The 2026 recap: what the evidence says about enterprise AI
Nine findings from Gartner, McKinsey, Deloitte, Forbes and J.P. Morgan on what changed in 2026, what it means for 2027, and what executives should do next.
Prashant Bhardwaj · 3 Oct 2026
37%
attribute some EBIT impact to AI, essentially unchanged from 2025
McKinsey
$3.64T
forecast worldwide AI spending in 2027, up from $2.67T in 2026
Gartner
74% vs 21%
expect to use agentic AI at least moderately within two years vs have a mature agent-governance model
attribute some EBIT impact to AI, essentially unchanged from 2025
Survey, n=1,719, 97 nations, fielded 4 May–8 Jun 2026
80%
say AI improved their own productivity
~6%
are 'high performers' (5%+ of EBIT from AI), flat year on year
Individual gains are now near-universal, but enterprise financial impact has not moved in a year. The small group of high performers stands out for redesigning workflows, visible leadership commitment and broader deployment, though the survey cannot isolate a single cause.
Mixed
Broad adoption, narrow returns; pressure on AI budgets to show P&L evidence
of large organisations (>$1B revenue) are scaling AI agents, up from 27%
Smaller organisations flat at 22%
32%
decided against buying software because coding agents let them build it
Agent scaling is concentrating in large enterprises. A third of respondents have already passed on buying at least one software product or feature because agentic coding tools made an in-house build viable.
Mixed
Pressure on point-solution software; more internal engineering and governance load
say AI operating costs, including tokens, have limited their AI use
60% still expect to raise AI investment in the next year
28%
spend more than 10% of enterprise ICT budget on AI
Cost has become a real constraint for about one organisation in five, even as most plan to spend more. The two facts together mean spend discipline, not spend appetite, will separate scaled programmes from stalled ones.
Headwind
Inference cost becomes a budgeting and governance discipline
Gartner Says Worldwide AI Spending Will Total $2.5 Trillion in 2026
15 Jan 2026
$3.64T
forecast worldwide AI spending in 2027, up from $2.67T in 2026
Gartner table, Sept 2026; headline rounds 2026 to $2.7T (+49.5%)
$29.2B → $65.5B
AI agents and assistants spend, 2026 to 2027
+$142.6B
upward revision to the 2026 figure between Jan and Sept (derived from the two tables)
Gartner puts GenAI in the Trough of Disillusionment this year, yet keeps raising spend. Most enterprise AI is arriving through features embedded by incumbent software vendors rather than new transformation projects, despite lock-in, data-sovereignty and cost risks.
Tailwind
Vendor-led AI adoption; infrastructure remains the largest share of spend
expect to use agentic AI at least moderately within two years vs have a mature agent-governance model
23% already use agentic AI at least moderately
73%
rank data privacy and security as their top AI risk
30%
say their risk and governance approach is highly prepared (+6 pts)
Planned agent deployment is running well ahead of governance. The risks leaders worry about most are all governance problems, and readiness on strategy and governance improved only modestly year on year.
Headwind
Regulated sectors face audit and supervisory scrutiny before capability limits
J.P. Morgan Asset Management — AI market theme hub (2Q26)
31 May 2026
~$800B
hyperscaler capex on track for 2026, from over $400B in 2025
Consensus-based; capex ~60% of operating cash flow in 2025, 90–100% in 2026
~$250B
expected hyperscaler bond issuance in 2026, from $121B in 2025
~2/3
of AI compute is now inference rather than training
The buildout is demand-led and increasingly debt-financed. Power, not chips, is described as the binding constraint, and the question for investors is whether returns arrive on the timeline markets expect.
Mixed
Capacity should keep growing; price and availability depend on power and financing conditions
J.P. Morgan Asset Management — AI market theme hub (2Q26)
31 May 2026
~1 in 5
firms have fundamentally redesigned even one workflow; those that did are roughly 3x as likely to see bottom-line impact
J.P. Morgan AM citing McKinsey (Nov 2025)
~50%
rise in AI spend per employee this year, to roughly $2,000 (Atlanta Fed, via J.P. Morgan AM)
J.P. Morgan frames adoption friction as three potholes: people, processes and price. Employees are ahead of their employers on tool use, most firms are speeding up existing tasks rather than redesigning work, and agent costs are hard to budget.
Mixed
Diffusion is slower and more uneven than capability; gains go to firms that redesign work
Forbes 2026 AI 50 List (staff-edited; sponsoring partner Mayfield)
16 Apr 2026
$60B vs $91M
funding listed for Anthropic and for Gamma on the same 2026 list
Gamma is described by Forbes as profitable since 2023 and valued at $2.1B
Forbes' 2026 list places labs funded in the tens of billions beside small vertical players that are already profitable. For buyers, a vendor's size or funding is a weak proxy for viability, and the list works as a watchlist rather than due diligence.
Mixed
Wide capital divergence across the enterprise AI supply chain; consolidation risk for smaller vendors
Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027
25 Jun 2025
>40%
of agentic AI projects predicted to be cancelled by end of 2027
Prediction made June 2025; open until end-2027
~130
of the many 'agentic' vendors Gartner judged to be genuinely agentic
Gartner attributed the expected cancellations to rising costs, unclear value and weak risk controls, and described much current activity as early experiments. This is a standing prediction to test against 2027 outcomes, not a 2026 finding.
Headwind
Shakeout among agent projects and vendors
Method
How we sourced this
—Only Gartner, McKinsey, Deloitte, Forbes staff-edited research and J.P. Morgan are used.
—Every figure was read on the institution's own page and the check date is shown on each card.
—Findings are paraphrased and link to the original.
—“Our view” marks Beyond the Black Box analysis and is never attributed to a source.
—Executive actions are our recommendations, not the sources'.
Planning 2027
Turn these findings into a plan your board can test.