Citigroup: Big Tech AI Spending to Exceed $2.8 Trillion by 2029

Citi’s upgraded AI capex call: why it matters now

Citigroup has raised its estimate for Big Tech AI spending to more than $2.8 trillion by 2029, up from a prior $2.3 trillion view, citing earlier-than-expected build-outs by hyperscalers and broadening enterprise deployment. The bank now sees AI capital expenditures reaching about $490 billion by 2026, underscoring how quickly the investment cycle is scaling beyond first-wave pilots.

What’s driving the upgrade

Citi points to two reinforcing forces: (1) hyperscaler race dynamics, Microsoft, Amazon and Alphabet accelerating data-center and model-inference footprints; and (2) tangible enterprise use cases that justify budgets outside consumer chatbots, from drug discovery to technical documentation and workflow automation. The result is a steeper near-term capex ramp with a longer investment tail into the late 2020s.

The infrastructure bill: silicon, steel and electrons

Behind the headline number is an industrial-scale build: racks of accelerators, high-voltage substations, liquid-cooling retrofits, and miles of fiber. Citi estimates global AI compute will require ~55 GW of additional power by 2030, with about $1.4 trillion of the spend in the U.S. alone, implying multi-year demand for generation, grid upgrades and onsite backup. For context, separate industry trackers also foresee trillion-dollar AI outlays through 2029, though with different scope (e.g., IDC’s broader AI-spend series at $1.3 trillion by 2029).

Follow the money: more debt in the mix

Citi highlights a notable financing shift: Big Tech is leaning more on borrowing to fund AI infrastructure rather than exclusively on operating cash flows. That keeps WACCs sensitive to credit spreads and creates a cleaner read-through to bond markets. It also explains why free cash flows can dip temporarily even as revenue and bookings rise, an effect equity investors will need to model carefully.

Market implications: who wins, who wobbles

Semiconductors & hardware

Near-term beneficiaries remain accelerator vendors, advanced packaging, HBM memory, power electronics and optical interconnects. If Big Tech AI spending hits $2.8 trillion by 2029, backlog visibility for leading-edge foundry and packaging capacity improves, while second-source suppliers gain optionality as buyers diversify supply. Longer term, efficiency gains (sparser attention, quantization, custom silicon) could temper unit growth but expand total workloads. (Inference anchored in Citi’s capex path and recent vendor roadmaps.)

Utilities, grid equipment & data-center REITs

The forecast 55 GW of incremental AI load implies sustained orders for transformers, switchgear, cables, and gas-peaker or renewable-plus-storage capacity, depending on region. Utilities with favorable rate frameworks and interconnection backlogs may see multi-year capex plans trend higher; data-center landlords with power-rich campuses keep pricing power as “MW-first” leasing spreads. The bottleneck is increasingly power and permits, not land.

Software & enterprise AI adopters

Citi flags named enterprises already extracting value from AI, reinforcing the “utility phase” narrative, validated ROI tends to unlock second-wave budgets in vertical software and workflow tools. Pricing pressure at the model/API layer (as vendors cut rates) can be margin-accretive for app-layer integrators who monetize outcomes rather than tokens.

Credit markets & Big Tech balance sheets

A debt-tilt in funding raises questions on duration and capital-return cadence. Expect more frequent jumbo prints from AA/A issuers, plus structured financing tied to data-center assets. Bond investors will focus on capex discipline, power procurement contracts, and utilization risk; equity investors will scrutinize free-cash-flow bridges and the timing of a potential opex-to-revenue flywheel.

How this forecast compares, and what could change it

Citi’s $2.8 trillion lens is specifically about AI-related infrastructure by Big Tech, not every category of AI software/services worldwide. Broader trackers (e.g., IDC) measure total AI spending across industries and products and therefore print different totals. Either way, the signal is the same: the AI buildout is large, front-loaded and power-constrained. Upside risks include faster agentic adoption or breakthrough efficiency in training; downside risks include policy friction, chip supply hiccups, and tariff or export-control disruptions that slow deployments.

What corporate finance desks and CIOs will watch next

Power SLAs & interconnect queues: Access to megawatts is now a gating item for AI capacity; watch utility filings and ISO/RTO queue reforms.
Model efficiency curve: If inference cost per token keeps falling, total demand likely elastic, cheaper use drives more use, supporting Citi’s top-line trajectory even as unit costs drop.
Financing cadence: Expect opportunistic bond deals ahead of major campus phases; duration and spread moves will influence capex timing.
Enterprise adoption proofs: Clear ROI at flagship adopters (life sciences, industrials, legal/tax content) tends to unlock sector-wide budgets.

Risks that could derail the path

Power scarcity: Delays in transmission, permitting or gas supply can push projects to 2027–29, flattening early-curve growth.
Policy shocks: Fresh tariffs, export restrictions or localization rules that balkanize supply chains.
Funding fatigue: If debt windows shut or spreads gap wider, some programs could slip, denting the front-loaded ramp.

Outlook

On Citi’s numbers, Big Tech AI spending is set to exceed $2.8 trillion by 2029, with $490 billion of capex already visible by 2026. The second half of the decade becomes a contest over power, permits and procurement, not just GPUs, placing utilities, grid vendors and balance-sheet engineering at the center of the AI story.