HSBC: AI Model Slowdown Won't Suppress Traditional Data Centers; Triple Tailwinds Support 11%-12% AFFO Growth
nashnova research
HSBC argues the real demand driver for legacy data centers is not frontier-model iteration but AI's commercial rollout into enterprise workloads; combined with power-supply bottlenecks and tightening regulation, data-center REITs could deliver 11%–12% AFFO per-share CAGR, well above their five-year average.
AI models are slowing down — why do legacy data centers still benefit?
HSBC's core thesis: what drives legacy data-center demand is not how fast frontier models advance, but how fast AI technology diffuses into real enterprise use cases.
This means → even if model development cools, AI inference — actually running trained models to handle live requests — and enterprise deployment keep pushing infrastructure demand higher.
Microsoft estimates that by 2032, AI-chip-driven capacity will account for less than 50% of new data-center builds. In plain terms = not all new capacity becomes high-density AI compute; traditional workloads still claim the larger share.
Cloud giants are still spending — where does the money go?
HSBC's tech team forecasts the six major cloud providers will grow capex by 37% in 2027 and 9% in 2028.
AI adoption transmits through three channels: cloud computing + data processing + enterprise infrastructure upgrades.
This means → capital spending does not only feed AI training clusters; legacy data centers capture a steady stream of incremental demand.
How severe is the power bottleneck?
Power access has become the binding constraint on new capacity. Multiple U.S. regions have raised electricity, environmental, and water requirements, forcing large power users to bear more of the cost for plants, transmission lines, and grid upgrades.
On the regulatory front: California revoked environmental-review exemptions for data centers and now requires water-use disclosure; Texas ordered a permit pause until the reliability council completes its audit.
This reflects a widening gap between demand and deliverable supply — permitting and grid-connection timelines are lengthening, compressing supply elasticity further.
How many projects have been delayed or canceled?
Data Center Watch reports that in Q2 2026 at least 45 projects worth ~$68 billion were delayed or scrapped; in Q1, at least 75 projects worth ~$130 billion.
In plain terms = what operators want to build cannot get built, so existing data-center capacity grows scarcer.
This means → rents and asset utilization both get a lift — the tighter supply is, the more valuable the installed base becomes.
Could rising rates eat into those gains?
HSBC's sensitivity analysis: if new-debt rates run 50 basis points above the base case, 2027–2029 AFFO — adjusted funds from operations, the key measure of a REIT's real earning power — per-share forecasts drop only 0.3%–0.8%, shaving roughly 10–40 bps off growth.
This means → rate pressure remains modest relative to the demand and supply drivers.
Can the three tailwinds last — and what should investors watch?
HSBC projects data-center REITs will deliver roughly 11%–12% AFFO per-share CAGR from 2025 to 2028, well above their five-year historical average.
The three structural tailwinds: demand gains from AI diffusion + supply constraints from power and regulation + robust AFFO growth.
The key verification points are two variables: ① whether enterprise AI adoption sustains inference-side workload growth, and ② whether supply-side constraints hold through 2027–2028. How these two evolve will directly determine whether the growth forecast materializes.
市场有风险，内容仅供研究参考，不构成投资建议。