Goldman Sachs has projected a staggering $7.6 trillion will be spent on AI infrastructure by 2031, with Nvidia poised to capture an estimated 75% of the compute layer. The investment is expected to cover three key areas: compute, data centres and power.
The report outlines a baseline model that forecasts annual spending on AI infrastructure to reach $765 billion in 2026, scaling up to $1.6 trillion by 2031. This growth is driven by explicit assumptions regarding hardware pricing, depreciation cycles and power availability. Readers can trace exactly how sensitive each projection is to changes in these variables.
Compute takes the largest share of spending at $5.1 trillion, covering AI chips, servers and supporting hardware for training and inference workloads. Goldman uses Nvidia's next-generation Rubin VR200 GPU as its baseline unit price, roughly double what the H100 cost at launch. Data centres account for $2.1 trillion, including construction, cooling upgrades, electrical infrastructure and connectivity.
However, power availability is identified as the bottleneck in this area. Power and energy infrastructure is the smallest line item at $358 billion, yet analysts consider it the single most binding constraint on deployment timelines. Nvidia's structural position in the market means that it will likely capture a significant share of the compute spend.
Goldman estimates that Nvidia will account for approximately 75% of the $5.1 trillion allocated to the compute layer, translating to roughly $3.8 trillion in cumulative revenue through 2031. Combined capex from Meta, Microsoft, Amazon and Alphabet has been raised to $5.3 trillion, up from a prior estimate of $4.5 trillion – an $800 billion upward revision.
The report highlights that power availability is a significant constraint on deployment timelines, with confirmed contracts and planned spending totalling $358 billion. However, this figure represents only part of the actual gap in power availability. AI data centres have already undergone a step-change in power density, with next-generation facilities designing for 500+ kilowatts per rack – a tenfold increase over traditional hyperscale racks.
Liquid cooling is becoming increasingly standard and construction costs have risen from $10 million per megawatt to $15-20 million per megawatt for next-generation facilities. The report notes that nuclear energy is being revisited by corporations, with Vistra Energy signing a 20-year contract with Meta covering over 2,600 megawatts.
Goldman's own caveat highlights the uncertainty in hardware depreciation assumptions, which creates a $1.76 trillion variance in the cumulative spending forecast. This uncertainty has been flagged by Goldman's equity research division, which notes that scale does not guarantee returns.
The report also warns that cloud AI compute pricing will likely trend upward for the next several years unless GPU energy efficiency improves materially or a credible Nvidia alternative emerges at scale. Organizations modelling AI infrastructure budgets should treat compute costs as a dynamic variable rather than a fixed line item.
