Meta swaps its infrastructure chief as capex heads for $135 billion to $145 billion. Watch usable compute
Meta is replacing the executive who runs its infrastructure just as its capital budget runs at $135 billion to $145 billion. The new name is the least important part of the story.
Santosh Janardhan is expected to leave in early 2027. Dave Brown, a former Amazon Web Services executive who joined Meta earlier in 2026, is set to succeed him. Janardhan will stay through the first quarter to support the transition. (Reuters)
CTOL's view, held with high confidence, is that this is a staged transfer of execution responsibility that changes oversight and does not yet justify higher expected infrastructure returns. It says little about whether Meta can build faster or earn more from its assets. Investors should leave their infrastructure-efficiency assumptions unchanged on the appointment alone. A change would take a comparison of capacity delivered into service against the schedule and cost set before the handoff.
The number everyone will quote is the budget. Meta reported $34.91 billion of third-quarter 2026 capital expenditure. It guided to $135 billion to $145 billion for the full year, with finance-lease principal payments included in its annual measure. That guidance covers more than AI infrastructure. It shows the scale of the investment program, and shows little of how much usable AI capacity the spending will produce. (Meta results release)
Spend measures money, capacity measures output
Finished construction and delivered accelerators are still short of useful computing capacity. Site readiness, electricity, networking, accelerator deployment and reliable operation have to converge. Power that arrives after hardware installation can defer usable output. So can a network that is unready when accelerators are commissioned. The capital has already been deployed in both cases.
That interval between investment and productive service is what the leadership announcement leaves out. A commissioning delay can postpone the benefit of capital already deployed while annual capex barely moves. Higher spending can also coexist with better delivery if more capacity reaches service. The headline budget settles neither.
The number that matters is a cohort
The right unit of analysis is a comparable deployment cohort, rather than Meta's whole annual investment total. An analyst needs, for each cohort, the planned and actual dates for power, network acceptance and reliable compute service. Capital costs must be assigned on a consistent basis, and finance-lease principal payments need handling consistently, kept apart from asset costs.
Meta's record offers no matched project-level schedule and cost baselines. Dividing total 2026 guidance by a future AI-capacity estimate would manufacture a unit-cost figure instead of measuring one.
Brown inherits a schedule he did not set
Janardhan's presence through March 2027 makes the transfer gradual. Brown can oversee the resolution of problems after the handoff. Projects commissioned later in the year may reflect site selection, equipment commitments and delivery schedules established earlier. Credit for a 2027 completion, or blame for a 2027 delay, may belong partly to his predecessor's decisions.
The calendar holds a second trap. The $135 billion to $145 billion figure is guidance for 2026, and it was never Brown's 2027 spending target. It cannot serve as the cost denominator for a 2027 productivity verdict until the capital is matched to the capacity it funds. Shareholders should assign the succession no higher execution value, and should use the 2026 budget to declare neither success nor failure under the next chief.
Brown's AWS experience may be well suited to coordinating a hyperscale buildout. The plausible upside is better removal of dependencies across power, network and equipment delivery. That is an operating hypothesis to test before it earns a return premium.
Meta's third-quarter 2027 results, covering the quarter ending September 30, give a dated checkpoint after the first full post-handoff quarters. A credible upward revision needs comparable project evidence. That means capacity entering reliable service on or ahead of pre-handoff schedules, at controlled attributable cost. A higher amount spent or a new management mandate would show only spending. The strongest case would show the same improvement across more than one deployment cohort.
The critical evidence will be whether scheduled capacity actually becomes usable compute, and at what cost. Until a cohort comparison exists, hold execution assumptions steady. Silence from Meta is no sign of failure, and a strong résumé earns no productivity premium.