructure with no return attributed (claim c22).
Before LLMs: expanded
Interest expense was in fiscal 2024 (shown as an expense), about a quarter, nearly all on the senior notes (). The size is the interest above that quarterly level, times an assumed AI share. The size is the change AI made, not the whole line.
“Interest expense, net of capitalized interest, recognized on the Notes was $683 million, $420 million, and $160 million for the years ended December 31, 2024, 2023, and 2022, respectively.”
Interest expense (negative: an expense) · 2026-CQ2
Interest expense, prior-year quarter (negative: an expense) · 2025-CQ2
Senior unsecured notes issued in May 2026 · 2026-CQ2
Long-term debt at quarter end · as-of 2026-06-30
What else could explain it
line composition: Interest is on all senior notes, whose proceeds are for general corporate purposes.
other: Long-duration capital funds initiatives with long time horizons in general, AI infrastructure among them (claim c22); no issue is tied to AI alone.
Quotes
“We have generally found it prudent to continue adding cost-efficient, long-duration sources of capital as we make investme nts in initiatives that themselves have long time horizons, especially AI infrastructure projects.”
“Obviously, our strong operating cash flow certainly has put us in a position of strength as it pertains to funding our infrastructure build-out. We've also been evolving our capital structure in recent years to include a greater mix of debt as we work to bring down our cost of capital.”
“Interest expense in the three and six months ended June 30, 2026, increased $542 million, or 225%, and $864 million, or 180%, respectively, compared to the same periods in 2025, due to higher long-term debt balances.”
“In funding these infrastructure investments, the strength of our balance sheet gives us the ability to attract capital from a wide range of markets to supplement the cash flow generated by our business.”
“We currently anticipate that our available funds and cash flow from operations and financing activities will be sufficient to meet our operational cash needs and fund our cash commitments for investing and financing activities, including investments in infrastructure and AI initiatives, as well as any return of capital to stockholders over the next 12 months and thereafter for the foreseeable future.”
Q1 2026described · described, no size · exploratory
Q2 2026direction only · described, no size · exploratory
Cost displaced by AI2 channels · $29mn to $897mn sized · $29mn to $897mn incremental · 1 not sized
partner support · cheap to verify
Advertiser account support handled by the Meta AI business assistant
Not sized
Silent quarter: no claim to size against.
inscrutabledisclosure: not mentioned· motive: efficiency· before LLMs: expanded
The quarter’s sources do not mention the Meta AI business assistant for advertisers; the business agents the call describes serve businesses’ own customers. The parts on the other side, when the channel is read, are the company’s own support staff and the professional-services vendors in marketing and sales.
Evidence: 0 quotes, 2 from before coverage
Support for advertisers (account issues, recommendations, campaign insights) handled by the Meta AI business assistant, tested with advertisers from Q4 2025 and rolled out to eligible advertisers in Q1 2026, which management credits with resolving common account issues at a higher rate. The parts on the other side are the company’s own sales support and customer service staff and the professional-services vendors that marketing and sales also pays; the cost sits in marketing and sales.
Why this motive
Carried from Q1: an implemented tool with a measured resolution rate.
Before LLMs: expanded
At the anchor marketing and sales ( in fiscal 2024) already held sales support, customer service and professional services for community and product operations (anchor claim meta-anchor-c8). Testing of the assistant with advertisers began in the 2025-CQ4 reference quarter (anchor claim meta-anchor-c26). An LLM assistant put into that work is expanded; the size is a saving against the prior cost per issue. The size is the change AI made, not the whole line.
“Marketing and sales expenses consist primarily of employee compensation which includes payroll, share-based compensation and benefits for our employees engaged in sales, sales support, marketing, business development, and customer service functions; marketing and promotional expenses; and professional services to support our community and product operations.”
Q1 2026direction only · our inference · efficiency · $1.5mn to $105mn
Q2 2026not mentioned · inscrutable · efficiency
engineering · cheap to verify
Engineering and product development work done with AI tools and agents
0.05% to 1.5% of the quarter’s revenue
Incremental total: counts in full.
our inferencedisclosure: described· motive: exploratory· before LLMs: expanded
The CEO says AI speeds up product development and the CFO that ranking agents raised launches and engineer productivity (claims c51, c53). The May reduction of about employees, with severance of , is not attributed to AI in any of the quarter’s sources. The size, , is the ledger’s: the hours the reference call’s output-per-engineer rate saves at most, times a small captured share.
Evidence: 6 quotes, 3 figures, 2 confounds, 3 from before coverage
Engineering payroll that AI coding tools and LLM-powered agents may displace or avoid: management describes accelerating output from engineers, agents that evaluate content and test ranking changes, and small teams building in a week what took far larger teams months. Management says it uses the gains to build more products rather than to cut cost, and does not attribute the May 2026 headcount reduction to AI; that reduction and its severance are confounds here.
Why this motive
Faster development and more launches described without a cost measure (claims c51, c53).
Before LLMs: expanded
At the anchor AI already powered tools to make product development more efficient, and technical headcount ( employees in total) drove research and development compensation. By the 2025-CQ4 reference call the CFO gave a increase in output per engineer since the start of 2025, mostly from agentic coding (anchor claim meta-anchor-c32). AI tools change output per engineer, not the existence of the line. The size is the change AI made, not the whole line.
“Our AI investments support initiatives across our products and services, helping power the systems that rank content in our apps, our discovery engine that recommends relevant content, the tools advertisers use to reach customers, the development of new generative AI experiences, and the tools that make our product development more efficient and productive.”
“The higher employee compensation was mainly from a 13% growth in employee headcount from 2023 to 2024 in engineering and other technical functions supporting our continued investment in our family of products and Reality Labs.”
“Since the beginning of 2025, we've seen a 30% increase in output per engineer, with the majority of that growth coming from the adoption of agentic coding, which saw a big jump in Q4.”
“Second, LLM-powered agents are also helping with engineering development by evaluating content quality, detecting trends, and testing ranking changes.”
“We've also invested in using LLM-based agentic approaches to transforming our recommendation system, and we grew the number of launches from our ranking agents this half. That also helps make our engineers more productive, and that's another path that we're very excited about as well.”
“We're building coding and developing an internal productivity tools partially because we need to build them ourselves, and we need to make sure that we have tools that are tuned for ourselves.”
“The higher employee compensation was mainly from increases in share-based compensation expense and severance expenses during the three and six months ended June 30, 2026.”
“We ended Q2 with over 75,000 employees down 3% from Q1. This total includes approximately 8,000 employees impacted by the May 2026 headcount reduction.”
Revenue arriving through AI8 channels · $129mn to $302mn sized · $0 incremental · 7 not sized
product ranking · cheap to verify
Advertising revenue gained from AI ad ranking and recommendation models
Not sized
AI named beside ad targeting and measurement tools, currency and macro conditions; no measure separates the ranking models’ share of the price gain.
Matched line moved : the whole line, not this channel.
described, no sizedisclosure: direction only· motive: offensive· before LLMs: relabelled
The CFO credits user understanding models combined with GEM with more ad clicks and a uplift in conversions on Facebook, and early LLM pilots with more app event conversions on Instagram (claims c28, c27). These are rates on particular surfaces. The 10-Q explains the price per ad rise of as mostly ad performance from ad targeting and measurement tools, together with currency (claim c31), and the CFO adds macro conditions (claim c30); no measure separates the models’ part, so the channel is unsized. Advantage+ end-to-end solutions run at over a year (claim c29), a product the anchor already sold, recorded as a metric. The channel is relabelled (anchor claims meta-anchor-c13, meta-anchor-c24); only the LLM pilot is specific to LLMs, and a lift of that kind at scale could re-tag it through a correction.
Evidence: 8 quotes, 8 figures, 4 confounds, 7 from before coverage
Advertising revenue that management credits to its ads ranking and retrieval models (Lattice, GEM, the adaptive ranking model, Meta Generative Recommender), which it says lift conversion and click rates on particular ad types and surfaces. The filing explains the higher average price per ad as mostly ad performance from ad targeting and measurement tools, together with currency, and the CFO adds macro conditions; no measure separates the models’ part, so the channel is left unsized. Advantage+, which management calls AI-powered and whose run rate it gives, is a product the anchor already sold; its revenue is recorded as a metric, never as this channel’s size. Advertisers pay from their marketing budgets.
Why this motive
Measured click and conversion movement that management attributes to LLM and GEM ranking models (claims c27, c28).
Before LLMs: relabelled
At the anchor AI already improved ad delivery, targeting and measurement (anchor claim meta-anchor-c4), the CFO credited the Meta Lattice ranking architecture with improving ad performance (anchor claim meta-anchor-c13), revenue through the Advantage+ end-to-end tools had grown fast (anchor claim meta-anchor-c14), Advantage+ Audience showed a lower cost per click in tests (anchor claim meta-anchor-c15) and conversions grew faster than impressions (anchor claim meta-anchor-c16). The 2025-CQ4 reference call gave a conversion lift from a runtime model (anchor claim meta-anchor-c24). Advertising revenue was in fiscal 2024. Coverage reports more rate lifts of the same kind from the same activity, so the tie-break gives relabelled; a later quarter showing an LLM-specific lift at scale could re-tag it through a correction.
“Our AI investments support initiatives across our products and services, helping power the systems that rank content in our apps, our discovery engine that recommends relevant content, the tools advertisers use to reach customers, the development of new generative AI experiences, and the tools that make our product development more efficient and productive.”
“Across all of these efforts, we are making significant investments in AI initiatives, including generative AI, to, among other things, recommend relevant content across our products through our AI-powered discovery engine, enhance our advertising tools and improve our ad delivery, targeting, and measurement capabilities, and to develop new products as well as new features for existing products.”
“One example is our new ads ranking architecture, Meta Lattice, which we began rolling out more broadly last year. This new architecture allows us to run significantly larger models that generalize learnings across objectives and surfaces in place of numerous smaller ad models that have historically been optimized for individual objectives and surfaces. This is not only leading to increased efficiency as we operate fewer models but also improving ad performance.”
“AI has also been a huge part of how we create value for advertisers by showing people more relevant ads. And if you look at our two end-to-end AI-powered tools, Advantage+ Shopping and Advantage+ App Campaigns, revenue flowing through those has more than doubled since last year.”
“And based on tests that we've ran, campaigns using Advantage+ Audience targeting saw, on average, a 28% decrease in cost per click or per objective compared to using our regular targeting.”
“This quarter, we introduced Meta Generative Recommender, a paradigm shift in how our ad system works. Rather than scoring every possible ad individually, we are now using LLMs to reason about ad content and user preferences together and predict the best ad for each person.”
“Combined with our GEM model for ads ranking and sequence learning, these advancements generated an 8.3% increase in ad clicks and a 15.7% uplift in conversions on Facebook.”
“The global average price per ad increased 12% year-over-year, driven by ad performance gains, improvements in macro conditions relative to Q2 of last year, and currency tailwinds.”
“The increases in average price per ad in the three and six months ended June 30, 2026 were driven by an increase in advertising demand, which we believe is mostly due to ongoing improvements to our ad performance from our ad targeting and measurement tools, and a favorable foreign currency exchange impact.”
Q1 2026direction only · described, no size · offensive
Q2 2026direction only · described, no size · offensive
search discovery · cheap to verify
Advertising revenue from engagement gained through AI content recommendations
Not sized
Time spent is credited to ranking improvements, and impressions to users, their engagement and ad frequency together; no measure ties the lifts to impressions or revenue.
Matched line moved : the whole line, not this channel.
described, no sizedisclosure: direction only· motive: offensive· before LLMs: relabelled
On Facebook, video time spent rose globally and over in the U.S. and Canada, where the CFO credits ranking improvements (claim c34); Instagram time spent grew on Feed and Reels recommendation improvements (claim c33), and the largest Reels ranking release added percent points to Instagram sessions (claim c36). Impressions grew , which the 10-Q explains by users, their engagement and ad frequency together (claim c37); no measure ties these lifts to impressions or revenue, so the channel is unsized. It is relabelled, as at the anchor (anchor claim meta-anchor-c18).
Evidence: 8 quotes, 4 figures, 4 confounds, 4 from before coverage
Advertising revenue from additional impressions when AI content recommendations (ranking models, LLM-based content understanding, AI translation and dubbing of videos) raise time spent and sessions. Management gives measured time-spent and session lifts that it credits to ranking improvements; the filing explains impression growth by users, their engagement and ad frequency together, and no measure ties the lifts to impressions or revenue, so the channel is left unsized. Users do not pay; advertisers pay for the impressions.
Why this motive
Measured time-spent and session movement that management attributes to recommendation and ranking improvements (claims c33, c34, c36).
Before LLMs: relabelled
At the anchor the filing credited an AI-powered discovery engine with engagement and monetization (anchor claim meta-anchor-c3), about of Facebook Feed posts and more than of Instagram content were AI-recommended (anchor claim meta-anchor-c17), and a new recommendation architecture had raised Facebook Reels watch time by to (anchor claim meta-anchor-c18). Advertising revenue was in fiscal 2024. Coverage reports time-spent lifts of the same kind from the same activity, with LLMs described as being incorporated but not measured apart, so the tie-break gives relabelled.
“Our AI investments support initiatives across our products and services, helping power the systems that rank content in our apps, our discovery engine that recommends relevant content, the tools advertisers use to reach customers, the development of new generative AI experiences, and the tools that make our product development more efficient and productive.”
“We are investing in Reels and in AI initiatives across our products, including our AI-powered discovery engine to recommend relevant content, which we have already seen results in improved user engagement and monetization of our products.”
“Right now, about 30% of the posts on Facebook Feed are delivered by our AI recommendation system. That's up 2x over the last couple of years. And for the first time ever, more than 50% of the content that people see on Instagram is now AI-recommended.”
“We started partially validating this model last year by using it to power Facebook Reels, and we saw meaningful performance gains, 8%-10% increases in watch time as a result of deploying this.”
“On Instagram, global time spent this quarter grew double digits year-over-year this quarter, largely driven by improvements to our Feed and Reels recommendations.”
“On Facebook, video time spent increased 9% globally year-over-year and over 10% within the U.S. and Canada, where it was driven by ranking improvements.”
“This drove a 15 basis point increase in sessions on Instagram, with particular strength in reshares and time spent, which are both strong indicators of better content to user matching.”
“Ad impressions delivered during the three and six months ended June 30, 2026 grew in all regions, especially in Asia-Pacific, which was driven by increases in users and their engagement as well as the frequency of ads shown on our products.”
“We are investing in Reels and in AI initiatives across our products, including our AI-powered discovery engine to recommend relevant content, which we have already seen results in improved user engagement and monetization of our products.”
Q1 2026direction only · described, no size · offensive
Q2 2026direction only · described, no size · offensive
marketing · cheap to verify
Advertising revenue gained from generative AI ad creative tools
Not sized
The quarter's only measures are counts: about small businesses using at least one AI ad creative tool (claim c40) and image generation adoption more than doubling (claim c41), with no measured lift this quarter. A count sizes nothing (the methodology rule for counts). Q1, which carries a tested conversion rate, stays sized.
described, no sizedisclosure: direction only· motive: product-defensive· before LLMs: expanded
About small businesses use at least one AI creative tool, and the CFO reports faster adoption of image generation (claims c40, c41). These are counts of adoption with no measured lift this quarter, which read directional and size nothing, so the channel is left unsized. The tools act through ad performance, as ranking does, and the ranking channel is unsized too.
Evidence: 2 quotes, 1 figure, 1 confound, 4 from before coverage
Free generative AI tools in the ads creation flow (image and video generation, text variations, end-to-end creative solutions) that may raise advertiser results and spend. Management reports adoption counts and a conversion lift in tests. The lift reaches revenue through ad performance, as the ranking channel’s does; that channel is unsized, so no sized channel contains this one’s dollars and it counts in the revenue total.
Why this motive
Carried: free tools inside the ads products, with adoption counts and no new measured lift (claims c40, c41).
Before LLMs: expanded
At the anchor generative AI ad creative features (text variations, image expansion, background generation) were already in the ads creation tools (anchor claim meta-anchor-c19), and the filing already listed advertiser tools among AI investments (anchor claim meta-anchor-c4); advertising revenue was in fiscal 2024. By the 2025-CQ4 reference call the video generation tools carried a combined revenue run rate of (anchor claim meta-anchor-c25), ad spend running through the tools and never this channel’s size. Coverage gives a measured conversion lift in tests, so it is expanded and sized as an increment. The size is the change AI made, not the whole line.
“Our AI investments support initiatives across our products and services, helping power the systems that rank content in our apps, our discovery engine that recommends relevant content, the tools advertisers use to reach customers, the development of new generative AI experiences, and the tools that make our product development more efficient and productive.”
“Across all of these efforts, we are making significant investments in AI initiatives, including generative AI, to, among other things, recommend relevant content across our products through our AI-powered discovery engine, enhance our advertising tools and improve our ad delivery, targeting, and measurement capabilities, and to develop new products as well as new features for existing products.”
“But right now, we have features supporting text variations, image expansion, and background generation. We're continuing to work to make those more performant for advertisers to create more personalized ads at scale.”
“The combined revenue run rate of video generation tools hit $10 billion in Q4, with quarter-over-quarter growth outpacing the increase in overall ads revenue by nearly three times.”
Small businesses using at least one AI ad creative tool · 2026-CQ2
What else could explain it
bundling: The tools are free features of the ads products; any effect reaches revenue through ad performance.
Quotes
“9 million small businesses on our platforms are now using at least one of our AI ad creative tools, and we're rolling out new end-to-end creative solutions that help advertisers translate performance data into their creative decisions.”
“Image generation, which now lets advertisers produce more creatives at scale from existing content, including a new ability to create images from video assets, saw adoption more than double this quarter.”
Q1 2026direction only · our inference · product-defensive · $28mn to $578mn
Q2 2026direction only · described, no size · product-defensive
product revenue · cheap to verify
Business AIs and Meta Business Agent in messaging
Not sized
No price, revenue or paying share is given; the product was free for most businesses in Q1 and the pricing described has no amount.
described, no sizedisclosure: direction only· motive: exploratory· before LLMs: new
Meta Business Agent launched globally and more than businesses use it each week (claim c42); the CEO describes subscriptions and volume pricing moving toward pay for results (claim c43), and an enterprise platform followed after the quarter (claim c44).
Evidence: 3 quotes, 1 figure, 3 from before coverage
AI agents that represent businesses in chats with their customers on WhatsApp, Messenger and Instagram, answering questions, recommending products and handling support, and from Q2 a platform for enterprises to build their own. Management says they are free for most businesses and describes future monetization by subscription, volume or results; no price or revenue is given.
Why this motive
Pricing described in outline with no revenue given (claim c43).
Before LLMs: new
At the anchor business AIs that represent a business in chats with its customers were in testing (anchor claim meta-anchor-c20), and the CEO expected a multi-year investment cycle before they became profitable services (anchor claim meta-anchor-c21); no revenue was given. By the 2025-CQ4 reference call they carried over weekly conversations, still with no revenue (anchor claim meta-anchor-c27). An LLM agent speaking for a business could not exist without LLMs.
“The longer-term piece here is around business AIs. We have been testing the ability for businesses to set up AIs for business messaging that represent them in chats with customers, starting by supporting shopping use cases such as responding to people asking for more information on a product or its availability.”
“I also expect to see a multi-year investment cycle before we fully scaled Meta AI, business AIs, and more into the profitable services I expect as well.”
“Finally, we're seeing good early traction with our business AIs in Mexico and the Philippines, with over one million weekly conversations between people and business AIs now happening on our messaging platforms.”
Businesses using Meta Business Agent each week, a floor · 2026-CQ2
Quotes
“We made Meta Business Agent available globally this quarter on WhatsApp and Messenger. There are already more than 1 million businesses using them to talk to their customers or complete sales every week.”
“In terms of how we will monetize these, we have a mix of subscriptions, volume-based pricing. I expect that we're going to continue to evolve more of these products to be like our ad systems, where businesses only pay us when we achieve results for them.”
“Earlier this month, we also introduced the Meta Business Agent Platform, which gives enterprises the infrastructure to build, customize, and deploy their business agent at scale on WhatsApp.”
Q1 2026direction only · described, no size · exploratory
Q2 2026direction only · described, no size · exploratory
product revenue
Meta AI consumer assistant
Not sized
Meta AI carries no price and no revenue is given.
described, no sizedisclosure: direction only· motive: exploratory· before LLMs: new
Daily users of Meta AI rose since the rebuild on Muse Spark (claim c45). No price or revenue is given.
Evidence: 1 quote, 1 figure, 3 from before coverage
The consumer assistant across the apps, the standalone Meta AI app and the glasses, powered from Q1 by Muse Spark. It carries no price; management names possible commission structures or a premium offering for later. Its serving cost sits in the infrastructure channels.
Why this motive
Carried: not monetized, usage growing (claim c45).
Before LLMs: new
At the anchor Meta AI had just been released in a new version across the apps (anchor claim meta-anchor-c23), unmonetized, and the CEO expected to grow investment before the new AI products made much revenue and a multi-year investment cycle before they became profitable services (anchor claims meta-anchor-c22, meta-anchor-c21). An LLM assistant could not exist without LLMs.
“I also expect to see a multi-year investment cycle before we fully scaled Meta AI, business AIs, and more into the profitable services I expect as well.”
“But realistically, even with shifting many of our existing resources to focus on AI, we'll still grow our investment envelope meaningfully before we make much revenue from some of these new products.”
“Last week, we had the major release of our new version of Meta AI that is now powered by our latest model, Llama 3. Our goal with Meta AI is to build the world's leading AI service, both in quality and usage.”
Increase in people interacting with Meta AI each day since the rebuild on Muse Spark · 2026-CQ2
Quotes
“Since we rebuilt Meta AI and integrated Muse Spark, we have seen a 60% increase in the number of people interacting with the assistant each day, that continues to grow quickly week-over-week.”
Q1 2026direction only · described, no size · exploratory
Q2 2026direction only · described, no size · exploratory
pricing packaging
Meta One subscription with AI features and tools
Not sized
Launched at about the time of the call, after the quarter, with no price or subscriber count; subscriptions are not split out of other revenue.
described, no sizedisclosure: described· motive: exploratory· before LLMs: expanded
The CEO announced Meta One as just launched, with tools and AI features, tiers to follow (claim c46); the CFO calls it an evolution of the subscription portfolio (claim c57). Family of Apps other revenue was against , which the 10-Q explains by WhatsApp paid messaging and subscriptions (claim c58).
Evidence: 3 quotes, 2 figures, 1 confound, 1 from before coverage
A subscription offering for users, businesses and creators that management describes as an evolution of its subscription portfolio, providing more tools and AI features across the apps, with tiers and pricing to follow. The AI part is bundled with other features. Subscriptions sit in Family of Apps other revenue beside WhatsApp paid messaging.
Why this motive
Launched about the time of the call with tiers and pricing to follow (claim c46).
Before LLMs: expanded
At the anchor paid subscriptions already existed as Meta Verified, inside other revenue with the WhatsApp Business Platform; no AI features were sold in them. A subscription that bundles AI features into a portfolio the anchor already sells is expanded under the tie-break, sized as a level with no traced quarter before AI. The size is the whole of an activity that existed before.
“Other revenue consists of revenue from WhatsApp Business Platform, Meta Verified subscriptions, net fees we receive from developers using our Payments infrastructure, and revenue from various other sources.”
Family of Apps other revenue (WhatsApp paid messaging, subscriptions) · 2026-CQ2
Family of Apps other revenue, prior-year quarter · 2025-CQ2
What else could explain it
bundling: AI features are bundled with other tools in the subscription.
Quotes
“We also just launched Meta One, a new subscription offering that provides more tools and AI features across our apps. As demand grows, we're going to offer a variety of different tiers and pricing options there as well.”
“Meta One is an evolution of our subscription portfolio to create more value for everyday users, businesses, and creators, so they get more features and AI tools to create, connect, and stand out.”
Launched at about quarter end; no price, volume or revenue is given and no reference class in the stored sources bounds a weeks-old API.
described, no sizedisclosure: described· motive: exploratory· before LLMs: new· layer: compute
The CFO says a high-intelligence model API launched at a competitive price and Muse Spark is on OpenRouter for developers in the United States (claim c47); the CEO says distribution through partner channels and coding agents follows (claim c61). Buyers are developers and coding-agent products, later enterprises.
Evidence: 3 quotes
Access to the Muse Spark models through the company’s own public API and through OpenRouter for developers in the United States, priced competitively, with enterprise availability to follow. The buyers are developers, coding-agent products and later enterprises; no volume or revenue is given. Selling compute directly, which management says it is being offered at a premium, has not begun and is not part of this channel. Layer compute: the CFO says the model API launched at a competitive price, so it is hosted model access sold to builders for a price.
Why this motive
A priced product described as encouraging, with no revenue or volume (claim c47); taken as exploratory while it is weeks old.
Before LLMs: new
At the anchor the Llama models were released openly for researchers and developers, with no paid API; the anchor sources give no revenue from models. A paid model API could not exist without LLMs.
Quotes
“We also recently launched a high-intelligence model API at a competitive price and are encouraged by the initial results. We recently made Muse Spark available on OpenRouter for U.S.-based developers, broadening its distribution and making it easier for developers to adopt the model.”
our inferencedisclosure: direction only· motive: offensive· before LLMs: relabelled
Reality Labs revenue was against , the rise credited to AI glasses as Quest sales fell (claims c48, c49); the new glasses ship with Muse Spark (claim c50). The size, , is the ledger’s share of the segment; the channel is relabelled.
Evidence: 3 quotes, 2 figures, 2 confounds, 1 from before coverage
Sales of the glasses the company calls AI glasses (Ray-Ban Meta, Oakley Meta, the Meta Ray-Ban Display and, from Q2, its own line of Meta glasses), which ship with the Meta AI assistant. Their revenue sits in Reality Labs revenue beside Meta Quest headsets, which are falling; it is not broken out. What AI adds to demand, as against camera, audio and frames, is not measured.
Why this motive
A separately priced product sold as AI glasses, whose sales growth management names (claim c48).
Before LLMs: relabelled
The glasses were already sold at the anchor as Ray-Ban Meta AI glasses featuring Meta AI, inside Reality Labs revenue of for fiscal 2024. Coverage reports sales growth without a measure of the AI part, so the activity reads as renamed.
“We have continued to advance our roadmap to include additional AI-enabled offerings such as the Ray-Ban Meta AI glasses, which feature Meta AI, our advanced conversational assistant, as well as other features such as hands-free interaction.”
mix shift: Segment revenue mixes the glasses with Meta Quest.
relabel: Growth may come from new styles and partners rather than the assistant.
Quotes
“Within our Reality Labs segment, Q2 revenue was $431 million, up 16% year-over-year due to strong growth in AI glasses revenue, partially offset by lower Quest headset sales.”
Q1 2026direction only · our inference · offensive · $121mn to $281mn
Q2 2026direction only · our inference · offensive · $129mn to $302mn
Reported lines, year-over-year growth
Revenue +28.0%
Q2 2026. Growing slower than revenue: marketing and sales (+15.2%). A displaced cost shows up as a line that stays under the dashed revenue line. These are the audited lines, as first reported; nothing here is attributed to AI by the filing. Not drawn: research and development, general and administrative, which one-time items move by more than 60% in a quarter; the values are in the table below.
Cost of revenueMarketing and salesTotal costs and expensesRevenue