Meta reallocates Reality Labs budgets to fuel a $135 billion AI infrastructure spree. Zuckerberg targets a $9 trillion valuation by 2028.
Why Meta Is Redirecting Reality Labs Capital
Meta is shifting spending away from Reality Labs and toward a $135 billion AI infrastructure buildout. The move treats compute, data centers, and model training capacity as the scarce resources that matter most for an AGI push, rather than headset hardware or metaverse product lines that still need years of consumer adoption to pay back. Zuckerberg has tied the strategy to a $9 trillion valuation target by 2028, which implies the company must show investors that AI infrastructure spend converts into durable platform power, not just higher operating costs.
Reality Labs was never free capital. It absorbed large, multi-year budgets with long payback cycles and unclear unit economics. Reallocating that budget does not automatically create value; it only frees capacity to bet on a different product surface. The risk is concentration: if AI infrastructure fails to produce products people pay for or use daily, Meta will have sold a slower, more patient bet for a faster, more capital-intensive one.
What a $135 Billion Infrastructure Spree Actually Buys
AI infrastructure spending is not abstract research. It buys chips, power, cooling, networking, storage, and the software stack that keeps training and inference jobs running at scale. That stack determines how quickly Meta can train larger models, how cheaply it can serve them to billions of users, and whether it can host third-party workloads that turn fixed capacity into recurring revenue. A $135 billion commitment is a bet that ownership of that stack will matter more than licensing capacity from others when AGI-class systems become productizable.
Infrastructure also locks in multi-year contracts and location decisions. Power availability, grid reliability, and land for data centers set the real ceiling on how fast the spend can turn into usable flops. Teams that treat this as a pure model problem miss the operational work: capacity planning, utilization targets, failure domains, and the gap between peak theoretical performance and sustained training throughput.
- Prioritize utilization over peak capacity: idle clusters burn cash without advancing model quality.
- Separate training clusters from product inference so consumer traffic does not starve research runs.
- Measure cost per useful training step and cost per served request, not only total spend.
- Keep Reality Labs talent and IP transferable where it helps spatial computing, simulation, or device-side AI rather than discarding everything wholesale.
Tradeoffs for Product, Talent, and Investors
Sunsetting or shrinking Reality Labs programs sends a signal to engineers, partners, and consumers who built around those roadmaps. Hardware roadmaps, content ecosystems, and enterprise XR pilots all depend on multi-year commitment. A clean cut can free budget and executive attention; a messy cut leaves stranded inventory, broken partner expectations, and internal teams competing for remaining scraps. The practical path is staged wind-down: freeze new long-horizon bets, finish work that has clear near-term product use, and migrate people into AI infrastructure and model product roles where their systems skills still apply.
For investors, the $9 trillion by 2028 framing is a forcing function. It requires AI spend to show up in growth, margins, or strategic optionality before the market loses patience with infrastructure depreciation and energy costs. That means Meta must connect the $135 billion program to concrete surfaces: better ranking and generation in social products, developer APIs, enterprise tools, or advertising systems that improve conversion. Capex without product pull-through is just a larger balance sheet risk.
How Leaders Should Read This Kind of Pivot
Capital reallocation from a long-horizon hardware platform to AGI infrastructure is a classic platform company move: abandon the bet with the longest, most uncertain consumer path and double down on the bet closest to the company’s distribution advantage. Meta already reaches massive audiences; AI infrastructure is the way to make every surface smarter, more automated, and harder to leave. The open question is execution discipline—whether the company can kill or shrink Reality Labs programs without cultural thrash, and whether the $135 billion buildout stays tied to measurable product outcomes instead of prestige model releases.
Outside Meta, the lesson is narrow and useful. When a company redirects a major division’s budget into infrastructure, treat public valuation targets as communication tools, not forecasts. Ask what gets deprioritized, who retains ownership of the new spend, and which product metrics must move before the next planning cycle. A sunset plus a $135 billion AI spree is only a strategy if the freed capital and the new capacity share one accountable roadmap through 2028.