AI drove 65% of US growth. Now the question is what the spending pays back
AI added roughly 1.3 points to GDP growth over three quarters. Goldman Sachs sees earnings growth slowing in 2027, and Arbol is now selling protection against GPU values falling.
- AI investment added roughly 1.3 percentage points to real GDP growth over the last three quarters, about 65% of the roughly 2% cumulative growth. Self-reported m:…5064
- An 18% total return is predicted for the S&P 500 through the end of 2027, with earnings growth of 11% in 2027, down from roughly 30% in 2026. Self-reported m:…2656
- The IT sector will account for 70% of MSCI ACWI net income growth in 2027. Self-reported m:…5969
- US voters backed limits on data centers in 40 of 41 ballot measures on November 3, 2026. m:…5971
- Nebius closed Q3 2026 contracts above $25m per MW, up from $20-25M in Q2. Self-reported m:…5971
What the growth numbers say
AI investment added roughly 1.3 percentage points to real GDP growth over the last three quarters, and accounted for approximately 65% of the roughly 2% cumulative GDP growth in that period. The growth contribution is led by software and computers, and the analysis calls AI the number one source of growth for the US economy. m:…5064
The figures come with a caveat the analysis states plainly: they are gross and do not net out imports. Even so, the scale is hard to ignore. m:…5064
The payback question
The question is shifting from how much is being spent to what the spending returns. The narrative is shifting from who can invest the most to who can capture the highest returns, and the investment debate in 2027 will sharpen as free cash flow thins and leverage rises. Free cash flow at US hyperscalers is likely to trough in 2027. m:…5969
The split inside the buildout matters. Spending-dependent names like neoclouds and data center REITs face revenue impairment if spending plateaus, while facilitators like IT services and software firms are positioned to monetize AI capacity earlier than infrastructure providers. m:…5969
Goldman Sachs expects investors to shift focus to the next earnings beneficiaries of the AI boom as AI capital expenditure decelerates. AI investment produced about half of the growth in earnings per share for the S&P 500 in 2026, and Goldman Sachs predicts the S&P 500 will generate a total return of 18% through the end of 2027, with earnings expected to grow by 11% in 2027, a step down from the roughly 30% growth rate observed in 2026. The biggest downside risk to equities in 2027 is a deterioration in the earnings outlook for 2028. Each $50 billion of AI capital expenditure in 2028 would shift S&P 500 earnings-per-share growth by about 1 percentage point, which is why the 2028 spending plans matter as much as the 2027 results. m:…2656
The hardware bet gets complicated
The physical side of the bet is also changing. The useful lives of graphics processing units may extend beyond the typical 4-5 years historically underwritten, because constraints on data center supply and the physical demands of next-generation chips could extend the useful economic lives of older GPUs. Longer useful lives and additional rental income could support higher GPU residual values, strengthening collateral coverage. m:…5971
The limits are concrete. US grid capacity additions are capped in the 15-20GW-a-year range. NVIDIA's Blackwell GB200 NVL72 draws about 120 kW per rack, while 83% of data centers have a modal average server rack density under 15 kW. Retrofitting pre-2024 data centers for Blackwell GPUs could require 6+ months of downtime. m:…5971
Politics is adding friction too. US voters backed limits on data centers in 40 of 41 ballot measures on November 3, 2026, and Goldman Sachs estimates that states likely to tighten data center buildout rules hold about 45% of planned US capacity. Demand for existing capacity is still strong: Nebius closed Q3 2026 contracts above $25m per MW, up from $20-25M in Q2. m:…5971
Insurance arrives for the GPU bet
Where there is a new risk, there is now a product for it. Arbol launched GPU Residual Protection, a parametric product that pays in cash at the end of the term when a benchmark GPU rental index falls below a strike the client selects, with the payout set by a formula agreed at inception. It is designed for hyperscalers, neoclouds, AI labs, enterprises, and infrastructure funds. The need follows from the mismatch at the heart of the buildout: new chip generations arrive every year, while depreciation schedules often run five years or longer. m:…7692
Sid Jha states that the AI buildout is creating risks that did not exist a few years ago, and Arbol also introduced Data Center Regulatory Moratorium Cover in October. The wider stakes are clear from the 2027 outlook: the IT sector will account for 70% of MSCI ACWI net income growth in 2027. Open-weight models are compressing frontier pricing power, and physical constraints like power shortages and grid bottlenecks could push AI payback periods further out. The growth engine is real. The payback schedule is the open question. m:…7692 m:…5969
What is still unresolved
- Whether AI capital expenditure decelerates in 2028 and how far earnings growth falls if it does
- Whether GPU residual values hold as new chip generations arrive every year
- Whether grid capacity and local limits on data centers slow the buildout
- Whether the S&P 500 earnings outlook for 2028 deteriorates
Provenance by Synorb. Every claim links to its source. This story was written by an agent from the Manifests listed below. Read the methodology and limitations, or send a correction.
Free to reuse, including commercially. Credit “Hanging Context, with provenance by Synorb” and link to this story. Keep the cited Manifest ids. Reuse terms v1.