Trillions of dollars now flow toward artificial intelligence infrastructure. Goldman Sachs Research lays out the numbers in stark terms. Its baseline projection calls for $765 billion in annual AI capital spending in 2026. That figure climbs to $1.6 trillion by 2031. Add it up and the cumulative total reaches roughly $7.6 trillion between 2026 and 2031. The outlays cover specialized chips, data centers and power generation. Yet so far the economic payoff looks thin.
Chief economist Jan Hatzius delivered the blunt assessment earlier this year. Massive investment in AI contributed “basically zero” to U.S. economic growth last year, he said in remarks covered by Yahoo Finance. The spending boosted suppliers in Taiwan and South Korea far more than domestic output. And. This disconnect raises pointed questions for technology executives, investors and policy makers who bet heavily on an imminent productivity surge.
Consensus forecasts for 2026 hyperscaler capital expenditure have already climbed to $527 billion. That mark sits well above earlier projections and reflects repeated upward revisions. Actual spending outpaced analyst expectations in each of the past two years. Growth exceeded 50 percent when forecasts called for roughly 20 percent. Strong balance sheets at the largest cloud providers support further increases. Still, the timing of any slowdown carries risk for valuations, Goldman analysts noted in their December 2025 report Why AI Companies May Invest More than $500 Billion in 2026.
But the bigger picture extends beyond hyperscalers. Goldmanās May 2026 analysis Tracking Trillions: The Assumptions Shaping the Scale of the AI Build-Out shifts attention from demand justification to supply-side realities. The $7.6 trillion cumulative estimate rests on four core assumptions. Economic useful life of AI silicon. Cost and complexity of next-generation data centers. Choices in chip architecture. And delays caused by physical bottlenecks.
Small shifts in any of those variables swing the total by hundreds of billions. Extend average silicon life from four years to six and replacement cycles drop sharply. Shorten it and spending jumps. Data center construction now runs $15 million to $20 million per megawatt. That compares with $10 million in earlier builds. Higher power densities demand advanced cooling and integrated systems. Even modest cost increases compound across thousands of facilities.
Power infrastructure adds another layer. New capacity carries a price tag around $2,500 per kilowatt. Queues for grid connections stretch long. Permitting hurdles slow deployment. Labor shortages and equipment lead times create further friction. In the base case these bottlenecks stretch the timeline without cutting the overall investment sum. In stress scenarios they feed back into corporate decisions and potentially shrink plans.
Chip architecture introduces its own dynamics. The baseline assumes NVIDIA maintains about 75 percent of compute spending. A move toward application-specific integrated circuits could lower per-unit costs. Whether that reduction trims total capital outlays or simply redistributes margins depends on demand elasticity. Goldman models the elastic case in its central projection.
Recent company earnings and market commentary reinforce the scale. Hyperscalers guided toward combined 2026 capital expenditure above $700 billion in some projections. That pace echoes peaks of prior technology cycles when spending reached 1.5 percent of GDP or more. Current AI-related capital expenditure sits near 0.8 percent of GDP. Room exists for further expansion. Yet corporate America shows mixed signals on returns.
Productivity data tell a cautious story. Goldman found no meaningful economy-wide link between AI adoption and productivity gains as of early 2026. Management teams that measured specific tasks reported median improvements around 30 percent. Those localized wins have not yet scaled. A 2023 Goldman forecast anticipated measurable GDP and labor productivity effects starting in 2027. That timeline still holds inside the bank. Widespread adoption could lift annual U.S. productivity growth by 1.5 percentage points over a decade.
Full incorporation of generative AI might raise labor productivity levels by roughly 15 percent in developed markets. The same analysis points to displacement of 6 to 7 percent of the U.S. workforce during the transition. New roles created by the technology could offset losses over time. Net employment effects so far show a modest drag. AI trimmed monthly U.S. payroll growth by about 16,000 jobs over the past year while augmenting certain positions added back 9,000.
Investors have started to rotate. Infrastructure plays that fund heavy capital programs through debt have given ground. Platform companies offering databases, development tools and software layers have outperformed. Productivity beneficiaries, those firms with high labor costs and exposure to automation, still trade at discounts. Uncertainty over the size and arrival of earnings benefits explains much of the hesitation.
Shareholder returns reflect the shift too. S&P 500 capital expenditure is projected to grow 33 percent in 2026 while gross buybacks expand just 3 percent. The pattern marks a clear pivot from returning cash to building future capacity. Big technology names show even sharper moves. Their combined capital plans point to an 83 percent year-over-year jump in some estimates.
Power constraints could cap ambition. Goldmanās chief information officer Marco Argenti highlighted the gigawatt ceiling in a January 2026 outlook. Access to the utility grid now determines scaling speed as much as available capital. Data center announcements and construction pipelines have swelled. More than 3,400 facilities are planned or underway in the United States alone. Many target the demands of agentic systems that operate continuously rather than on demand.
Those always-on workloads change the equation. Persistent compute, 24-hour cooling and heightened security requirements drive fresh infrastructure needs. Goldman Asset Management analysts described the shift as a total rebuild of outdated systems. They see the move toward autonomous AI as the dominant force behind more than 90 percent of future digital infrastructure demand.
Non-hardware spending adds another dimension often overlooked in headline figures. Intangible investments in workflow redesign, software integration and organizational change may match the scale of physical capital outlays. Goldman economists argue that true costs and eventual benefits both exceed current capex narratives. The combined spending could reshape corporate balance sheets for years.
Market sentiment shows signs of fatigue. One Goldman strategist recently described the AI-fueled rally as in danger of becoming “one big trade.” Concentration in a handful of names has grown. Positioning in semiconductors sits at extremes while software valuations lag. Any disappointment in near-term returns could amplify volatility.
Policy questions loom as well. Displacement effects, even if temporary, will test labor markets and safety-net programs. Gains in productivity, when they arrive, tend to lift wages only after a lag. The interim period could prove politically charged. Tax treatment of capital investment, energy policy and trade rules on semiconductors all influence the pace and location of this buildout.
History offers perspective. Past technology waves, from railroads to the internet, delivered large societal benefits only after massive upfront spending and frequent forecast misses. AI may follow a similar path. The difference lies in speed. Model capabilities improve quarterly. Hardware refreshes arrive annually. Economic obsolescence now competes with physical depreciation as the binding constraint on asset lives.
Executives face a high-stakes calibration exercise. Spend too little and fall behind on capability. Spend too much and erode returns for years. The Goldman framework supplies a quantitative map for that decision. Its sensitivity analysis shows how different assumptions alter the trillion-dollar totals. Companies that master the interplay of silicon longevity, power access and organizational change stand to capture disproportionate gains.
For now the investment wave rolls forward. Consensus keeps rising. Project pipelines expand. Productivity metrics remain stubbornly flat at the macro level. The gap between spending and measured output defines the current chapter of the AI story. How and when that gap closes will determine whether the $7.6 trillion bet delivers the widely anticipated economic transformation or joins earlier cycles remembered more for their costs than their lasting returns.
WebProNews is an iEntry Publication