Even as AI-generated content floods social apps — producing confusion, distrust and in some cases revulsion — platforms keep doubling down on AI investment. A new report says the scale of that investment is considerably larger than what has been disclosed.
The Wall Street Journal reported that Meta and other AI developers are spending far more on AI than they currently communicate, because of back-end deals that are not reflected on their books. Per the WSJ, Meta holds roughly $693 billion in off-balance-sheet commitments, largely related to AI expenditure. That sits on top of the financial liabilities listed in Meta's quarterly reports.
In its Q2 update last month, Meta capped 2026 AI infrastructure spend at $145 billion. That figure excludes expenses that kick in once these projects are completed.
The WSJ's example is Hyperion, Meta's gigantic data center project in Louisiana — about the size of 1,700 football fields.
The structure works like this. Meta initially agreed to lease Hyperion for a four-year term starting in 2029, with options to renew for up to 20 years. It guaranteed that it would make bondholders whole if it does not stay the entire two decades. Because the company does not think payments under that guarantee are probable, it has recorded no liability on its balance sheet.
In other words, no financial liability from the project needs to be listed until 2029. The WSJ estimates roughly $347 billion in future lease obligations that Meta can keep off its official accounting for now.
Meta's investment rests on a foundational belief that its AI projects will be a massive ongoing revenue opportunity. If that does not materialize, the exposure could be ruinous.
Put differently: Meta really needs people to be excited about its AI tools. Which is why it will keep pushing them in its apps, in stores, on billboards and everywhere else. Meta needs personal superintelligence to happen, or it will have paid trillions for a failed technology.
The obstacle is trust. Meta's poor record on data handling and privacy, plus generally low consumer trust, works directly against the vision. A 2025 Forrester survey ranked Meta among the least trusted corporations in the world. It is also in the middle of major litigation over social media harms, and coverage of that — along with the ongoing push to restrict teen social media use — has fed a broader negative view.
For the plan to play out, millions of people would have to let Meta's AI tools access their personal health data, financial records and personal information, so that personal AI agents could deliver custom advice and daily action plans. It is an interesting concept. At this stage, enough people extending that level of trust to Meta looks unlikely.
Read purely as financial news, the practical implication gets missed. What matters to advertisers is not the number but the direction of the pressure it creates.
First, Meta's push on AI features will intensify. Justifying commitments at this scale requires AI adoption rates and AI-powered ad product performance. Advertisers should expect automated products to keep arriving as defaults, with manual controls pushed further back. Even as the toolset expands — as with Meta opening its Creator Studio app to all U.S. and Canada iOS creators — the direction those tools point is automation.
Second, data access demands grow. The vision presupposes broad consent to personal data access, and the rules brands operate under when handling first-party data inside Meta's ecosystem move with that current.
Third, reputation risk becomes ad environment risk. Meta is already running reputational defense, whether by giving 15,000 AI glasses to blind and low-vision users or by actioning 750,000 underage accounts in Australia. That such defense is needed means the brand safety checklist gets longer.
The operational conclusion is not to cut Meta budget. It is to plan on the assumption that platform policy and product mix will keep shifting for years. Campaign architectures that depend entirely on one ad product are fragile in that environment.
Per the Wall Street Journal, it is largely AI-related and separate from the financial liabilities listed in Meta's quarterly reports — contracted obligations such as data center leases whose payment duties are not yet recorded in the accounts.
Take the Hyperion project in Louisiana. Meta agreed to a four-year lease starting in 2029 with renewal options up to 20 years, guaranteeing bondholders would be made whole on early exit. Because payment under that guarantee is not considered probable, no liability was recorded.
It capped 2026 AI infrastructure spend at $145 billion in its Q2 update. That figure does not include the off-balance-sheet costs that begin after projects are completed.
Assume Meta will keep pushing AI features and automated products hard. Avoid campaign structures wholly dependent on a single ad product, and plan operations on the assumption that platform policy and product mix will keep shifting for several years.
To apply what you just read to your own site, start with a free audit of where things are now.
A strategist replies within 24 hours on business days.
TikTok owner ByteDance and the MPA reached a memorandum of understanding on guardrails for generative AI models like Seedance and Seedream — a deal seeded by a viral Brad Pitt deepfake.
LLM visibility depends on your whole digital footprint, not your website. Teams are still built by channel. Mordy Oberstein's uncomfortable advice: do not reach for an action plan yet.
Review content incrementality is the metric affiliate platforms cannot give you. Here is how one creator ends up paid four times over, and the six questions that separate acquisition from margin leakage.