Halfway through 2026, the story of artificial intelligence has quietly changed shape. The early narrative, dominated by chatbots and model releases, has given way to a harder question: can enterprises actually make AI pay off, and can the infrastructure underneath it keep up with demand. Investment is still climbing, adoption is broadening well past the early movers, and the competitive landscape is reorganizing itself around compute, inference, and who can prove return on investment fastest. Understanding where each of these threads stands, rather than where the hype cycle says they should be, is what separates a sound 2026 AI strategy from a speculative one.
INVESTMENT IS SHIFTING FROM TRAINING TO INFERENCE
For the last several years, AI capital expenditure was primarily a training story: massive spending on chips and data centers to build ever larger models. That is changing. Brook Dane and Sung Cho, co-heads of the US Mid and Large Cap and Technology Equity businesses at Goldman Sachs Asset Management, returned from their annual Silicon Valley research trip in June with a clear message. Enterprise deployment of AI is accelerating, and the heaviest users are pulling away from everyone else. According to executives they met with, the top 5% of companies by usage are consuming three times the tokens of the median company, and that gap keeps widening.
That shift matters because inference, the process of running a trained model against fresh data to actually perform a task, draws on a different part of the compute stack than training does. As Cho put it, moving from a training-dominated world to an inference-heavy one puts pressure on parts of the infrastructure that previously saw little strain. Dane and Cho describe the current environment as compute constrained in a way that is real and durable, extending beyond GPUs into ASICs, memory chips, and the supply chains behind both. One knock-on effect they flagged is a coming shift in data center wiring, as copper connections give way to fiber optics to keep pace with faster processing speeds, a trend they expect to run for five-plus years.
Investors have also revised an older fear. A year ago the concern was that large language models would cannibalize search advertising. Instead, Dane and Cho report the opposite: AI queries generate more contextual depth about what a user wants, which makes ad placement more valuable rather than less. Combined with growing enterprise appetite for agentic commerce, where AI agents transact on a user’s behalf, the investable AI market looks broader in 2026 than the narrow infrastructure story of a year or two ago.
THE ENTERPRISE ADOPTION CURVE IS MATURING, NOT PLATEAUING
Investment enthusiasm only matters if adoption keeps pace, and the data suggests it is. NVIDIA’s 2026 State of AI research, drawing on more than 3,200 responses across financial services, retail, healthcare, telecommunications, and manufacturing, found that 64% of organizations are now actively using AI in their operations, up from the assessment-heavy posture of prior years. North America leads at 70% active usage, followed by EMEA at 65% and APAC at 63%. Company size remains the clearest predictor of maturity: 76% of large organizations, those with more than 1,000 employees, report active AI usage compared with far lower shares among smaller firms, largely because bigger companies can fund the infrastructure and talent needed to move use cases from pilot to production.
The payoff is showing up on the balance sheet. NVIDIA’s research found 88% of respondents said AI had increased annual revenue in at least part of the business, with 30% reporting gains greater than 10%. A similar 87% reported reduced annual costs. Confidence in that trajectory is feeding directly back into budgets, with 86% of organizations planning to increase AI spending in 2026 and nearly 40% expecting increases of 10% or more.
Agentic AI, systems that reason, plan, and execute multistep tasks with minimal human direction, is the fastest-growing category within that spend. Telecommunications leads adoption at 48%, with retail and consumer packaged goods close behind at 47%. Open source and open weight models are underpinning much of this build-out too, with 85% of respondents calling open source moderately to extremely important to their AI strategy, a preference especially pronounced among smaller companies that would rather fine-tune existing tools than build from scratch.
None of this comes without friction. The same research identifies data readiness as the leading obstacle, cited by 48% of respondents, followed by a shortage of AI experts and data scientists at 38%. Nearly a third of organizations are still stuck in pilot or assessment mode, a reminder that averages can obscure a wide gap between leaders and laggards.
THAT GAP IS SHOWING UP EVEN AMONG THE MOST DIGITAL-NATIVE COMPANIES
A benchmarking study conducted by The Economist for Databricks, surveying more than 1,220 executives across eight industries including 150 leaders at digital native companies, found a similar pattern playing out inside the businesses that would seem best positioned to move fast. Digital native companies report the strongest AI-scaling ambitions of any sector surveyed, and 92% say their AI return on investment is running ahead of plan, compared with 84% across all industries. Yet ambition and breadth of deployment have not translated into full operational maturity. Digital natives lead on fully embedding AI at scale in only one of eight business functions studied, meaning sectors such as telecommunications, media, manufacturing, and energy are, in some respects, further along at institutionalizing AI within specific parts of the business. Scaling AI, in other words, is turning out to be a distinct discipline from experimenting with it, and even the companies built on data are still working out the architecture to close that gap.
ORGANIZATIONAL READINESS IS THE REAL BOTTLENECK
Consulting firm CapTech, writing in a sponsored analysis for Harvard Business Review, argues that AI has broken the usual technology hype cycle entirely. Rather than following the familiar arc toward a trough of disillusionment before recovering, AI keeps advancing month over month, faster than most organizations can absorb it. CapTech’s Brian Bischoff and Bree Basham frame the core 2026 challenge as organizational rather than technical: AI is not overpromised, it is underutilized, and the deciding factor is whether leaders can adapt their processes as fast as the technology evolves.
Their research also points to people, not algorithms, as the biggest variable in whether AI transformation succeeds. Poor change management, employee resistance, and compliance paralysis are named as recurring barriers, and CapTech’s own 2025 consumer survey found 65% of consumers highly concerned about privacy and data security in AI-driven products. The firm’s prescription is to treat AI as a tool for job crafting, letting employees reshape their own roles around it, rather than a blunt instrument for headcount reduction. CapTech also describes a fast-forming prototype economy, where AI compresses product development cycles that once took weeks into a matter of hours, though the same speed introduces what the firm calls a velocity trap, in which teams sacrifice thoughtful design for the sake of moving fast.
THE MACRO PICTURE IS MORE STABLE THAN THE DISPLACEMENT NARRATIVE SUGGESTS
Amid the investment and adoption numbers, a parallel debate has emerged over what all this AI spending means for jobs and the wider economy. Citadel Securities has pushed back directly on the idea of an imminent labor displacement crisis. Drawing on Real Time Population Survey data from the St Louis Fed, the firm’s analysis found no evidence of a non-linear spike in daily AI use for work, describing adoption trends as following the same S-curve that past general purpose technologies, including the internet and personal computers, have followed. Software engineering job postings, often cited as an early casualty of AI, were actually up 11% year over year at the time of the report.
Citadel’s broader argument is that recursive improvement in AI capability does not automatically translate into recursive economic deployment, since displacing large amounts of white-collar labor would require far more compute capacity than currently exists, and rising compute costs create a natural ceiling on substitution. The firm also frames AI-driven automation as a productivity shock rather than a demand shock, pointing to accelerating new business formation and continued strength in labor market data, including a pickup in construction hiring tied to data center buildouts, as evidence that AI is functioning as a complement to labor in most sectors rather than a wholesale replacement for it.
WHAT THIS MEANS HEADING INTO THE SECOND HALF OF 2026
Taken together, these five vantage points, from asset managers, chipmakers, consultants, macro strategists, and enterprise researchers, describe an AI market that is deepening rather than cooling. Capital is following inference and enterprise deployment rather than training alone. Adoption is broad but uneven, with company size and organizational readiness doing more to explain outcomes than the technology itself. And the fear that AI investment is disconnected from real economic value looks, at least so far, less supported by the data than the fear that organizations simply are not ready to absorb what they have already bought. The competitive edge in the second half of 2026 is likely to belong to whichever organizations close that readiness gap first.
References and Further Reading
- Goldman Sachs, AI Investment Is Shifting as Inference, Enterprise Adoption Accelerate (July 2026) — goldmansachs.com/insights/articles/ai-investment-is-shifting-as-inference-enterprise-adoption-accelerate
- Citadel Securities, The 2026 Global Intelligence Crisis, Global Macro Strategy (February 2026) — citadelsecurities.com/news-and-insights/global-macro-strategy/2026-global-intelligence-crisis
- Databricks / The Economist, Making AI Deliver: A Benchmarking Framework on How Leading Companies Operationalise AI for Impact (2026) — databricks.com/resources/analyst-research/making-ai-deliver
- NVIDIA, How AI Is Driving Revenue, Cutting Costs and Boosting Productivity for Every Industry in 2026 (March 2026) — blogs.nvidia.com/blog/state-of-ai-report-2026
- CapTech, Four Trends in AI Experimentation, Adoption, and Transformation, sponsored content in Harvard Business Review (March 2026) — hbr.org/sponsored/2026/03/four-trends-in-ai-experimentation-adoption-and-transformation
