2026 is the year AI has to prove it

The single most-quoted number of 2026 came from PwC's 29th Global CEO Survey, published at Davos in January. Of 4,454 chief executives in 95 countries, 56% said AI had delivered neither higher revenue nor lower costs in the previous twelve months. Only 30% reported any revenue increase, 26% reported lower costs, and 22% said their costs had actually gone up. Just 12%, which PwC calls "the vanguard", achieved both revenue growth and cost reduction at once.

That headline travelled fast. Fortune ran it as "56% of companies getting nothing out of AI", The Register called it "zero payoff from AI splurge", and Futurism said CEOs were "alarmed". PwC's chairman Mohamed Kande framed it differently: the problem is not the technology but the basics. Companies that built strong AI foundations (a Responsible AI framework, an integration-ready technology environment, enterprise-wide deployment tied to strategy) were three times more likely to report meaningful financial returns. Vanguard companies had applied AI to products, services and customer experiences at 44%, against 17% for everyone else.

Capgemini's picture, published the same week, looks almost like a different world. Its "multi-year AI advantage" study found that 66% of organisations report measurable gains in productivity and decision quality from human-AI collaboration, that 38% have already operationalised generative AI use cases, and that AI will rise from about 3% of total business budgets in 2025 to about 5% in 2026. More than half are committing to ongoing investment, many on a five-year horizon. An earlier Capgemini operations study put the average return on AI investment at 1.7x, with two in five expecting positive ROI within one to three years.

The two datasets are not actually in conflict. PwC asked every CEO it could reach, including those who have barely started. Capgemini surveyed 1,505 executives at companies with more than $1 billion in revenue that had already deployed AI at limited or full scale. Capgemini's respondents are, roughly speaking, PwC's top third. Read side by side, they say the same thing: most of the economy has not yet seen returns, but the companies that have crossed the deployment threshold are now budgeting for AI as a permanent line item.

Where they genuinely differ is in tone and in what they forecast. PwC avoids point forecasts and leans on regression-style analysis of what distinguishes winners (a separate PwC analysis found companies applying AI widely to products and customer experience achieved nearly four percentage points higher profit margins). Capgemini publishes dated projections and dollar figures. Both treat the next twelve months as a reckoning: PwC says the "window to capture value is narrowing"; Capgemini warns that "hastily assembled toy agents risk renewing disappointment". ⚠ Capgemini's 1.7x ROI figure comes from a June 2025 study, and a "42% average ROI" quoted by Capgemini Mexico executives in December 2025 could not be verified on a Capgemini page.

An AI agent is a program that is given a goal, breaks it into steps, uses tools and data, and completes the work with limited human involvement. In 2025 the idea went mainstream. In 2026 both firms are asking a harder question: does it work in production?

Capgemini has the most detailed baseline. Its flagship agentic study of 1,500 executives found that only 2% had fully scaled agent deployments, 12% had scaled them partially, 23% were running pilots, and 61% were still exploring. Even so, 93% believed that scaling agents in the next twelve months would give them a competitive edge, and nearly half had no implementation strategy at all. Capgemini sizes the prize at up to $450 billion of economic value by 2028, and projects that scaled implementers will generate about $382 million each over three years compared with $76 million for the rest.

The most striking Capgemini finding is about trust. In a single year, the share of executives who trust fully autonomous agents fell from 43% to 27%. Only 40% trust agents to manage tasks on their own, 71% say they cannot fully trust autonomous agents for enterprise use, and nearly two in five think the risks outweigh the benefits. The firm's reading is that trust fell because companies learned more, not less: the closer agents get to real work, the more executives see what can go wrong.

Yet adoption keeps rising. Capgemini's Top Tech Trends report says the use of AI agents in business operations more than doubled, from 10% in 2024 to 21% in 2025, agentic projects grew 48% in 2025, and 82% of organisations plan to integrate agents by 2027. In financial services, only 10% of banks and insurers have agents at scale but 48% are creating new roles to supervise them. In automotive engineering (Capgemini's most recent study, October 2026), 68% of companies already use or pilot agentic AI for product development, but only 5% describe their AI governance as fully mature.

Why the different postures? Capgemini is a technology services company and its research tracks a market it sells into; it surveys firms already exploring agents and projects their intentions forward. PwC advises boards and audits companies; its instinct is to demand evidence before extrapolating. The useful synthesis is that Capgemini's data describe the pipeline and PwC's advice describes the gate. Both agree on the obstacles: Capgemini finds 80% of organisations lack mature AI infrastructure and fewer than one in five report high data readiness; PwC finds the same foundations gap behind its 56%.

⚠ Capgemini's headline agentic numbers (2% scaled, trust 43%→27%, $450bn) are from fieldwork in April 2025. No general-purpose 2026 update had been published as of 23 October 2026; the 2026 figures above come from sector studies (financial services, automotive) and from the Top Tech Trends report.

PwC's 2026 Global AI Jobs Barometer analysed more than one billion job advertisements across 27 countries. Its central finding is that companies most exposed to AI are growing faster on every measure: productivity growth is about 40% higher (33.5% versus 24.0% since 2018), headcount grew 52% versus 36%, and wages grew 24% versus 17%. The top fifth of AI-exposed companies achieved 163% productivity growth. Job postings asking for AI skills rose 69% in 2024–25, against 9% for all jobs. PwC's conclusion: "far from being a job killer, AI may actually be a job expander".

But the gains are not evenly shared. Workers with AI skills earn a 62% wage premium on average, and that premium varies enormously by sector, from 118% in consumer markets to 16% in the public sector. PwC also finds a "two-track" market: jobs that AI "professionalises" (22% of advertised roles) are growing twice as fast and seeing salaries rise 42% faster than jobs that AI "democratises" (52% of roles). Skills requirements in the most AI-exposed jobs are changing 2.2 times as fast as in the least exposed.

The Barometer also documents what is happening to the career ladder. In the United States, AI-exposed entry-level roles are seven times more likely to demand skills traditionally expected of senior staff. Those "seniorised" junior roles have grown 35% since 2019, while non-seniorised entry-level roles have shrunk 10%. Routine work that used to serve as an apprenticeship is disappearing, and judgement, leadership and adaptability are being asked of people much earlier in their careers.

PwC's Hopes and Fears survey of 49,364 workers in 48 countries, published in September, shows how this feels from the inside. 64% of workers used AI at work in the past year, up ten points; daily generative AI use rose from 14% to 22%. PwC then splits the workforce into four groups by whether a person has scarce skills and whether they have an "AI advantage". Front-runners (14%) have both; AI insurgents (18%) use AI heavily without scarce skills; Indispensables (11%) have scarce skills but little AI engagement; and the "engine room" (56%) has neither.

The contrast between the top and bottom groups is stark. Three-quarters of front-runners trust top management; only a third of the engine room does. 51% of front-runners use generative AI daily; only 11% of the engine room does, and most have never used AI at all. Nearly 80% of front-runners have access to the learning they need; fewer than 40% of the engine room do. Front-runners are also a flight risk: 29% say they are very or extremely likely to change employer within a year.

Across the whole workforce, the capacity to absorb change is weakening. Access to needed learning and development fell from 59% to 51%. Trust in top management fell seven points year on year, to below half. Only 34% can pay their bills with money left over, down eight points. Workers fear economic volatility (57%) more than AI taking over their tasks (44%). PwC's Global Workforce Leader Pete Brown summarised it as "a real risk that the global workforce is starting to move at different speeds".

One PwC finding cuts against a popular narrative. At Davos, Kande himself said AI was breaking the entry-level "apprenticeship model". Five months later the Barometer found that AI-exposed junior roles grew 35% since 2019; what has flatlined is early-career hiring only in the top quartile of AI exposure. The honest reading is that junior jobs are not vanishing but changing shape, and the people who get them are expected to be more senior on day one. ⚠ The Barometer's firm-level findings are correlational and PwC notes survivorship bias in the company data.

In PwC's CEO Survey, the share of chief executives reporting high or extreme exposure to significant financial loss from cyber attack rose from 21% two years ago to 24% last year and 31% in 2026, putting cyber level with macroeconomic volatility as the top threat. 84% of CEOs plan to strengthen enterprise-wide cybersecurity specifically in response to geopolitical risk. Exposure is uneven: 34% of German CEOs say they are highly exposed against 16% in the UK.

PwC's dedicated cyber survey, the Global Digital Trust Insights (3,887 executives, 72 countries, published October 2025 as the 2026 edition), explains why the worry is justified. Only 6% of organisations say they are "very capable" of withstanding attacks across every vulnerability surveyed; roughly half are at best "somewhat capable". Only 24% spend significantly more on proactive measures than on reacting to incidents, which PwC calls the ideal ratio; 67% spend about the same on both. 60% put cyber investment in their top three strategic priorities because of geopolitical uncertainty, and 78% expect budgets to rise. On quantum computing, 49% have not started on quantum-resistant measures, 22% are implementing, and only 3% have implemented all seven measures PwC tracks.

A year on, the 2027 edition (3,934 leaders, published 1 October 2026) shows the threat moving. The number-one threat organisations feel least prepared for is now attacks targeting AI systems, named by 50% of security leaders, ahead of cloud (40%), third-party breach (34%) and ransomware (33%). The AI-enabled attacks they fear most are autonomous botnets (53%), adversarial attacks (52%) and data poisoning (52%). 84% of security and finance leaders expect budgets to rise, and 58% rank AI among their top budget priorities. Data foundations are getting worse, not better: companies have implemented on average three of seven data-risk measures, and only 5% have all seven, down from 7%.

Trust is the second thread. 66% of CEOs experienced stakeholder trust concerns in the past year, about AI safety, data privacy, transparency or climate. PwC's regression-adjusted analysis found that public companies with the fewest trust concerns delivered twelve-month total shareholder returns about nine percentage points higher than those with the most. In the workforce survey, trust in top management fell seven points. PwC's response is to make trust a boardroom topic across three forms: operational, accountability and digital.

On trust in AI specifically, the firms measure different things and reach compatible conclusions. PwC's number is about stakeholders trusting the company; Capgemini's is about executives trusting the machine (trust in fully autonomous agents fell from 43% to 27%; 71% cannot fully trust autonomous agents; only 46% have AI governance policies, and those are seldom followed). Both say the same thing about the remedy: Responsible AI governance is the enabler of value, not a brake on it. Capgemini's two-thirds of executives who want clearer accountability frameworks are asking for what PwC calls "Responsible AI moving from talk to traction".

Quantum is the shared blind spot. PwC finds 49% have not begun post-quantum work in both its 2026 and 2027 editions. Capgemini's 2025 post-quantum study found about two-thirds of organisations see quantum computing as the most critical cyber threat within three to five years, and 70% are assessing or deploying quantum-safe measures, but roughly 30% still downplay the urgency. ⚠ PwC's global Digital Trust Insights figures were read from PwC member-firm pages and a press release because pwc.com blocked direct retrieval; one source gives 60% and another 61% for the "top-3 priority" statistic.

CEO confidence tells the story in one line. The share of CEOs very or extremely confident about revenue growth over the next twelve months fell from 56% in 2022 to 38% in 2025 and 30% in 2026, the lowest in five years. Yet 61% expect global growth to improve, 51% plan international investment, and 41% plan a major acquisition within three years. PwC's reading is that leaders are gloomy about the near term and still betting on reinvention. Its warning is about paralysis: the 15% of "cautious" companies that neither invest nor acquire grow two points slower with margins three points lower.

Tariffs are the new, uneven risk. 20% of CEOs say they are highly or extremely exposed to significant loss from tariffs, but that ranges from 6% in the Middle East to 22% in the US, 28% in mainland China, 30% in Turkey and 35% in Mexico. 29% expect tariffs to reduce their net margin (most by less than 15%), 60% expect little or no change, and 6% expect an improvement. 32% say geopolitical uncertainty makes them less likely to make large new investments. PwC's US C-suite survey in April found 86% of executives now treat tariffs as permanent.

PwC's economists put numbers on the whole picture. The September 2026 Global Economy Watch projections have world growth at 2.4% in 2026 at market exchange rates, rising to 2.6% in 2027, with a two-speed pattern: emerging E7 economies at 4.0% against the G7 at 1.5%. India leads at 6.6%, China 4.6%, the US 2.0%, the Eurozone 0.9%, the UK 1.0%. Inflation re-accelerated in 2026 to 3.4% globally, and both the Federal Reserve (3.75–4.00% after a September hike) and the ECB (2.50%) are expected to hike again within six months, a reversal of the easing cycle. PwC explicitly notes the table does not account for the recent Middle East unrest.

The Middle East is the shock of the year. PwC's US Quarterly Outlook for Q2 2026 is built around the Iran conflict and disruption of the Strait of Hormuz, through which nearly 20% of global daily oil demand and LNG pass. Its baseline cuts US 2026 growth from 2.1% to 1.9% with WTI crude around $80; an adverse scenario has oil at $96–100; a severe scenario at $120–150 would slow global growth to about 2.2% and "likely tip the world into recession". Euro-area growth was cut to around 0.5%. PwC's August CEO snapshot confirms the pass-through: about 70% of CEOs said energy and non-energy costs rose because of global shocks in 2026, and 19% saw energy costs rise more than 10%. Energy, PwC says, is the biggest financial surprise of 2026.

Capital is behaving in a K-shape. PwC's mid-year M&A outlook has global deal value on track for about $4 trillion in 2026, up 13% and the strongest since 2021, while the number of deals falls 13% to about 42,000. Megadeals over $5 billion account for 48% of all value, up from 39% in 2025 and 26% in 2024; strip them out and value is down 4%. Alphabet, Amazon, Meta and Microsoft are expected to spend more than $700 billion on AI infrastructure in 2026 alone, and PwC estimates $5–8 trillion of AI capex will be needed by 2030. Its advice is to plan "as if markets are in a bubble" even if AI proves structural.

Where they differ: PwC quantifies tariff damage as moderate (only 29% see a margin hit, mostly small) and warns against "denial" and paralysis; Capgemini treats sovereignty and reshoring as a multi-trillion investment wave and, candidly, as a market for its services. Where they agree: geopolitics is structural rather than transient, the AI build-out is the main growth engine and also the main energy risk, and the US and India are where both capital (PwC) and production (Capgemini) are heading. ⚠ PwC's Annual Outlook 2026 page shows no publish date; it is treated as December 2025. The "phantom load" survey figures come from a Capgemini report whose full text was not retrieved.

PwC's CEO Survey titles its climate chapter "from climate risk to climate value". 42% of CEOs say they are at least moderately exposed to significant financial loss from climate change, rising to 51% in insurance and 67% in power and utilities. The gap is in process. Only 24% have defined procedures for bringing climate into supply-chain and sourcing decisions, 24% for product design, and 20% for capital allocation including M&A. Fewer than one in three insurance CEOs, the industry most exposed, have climate-informed capital allocation. PwC's finding is that the companies that do have these processes get products to market faster.

PwC's "Value in Motion" research gives the macro frame. Its model suggests AI could make the global economy up to about 15% larger by 2035 than otherwise expected, while physical climate costs could make it nearly 7% smaller than business as usual, and stranded-asset costs from decarbonisation could exceed 3% of global GDP. The climate damage, PwC notes, is largely locked in for the decade regardless of the emissions path; the AI dividend is the swing factor. Its Net Zero Economy Index measured a global decarbonisation rate of just 1.02% in 2023, the lowest since 2011, against the 20.4% a year now needed to stay within 1.5°C. ⚠ The Net Zero Economy Index 2024 is the latest edition found; no 2025 or 2026 update was located.

Capgemini's fifth "A world in balance" study (2,100 executives at 701 organisations, fielded June–July 2026) is the most detailed account of what companies are actually doing. The headline is that sustainability now pays: 69% report positive ROI, up from 49% a year earlier; 64% say it boosted sales (47% in 2025); 67% report lower operating costs, averaging 6%; 74% stronger brand equity. Spending runs at 1.04% of revenue, above the 0.8% planned. 73% cite value creation as a key reason to invest. Capgemini's phrase is that sustainability has become "value-driven" rather than "values-driven".

The reason is resilience. 87% of organisations experienced climate-related supply-chain disruption; a companion study with Sweep found 94% suffered a financial loss from it within 24 months. 71% say access to critical resources (energy, water, materials) now influences decisions more than emissions targets, and 61% expect water scarcity to be a bigger constraint on growth than energy availability within five years. Adaptation is where the money is going: 68% actively prioritise it (56% in 2025), and 83% will raise sustainability or adaptation spend in the next 12–18 months.

The uncomfortable half of Capgemini's report is about execution. 84% have set science-based targets, but only 42% are on track for their 2030 or interim goals. The share of organisations behind on net zero jumped from 1% to 11% in a year, and 29% have postponed their net-zero target, up from 8%. Coverage of Scope 3 (value-chain) emissions measurement fell from 54% to 34%. Capgemini's overall sustainability maturity index dropped from 111 to 105. Only 15% have fully quantified the financial impact of climate disruption, 27% have assessed climate risk across the extended value chain, and 28% have deployed climate-risk analytics. 36% report increased reliance on carbon-intensive energy or materials.

AI sits on both sides of the ledger. 64% use AI for sustainability work, mostly in supply chain (61%) and energy efficiency (52%), and 71% believe generative AI's benefits outweigh its footprint, up from 57%. But 45% say AI has significantly increased their greenhouse-gas emissions, and few are measuring it: 38% track AI energy use, 34% AI carbon, 25% disclose AI water consumption.

The two firms agree on the reframing (sustainability is a business variable, not a reporting exercise), on the measurement gap (PwC: a quarter or fewer have climate processes; Capgemini: 15% have quantified the cost), and on AI's energy footprint as a live constraint. They differ in what they emphasise. PwC confines climate to one chapter of a CEO survey and stresses opportunity and investor appetite; Capgemini devotes a full annual study to it and reports, without softening, that net-zero commitments are being postponed even as adaptation spending rises. Read together: the money is moving from mitigation promises to adaptation and resource security, and the companies that quantify their exposure are the ones making that shift deliberately rather than by default.