Goldman lifts AI capex to $800bn

Published May 27, 2026 by The Daily Scout

- Goldman Sachs raised its 2026 U.S. business-investment forecast on May 26, citing AI infrastructure spending that it said is tracking above $800 billion. - Goldman said AI spending was running at an annualized $650 billion in the first quarter and broadened beyond chips into servers, power and data centers. - Goldman’s May 1 AI build-out framework projected roughly $7.6 trillion of cumulative AI capital spending from 2026 through 2031.

Why it matters

Goldman Sachs has pushed the AI spending debate into a bigger arena. In a note published Monday, the bank raised its 2026 forecast for U.S. business investment growth to 7.8% from 6.5%, saying AI-related spending was running at an annualized $650 billion in the first quarter and was on track to exceed $800 billion by year-end. That figure matters because Goldman is not describing a narrow chip cycle. The bank said the spending wave now spans semiconductors, servers, memory, power infrastructure, data centers, software and research and development. ### Where is the $800 billion actually going? Goldman said the build-out is spreading across multiple layers of infrastructure rather than concentrating only in graphics processors. (benzinga.com) Its May 26 note tied the higher forecast to equipment and structures investment, with AI demand supporting spending on servers, storage, data-center construction and power systems as well as chips. (newsbreak.com) Goldman’s own longer-range work points the same way. In a May 1 analysis, the firm’s Global Institute and Global Investment Research teams estimated a baseline of about $7.6 trillion in aggregate AI capital expenditure from 2026 through 2031 across compute, data centers and power. ### Why did Goldman change its forecast now? (benzinga.com) The bank said first-quarter spending was already running at a pace that justified a higher call. Benzinga, citing the Goldman note, reported that AI-related outlays were annualizing at roughly $650 billion in the first quarter, enough to support a forecast of more than $800 billion by year-end. (goldmansachs.com) Other recent Goldman research had already been moving in that direction. In a December 2025 article, Goldman said consensus estimates for hyperscaler capital spending in 2026 had risen to $527 billion from $465 billion earlier in the earnings season, and wrote that spending constraints were more likely to come from supply bottlenecks or investor appetite than from balance-sheet capacity. (benzinga.com) ### Is this a macro call or a tech-sector call? Goldman framed the change as a macro investment call. The bank raised its forecast for overall U.S. business investment growth, not just technology-sector capital expenditure, and said AI spending would continue to support equipment and structures investment in 2026. That framing puts AI alongside other economy-wide capex drivers. (goldmansachs.com) MarketDash, summarizing the note, said Goldman expected AI infrastructure spending to add about 3.3 percentage points to capex growth this year, while tax incentives and easing manufacturing headwinds also supported the outlook. ### What does the forecast change for executives? (benzinga.com) The new number shifts the practical question from which model wins to which parts of the stack are being funded. Goldman’s May 1 paper said the size of the AI investment wave is not a single fixed number and depends on assumptions across compute, data centers and power, underscoring that capital is being allocated across interlocking bottlenecks rather than one product category. (marketdash.com) That means return-on-investment questions will increasingly attach to infrastructure layers, timelines and utilization. Goldman’s earlier work on hyperscaler spending said investors were already becoming more selective about AI stocks even as capex projections kept rising. ### What comes next after this forecast? Goldman’s next proof points will come from company capex disclosures, hyperscaler earnings and infrastructure order trends. (goldmansachs.com) The firm’s May 1 build-out analysis said its 2026-2031 estimate rests on assumptions that are subject to change, with compute, data-center and power spending all capable of shifting the total. (goldmansachs.com) The immediate benchmark is whether quarterly spending continues to support Goldman’s annualized first-quarter pace. If it does, the bank’s more-than-$800 billion year-end figure will be tested against reported spending by the largest AI infrastructure buyers over the rest of 2026. (benzinga.com) (goldmansachs.com)

Key numbers

business-investment forecast on May 26, citing AI infrastructure spending that it said is tracking above $800 billion.
Goldman said AI spending was running at an annualized $650 billion in the first quarter and broadened beyond chips into servers, power and data centers.
Goldman’s May 1 AI build-out framework projected roughly $7.6 trillion of cumulative AI capital spending from 2026 through 2031.
In a note published Monday, the bank raised its 2026 forecast for U.S.

What happens next

(benzinga.com) Its May 26 note tied the higher forecast to equipment and structures investment, with AI demand supporting spending on servers, storage, data-center construction and power systems as well as chips.
In a May 1 analysis, the firm’s Global Institute and Global Investment Research teams estimated a baseline of about $7.6 trillion in aggregate AI capital expenditure from 2026 through 2031 across compute, data centers and power.
(goldmansachs.com) MarketDash, summarizing the note, said Goldman expected AI infrastructure spending to add about 3.3 percentage points to capex growth this year, while tax incentives and easing manufacturing headwinds also supported the outlook.

Sources

Quick answers

What happened in Goldman lifts AI capex to $800bn?

Goldman Sachs raised its 2026 U.S. business-investment forecast on May 26, citing AI infrastructure spending that it said is tracking above $800 billion. Goldman said AI spending was running at an annualized $650 billion in the first quarter and broadened beyond chips into servers, power and data centers. Goldman’s May 1 AI build-out framework projected roughly $7.6 trillion of cumulative AI capital spending from 2026 through 2031.

Why does Goldman lifts AI capex to $800bn matter?

Goldman Sachs has pushed the AI spending debate into a bigger arena. In a note published Monday, the bank raised its 2026 forecast for U.S. business investment growth to 7.8% from 6.5%, saying AI-related spending was running at an annualized $650 billion in the first quarter and was on track to exceed $800 billion by year-end. That figure matters because Goldman is not describing a narrow chip cycle. The bank said the spending wave now spans semiconductors, servers, memory, power infrastructure, data centers, software and research and development. Where is the $800 billion actually going? Goldman said the build-out is spreading across multiple layers of infrastructure rather than concentrating only in graphics processors. (benzinga.com) Its May 26 note tied the higher forecast to equipment and structures investment, with AI demand supporting spending on servers, storage, data-center construction and power systems as well as chips. (newsbreak.com) Goldman’s own longer-range work points the same way. In a May 1 analysis, the firm’s Global Institute and Global Investment Research teams estimated a baseline of about $7.6 trillion in aggregate AI capital expenditure from 2026 through 2031 across compute, data centers and power. Why did Goldman change its forecast now? (benzinga.com) The bank said first-quarter spending was already running at a pace that justified a higher call. Benzinga, citing the Goldman note, reported that AI-related outlays were annualizing at roughly $650 billion in the first quarter, enough to support a forecast of more than $800 billion by year-end. (goldmansachs.com) Other recent Goldman research had already been moving in that direction. In a December 2025 article, Goldman said consensus estimates for hyperscaler capital spending in 2026 had risen to $527 billion from $465 billion earlier in the earnings season, and wrote that spending constraints were more likely to come from supply bottlenecks or investor appetite than from balance-sheet capacity. (benzinga.com) Is this a macro call or a tech-sector call? Goldman framed the change as a macro investment call. The bank raised its forecast for overall U.S. business investment growth, not just technology-sector capital expenditure, and said AI spending would continue to support equipment and structures investment in 2026. That framing puts AI alongside other economy-wide capex drivers. (goldmansachs.com) MarketDash, summarizing the note, said Goldman expected AI infrastructure spending to add about 3.3 percentage points to capex growth this year, while tax incentives and easing manufacturing headwinds also supported the outlook. What does the forecast change for executives? (benzinga.com) The new number shifts the practical question from which model wins to which parts of the stack are being funded. Goldman’s May 1 paper said the size of the AI investment wave is not a single fixed number and depends on assumptions across compute, data centers and power, underscoring that capital is being allocated across interlocking bottlenecks rather than one product category. (marketdash.com) That means return-on-investment questions will increasingly attach to infrastructure layers, timelines and utilization. Goldman’s earlier work on hyperscaler spending said investors were already becoming more selective about AI stocks even as capex projections kept rising. What comes next after this forecast? Goldman’s next proof points will come from company capex disclosures, hyperscaler earnings and infrastructure order trends. (goldmansachs.com) The firm’s May 1 build-out analysis said its 2026-2031 estimate rests on assumptions that are subject to change, with compute, data-center and power spending all capable of shifting the total. (goldmansachs.com) The immediate benchmark is whether quarterly spending continues to support Goldman’s annualized first-quarter pace. If it does, the bank’s more-than-$800 billion year-end figure will be tested against reported spending by the largest AI infrastructure buyers over the rest of 2026. (benzinga.com) (goldmansachs.com)