Enterprise AI data brief: the spending-vs.-returns chasm
Research brief delivering verified statistics, source assessments, and corrections for the enterprise AI spending-vs.-returns gap. Every major claim checks out directionally, but several specific figures need updating or reattributing. The most important correction: Goldman Sachs’ $527 billion hyperscaler capex estimate is already obsolete.
Gartner’s $2.52 trillion forecast holds up — with a crucial caveat
Gartner’s January 15, 2026 press release confirms worldwide AI spending of $2.52 trillion in 2026, a 44% year-over-year increase from roughly $1.76 trillion in 2025. The trajectory continues to $3.34 trillion by 2027. Spokesperson John-David Lovelock, Distinguished VP Analyst, anchors these figures.
The 41% of global IT spending claim is mathematically valid but requires careful framing. Gartner’s separate February 3, 2026 forecast pegs worldwide IT spending at $6.15 trillion for 2026 (up 10.8%). Dividing $2.52T by $6.15T yields ~41%. However, Gartner never explicitly states this percentage. The 41% represents IT spending that contains an AI component — not money “earmarked for AI.” Much of this is existing software with AI features baked in at higher price points. Third-party interpretations describe it as “AI represents roughly 41% of all new technology dollars,” which is a reasonable but imprecise gloss. The $401 billion “AI foundations” figure is confirmed exactly from the same Gartner press release, referring to incremental infrastructure spending from providers building out AI foundations. This is distinct from total AI infrastructure spending, which Gartner puts at ~$1.37 trillion in 2026.
Gartner’s broader spending breakdown: AI infrastructure leads at $1.37T, followed by AI services at $589 billion, AI software at $452 billion, AI cybersecurity at $51.3 billion, data science/ML platforms at $31.1 billion, and AI models at $26.4 billion. AI-optimized servers alone will see a 49% spending increase in 2026.
Hyperscaler capex: Goldman’s $527 billion is already stale
Goldman Sachs published the $527 billion consensus estimate for 2026 hyperscaler capex on December 18, 2025. But Goldman’s own analysts flagged it as likely too low — noting consensus had undershot actual spending for two consecutive years and suggesting up to $200 billion in upside. By February 2026, actual company guidance confirmed Goldman’s suspicion. Updated figures from Q4 2025 earnings:
| Company | 2026 Capex Guidance |
|---|---|
| Amazon | ~$200 billion |
| Alphabet/Google | $175–185 billion |
| Microsoft | ~$120–145 billion |
| Meta | $115–135 billion |
| Oracle | ~$50 billion |
Bloomberg reported ~$650 billion for the four largest hyperscalers (February 6, 2026). Futurum Group pegged the top five at $660–690 billion. CreditSights estimates roughly 75% (~$450B) is AI-related infrastructure, with the remainder covering general cloud and data center expansion. Goldman’s cumulative forecast of $1.15 trillion in hyperscaler capex from 2025–2027 — more than double the $477 billion spent 2022–2024 — still captures the trajectory well. For the article, cite the $527B as Goldman’s December 2025 consensus, then note the February 2026 company guidance of $635–690B.
BCG and McKinsey ROI data: confirmed with rich additional detail
BCG’s “Build for the Future 2025” report (published September 30, 2025; 1,250+ executives across 68 countries) confirms all three claims precisely. 60% of companies generate no material value from AI, reporting “minimal revenue and cost gains despite substantial investment.” Only 5% qualify as “future-built,” achieving five times the revenue increases and three times the cost reductions of other companies. The remaining 35% are “scalers” beginning to generate value — up 13 percentage points from 2024. Future-built companies also show 1.7× revenue growth, 3.6× three-year TSR, 2.7× ROIC, and 1.6× EBIT margins versus laggards. BCG’s earlier AI Radar 2025 adds a critical insight: leaders focus on an average of 3.5 use cases versus 6.1 for others — depth over breadth.
McKinsey’s State of AI 2025 (1,993 participants across 105 nations, surveyed June–July 2025) confirms that only ~39% of organizations report any measurable EBIT impact from AI, and most of those report less than 5% of EBIT. Only about 6% qualify as “high performers” — organizations reporting 5%+ EBIT impact with “significant” value. Despite 88% of organizations using AI in at least one function (up from 78% in 2024), roughly two-thirds remain stuck in experiment or pilot mode. High performers are 3× more likely to have strong senior leadership engagement and 3× more likely to have redesigned workflows — the single strongest variable McKinsey tested out of 31.
The claim about high-performing organizations investing 20%+ of digital budgets is confirmed: McKinsey found more than one-third of high performers allocate over 20% of digital budgets to AI, making them 5× more likely to make a big AI bet than average organizations (where only ~10% allocate at that level).
Benchmarkit data and the cost misestimation crisis
Benchmarkit’s “2025 State of AI Cost Management” report (September 10, 2025; 372 enterprise organizations, in partnership with Mavvrik) delivers striking findings. 85% of companies miss AI cost forecasts by more than 10%. Drilling deeper: 80% miss AI infrastructure forecasts by more than 25%, 56% miss by 11–25%, and a startling 24% miss by more than 50%. The margin impact is severe: 84% report significant gross margin erosion (>6%) tied to AI workloads, with over 25% of companies seeing margin drops of 16% or more. Large enterprises were just as likely to miss forecasts by wide margins as small companies. Ray Rike, Benchmarkit’s CEO, stated: “These numbers should rattle every finance leader.”
The failure rate picture is even grimmer than the original claims suggest
Multiple independent sources converge on devastating failure rates. Gartner initially predicted 30% of GenAI projects would be abandoned after proof-of-concept by end of 2025 (July 2024), but later revised upward to 50%. Gartner also predicts over 40% of agentic AI projects will be canceled by end of 2027 (June 2025). RAND Corporation found more than 80% of AI projects fail — twice the rate of non-AI IT projects. MIT’s GenAI Divide report (July 2025) claims 95% of GenAI pilots delivered zero measurable P&L impact, though its methodology (52 executive interviews, 6-month measurement window) has been questioned. Forrester’s 2026 predictions add that fewer than one-third of AI decision-makers can tie AI value to financial growth, and only 15% reported an EBITDA lift in the past 12 months. Forrester predicts enterprises will defer 25% of planned AI spend into 2027 as CFOs tighten oversight.
Deloitte’s State of AI in the Enterprise 2026 (3,235 leaders surveyed August–September 2025) found that only 34% are deeply transforming with AI, while 37% use AI at a surface level with little process change. A notable finding: 93% of AI spending goes to data, tech, and infrastructure — not people and change management, while companies that embrace work redesign are 2× more likely to exceed ROI expectations.
Corrections and source-quality summary
- Goldman Sachs $527B: Cite as December 2025 consensus, note actual February 2026 company guidance of $635–690B. Goldman itself predicted upward revision.
- Gartner 41% of IT spend: Frame carefully as “IT spending containing an AI component,” not money earmarked specifically for AI.
- Gartner GenAI project abandonment: Originally predicted 30% by end of 2025; actual/updated figure is 50% — significantly worse than initially forecast.
The strongest, most citable primary sources: Gartner’s January 15, 2026 and February 3, 2026 press releases; BCG’s “Build for the Future 2025” (September 30, 2025); McKinsey’s State of AI 2025; S&P Global’s Voice of the Enterprise survey (March 2025); Benchmarkit’s State of AI Cost Management (September 2025); and Forrester’s 2026 Predictions (October 2025).