iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Morgan Stanley chairman and CEO Ted Pick expects artificial-intelligence infrastructure spending to remain extraordinarily large for five to 10 years, and possibly to form a roughly decade-long investment cycle. His thesis is not a promise that spending rises every year: data-center construction, power, financing, technology changes and customer adoption will advance at different speeds.
In Pick’s 2026 earnings discussions, Morgan Stanley’s estimates moved to about $850 billion of data-center capital expenditure in 2026, $1.3 trillion in 2027 and as much as $1.5 trillion in 2028. He also said the market may be only 10%–15% through the investment cycle.
What Pick’s decade thesis actually means
Pick’s argument is that AI is still near the beginning of a physical buildout. The spending includes data centers, servers and networking equipment, electricity infrastructure, cooling systems and the financing needed to put those assets in service. It does not mean that Morgan Stanley has a fixed $10 trillion budget or that annual spending will follow a smooth upward line.
In the Q2 2026 earnings call, Pick used a tenfold increase from roughly $1 trillion of existing cloud-compute capacity as an analogy for about $10 trillion of AI compute. He said $10 trillion over 10 years was imaginable, while stressing that both the eventual amount and its timing are uncertain. He described the cycle as only 10%–15% complete, an estimate attributed to Morgan Stanley research rather than a contractual industry target.
#1 Best Overall
In the Q3 2026 discussion, Pick framed the outlook more flexibly: annual AI capital expenditure could remain around $1 trillion for five to 10 years and contribute about 50–70 basis points to U.S. real GDP. Fifty to 70 basis points equals 0.50–0.70 percentage points of GDP growth or level contribution, depending on how the estimate is measured; it is not a forecast of total economic growth.
How the 2026–2028 estimates changed
The figures below are management discussions and scenarios from Morgan Stanley’s 2026 earnings calls, not guaranteed industry spending.
| Year | Earlier figure discussed | Updated or upper scenario | What it represents |
|---|---|---|---|
| 2026 | $575 billion | About $850 billion | Data-center capital-expenditure forecast cited by Pick in Q2 2026 |
| 2027 | About $700 billion | $1.3 trillion | Data-center capital-expenditure forecast cited by Pick in Q2 2026 |
| 2028 | Not stated | Up to $1.5 trillion | Scenario discussed in the Q2 2026 call |
Pick later cautioned that the precise 2028 hyperscaler-capital figure could be $1.0 trillion, $1.2 trillion or $1.4 trillion. That range illustrates why the numbers should be read as scenarios rather than fixed commitments. The Q2 figures also combine a broad data-center-capex outlook with estimates that can change as companies revise plans, prices and deployment schedules.
Why the cycle could last another decade
Compute demand keeps expanding
Training and running AI models require different kinds of processors, memory, storage and high-speed networking. New model architectures, longer context windows, multimodal systems and software agents can increase demand even when the cost of an individual computation falls. If useful AI applications spread through customer-service, software, research, industrial and public-sector workloads, providers must keep adding capacity.
Physical construction takes years
A large data center cannot be delivered as quickly as a software release. Sites need land, permits, buildings, electrical equipment, cooling and fiber connections. Orders for transformers, switchgear, generators and specialized cooling can have long lead times. A multi-year queue means spending can remain elevated after a company has already decided that it wants more compute.
Power is a separate bottleneck
AI campuses need dependable electricity at a scale that can exceed the capacity available on an existing connection. Utilities and developers may need new substations, transmission upgrades, generation and long-term power contracts. A shortage of power can delay otherwise funded projects, making the cycle appear too slow even when demand is strong.
Financing must scale with the assets
AI infrastructure is capital intensive and often requires debt, leases, project finance or equity in addition to operating cash flow. Higher interest rates, tighter credit or uncertainty about the useful life of accelerators can delay orders. Conversely, abundant financing can push companies to build ahead of proven demand, increasing the risk of excess capacity.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →Adoption determines returns
Infrastructure spending only earns attractive returns if customers pay for the resulting compute or if AI-driven products create enough new revenue and productivity. Enterprise procurement, data-governance rules, reliability requirements and integration costs can make adoption slower than headlines suggest. A long investment cycle therefore depends on actual usage, not only on the number of announced projects.
Why the path will be uneven
Pick explicitly expects periods when capital investment or technology runs ahead of adoption. In his Q3 remarks, he said the cycle could feel “too fast” because of overbuilding, or “not fast enough” because another constraint has appeared. He also pointed to poor allocation, power limits, financing stress and other bottlenecks.
- Overbuilding: Developers may add capacity for demand that arrives later, leaving some facilities underused and pressuring prices.
- Technology turnover: Faster or more efficient chips can reduce the value of installed equipment before its expected life ends, forcing write-downs or accelerated replacement.
- Capital-allocation mistakes: A project can be technically feasible but economically weak if its location, power price, cooling design or customer mix is wrong.
- Funding stress: Expensive debt or a loss of investor confidence can interrupt construction even while long-term demand remains intact.
- Adoption delays: Businesses may run pilots without moving to production, postponing the revenue needed to justify the next wave of capacity.
These risks make a sustained cycle different from a guaranteed straight-line boom. Spending can pause, shift between regions and suppliers, or fall in one segment while another expands.
Bull case versus bubble or slowdown case
| Question | Long-duration investment case | Bubble or slowdown case |
|---|---|---|
| Spending scale | Annual AI capex stays near $1 trillion for several years as capacity catches up with demand. | Announcements exceed orders, and spending drops after an initial buildout. |
| Adoption timing | AI services move from experiments into recurring production workloads. | Productivity gains and customer revenue arrive too slowly to support planned capacity. |
| Power and construction | Grid upgrades and new generation unlock additional campuses. | Interconnection queues, equipment shortages or permitting hold projects back. |
| Financing | Debt and equity markets keep funding projects with credible customers. | Higher rates, weaker credit or falling hardware values make projects uneconomic. |
| Capital allocation | Operators learn where compute, cooling and power produce the best returns. | Overbuilding and rapid hardware obsolescence create stranded assets. |
| Economic payoff | AI investment adds roughly 50–70 basis points to U.S. real GDP at around $1 trillion of annual capex. | Economic gains are delayed, concentrated or insufficient to offset the capital cost. |
What Morgan Stanley’s role is in this buildout
Morgan Stanley’s 2026 shareholder letter says accelerating AI deployment may fundamentally change how the firm originates and executes business while its trusted-advisor role remains unchanged. Pick presents the firm as an adviser, structurer, financier and allocator of capital across public and private markets.
That role helps explain why Pick emphasizes both the opportunity and the plumbing required to realize it. Investment banks can advise data-center operators and technology companies, arrange debt and equity, structure partnerships, and help investors assess power, infrastructure and adoption risks. It does not make Morgan Stanley’s capex scenarios commitments by the companies that ultimately spend the money.
Best Value
How to interpret the forecasts
Treat every number as a dated scenario
The $850 billion 2026 figure, $1.3 trillion 2027 figure and up-to-$1.5 trillion 2028 figure came from management discussions in 2026. Pick’s later $1.0 trillion–$1.4 trillion range for 2028 shows that the estimate is sensitive to assumptions. Readers should not add the figures together as a guaranteed three-year total.
Watch physical and financial indicators
Useful checks include utility interconnection timelines, transformer and cooling-equipment orders, data-center construction starts, accelerator delivery schedules, vacancy and pricing for compute, customer usage and the availability and cost of project financing. A strong announcement pipeline without corresponding power, equipment and customer commitments would support the slower interpretation of the thesis.
Separate infrastructure demand from investor returns
Even if total AI compute demand rises for a decade, individual projects, chip generations or financing structures can lose money. The cycle describes aggregate investment, not a guarantee that every supplier, facility or investor benefits equally.
Pick’s central message is therefore durable but conditional: AI infrastructure may require a once-in-a-generation flow of capital, with roughly $1 trillion a year plausible for five to 10 years, yet bottlenecks, overbuilding, financing conditions and adoption can repeatedly change the pace. The most defensible reading of “another decade” is a long, volatile buildout—not a promise of uninterrupted annual growth.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
