Amer­icas CIO View

Happy Birthday Chat GPT! At just 3 years old, this big boy is outgrowing the house

On November 30th, 2022, Open AI (Artificial intelligence) introduced Chat GPT to the world. This marked the birth of AI capabilities available and intended for the mass market. Ever since Chat’s acclaimed birth, the Tech titans raced to become proud new parents of their own AI bots. These children are growing rapidly and consuming resources that even their super affluent parents might struggle to afford.

At this year’s Thanksgiving, AI children will be out of the cradle and sitting at the table, talking and eating. Best to cook an extra turkey and cake for each of these very hungry and precocious children to avoid drama. And surrender charge of the music and TV.  Next year, the feast likely requires a much larger house and the costs of food and festivities will require that guests chip in.

U.S. AI datacenter capex: $200b 2024, $350b 2025, $450b 2026…$750b 2030

We estimate that US investment spending on new AI data center construction, including all internal and onsite equipment could be roughly $350 billion in 2025 or up ~75%. This includes all computer servers, all chips, routers, wiring, computing control systems/software, cooling, electrical and on-site power equipment. We estimate chips are near 60% of overall 2025 costs.

In 2025, this investment spending arguably contributed roughly 0.5% to US gross domestic product (GDP) growth. While most of the chips were imported, most of this spend was captured in US chip designer profit margins. Furthermore, the extended effects of this capital expenditure (Capex) to production activity at certain manufacturers and electric utilities already at high capacity utilization may suggest additive capex. Additionally, the wealth effect from the stock market’s enthusiastic view on this investment spending and potential future profits boosted household consumption spending power.  US GDP now appears to be AI sensitive.

The S&P 500 appears increasingly a digital business & very AI capex sensitive

After terrific 3Q earnings results from Tech companies and reviewing their more aggressive plans for multiple years of robust capex growth on AI related infrastructure, we raise our 2025, 2026, 2027 S&P earnings per share (EPS) estimates by about 3% and introduce estimates through 2030. We expect an EPS compound Annual Growth Rate (CAGR) of 10% for the S&P 500 through 2030, fueled by an 18% CAGR at 10 Magnificent firms involved in digital businesses, AI services and advanced tech equipment and microchips. Excluding these Magnificent 10, we expect an EPS CAGR of 6–7% for the rest of S&P to 2030.

Our 2026 S&P EPS estimate of $307, factors in DWS economist forecasts for 2.1% US GDP growth (1.3% prior), yet unemployment climbing to 4.6% with inflation averaging 2.9%. We also factor in DWS forecasts for three U.S. Federal Reserve (Fed) cuts to 3.0-3.25%, EUR/USD foreign exchange (FX) rate at 1.15 and $57/bbl oil. While these macro inputs remain important, because S&P EPS has become highly sensitive to data center capex, we introduce a rough model of our capex expectations through 2030. This capex outlook is very uncertain and its future Return on Invested Capital (ROIC) is especially difficult to predict.  Current S&P firms are highly exposed to this capex as revenue sources and we believe that both S&P 500 firms and others will be exposed to the ROIC. Our S&P EPS forecasts are for current constituents.

A.I.’s Tech titan parents are asking extended family and friends to help pay bills

The magnitude of capital expenditures planned through 2030 is so large that it may require more than cash flow and moderate debt capacity use from Tech titans to comfortably fund. Five Tech data center hyper-scalers issued $121billion of debt in 2025 and are establishing long-term leases with data center landlords who are themselves forming joint ventures to tap into other sources of debt like capital. These funding developments may suggest that the hurdle rate for a sufficient return on these data centers may be moving from a cost of equity to more of a cost of debt. There also may be tax efficiencies to moving these assets to Real Estate Investment Trust (REIT) and private real estate structures, however, investors should consult their tax advisors regarding their individual circumstances, DWS does not provide tax advice.

We think the business of data centers (perhaps also AI chips ownership) might become more distinct from AI services over time. Some companies likely choose to be more of one than the other and leases might help separate the economics. Yet, we think that certain Tech titans will choose to stay integrated and have special expertise in both business types. Ultimately, the health of the data center business likely will be closely linked to the size and success of AI services. We believe that the current cost of chips is a challenge to achieving strong ROIC on AI services (at least in the early years), or for whoever owns the chips, which could also pressure data center pricing if the build out is faster than profitable AI services emerge. We think greater computing and energy efficiency in chips is important toward attractive economics of data center/AI service business models and/or more competitive chip pricing over time. The prices and useful life of the chips and their power consumption are important variables as well as the size of AI revenue. There is upside and downside on our S&P EPS outlook through 2030, but our forecasts are more dependent on the capex than the return on capex, especially for the next couple of years.

2026 yearend S&P 500 target: 7500 = 22x forward EPS of $340

We think S&P could sustain a 10%+ premium to a steady-state forward price-to-earnings (P/E) ratio compares a company's current share price to its earnings per share, with 10% EPS growth to 2030.