The gap between ambition and plumbing
The headline adoption number is unambiguous: within two years, 100% of surveyed organizations plan to use agentic AI, and 69% expect to use it widely.1 The constraint is data. On average only about 45% of company data is accessible to AI, and in organizations categorized as 'data laggards' that falls to 30% or less.1 Only about half of organizations trust their agents' decisions.1 For practitioners, the bottleneck in this picture is access to trustworthy, structured data rather than model capability.
A caution about the source. The report, MIT Technology Review's "Scaling AI agents with trustworthy data," has a measured reliability record of 0% of 11 checked claims holding up.1 We are not saying the figures are wrong. We are saying nobody has yet confirmed them against an independent source, so read them as a direction of travel and not as settled measurements. The pattern is plausible and consistent with the deal activity below, but the exact percentages should be treated as unverified.
Incumbents are buying governance, not just models
The big-company announcements in the dossier are about control as much as capability. Manulife and Microsoft announced a five-year agreement under which Manulife adopts Microsoft's Frontier Suite, deploys Microsoft Agent 365, and expands Microsoft 365 Copilot to more than 30,000 employees.2 Shamus Weiland called the partnership "a critical enabler of Manulife's continued evolution into a truly AI-driven organization."2 The announcement's own title leads with "Enterprise AI Governance."2
Box's July 21 announcement points the same way. It introduced agent guardrails, third-party agent activity oversight, prompt injection detection, and agent classification-based access policies.3 Nomura Research Institute's Tatsutoshi Murata said the company expects Box "to provide the administrative features needed to safely leverage this new era of AI."4 He also valued Box's multi-vendor support, which lets NRI switch flexibly between AI models.4 The customer concern in both cases is how to let agents touch company content without losing control of it, which is the same trust gap the survey describes.
Both press releases come from NewsEOD, whose measured reliability is 57% of 4,954 checked claims holding up.2 They are company announcements, so they show what the companies want to be seen doing and not how well it works. Siemens and NVIDIA are also reported to be deepening ties, but the dossier gives no detail beyond that.5
Startups are pricing against labor, not software
CB Insights' Enterprise AI interviews show how vertical players size their opportunity. Covecta's Ben Thomas says its market is "tens of thousands of financial institutions globally," and that "we are not just disrupting their software budget but their labor budget as well."6 Covecta deploys what it calls seasoned banker agents for workflows and portfolio activities, serving banks, non-bank lenders, building societies, credit unions and private credit firms in the US and UK.6
Penguin AI makes the same move in healthcare. Glenn Herzberg said the company defines its market as administrative labor spend, not the healthcare IT software budget. He put US healthcare administration at about a trillion dollars a year, about a quarter of total health spend, and said published estimates put around $570 billion of that in work that has no effect on health outcomes.7 Those are the company's claims and estimates it cites, not figures we have checked.
Maisa AI's David Villalon describes a similar target: process automation of core business tasks at regulated industries, where work must be auditable, reproducible and hallucination resistant.8 The requirement that agents be auditable echoes the governance demands from Manulife and Box. It also suggests that vertical agents will be judged on how well their work can be checked, as well as on how much work they do.
Funding and attention are flowing in. Latitude, a payments platform founded by an ex-Stripe crypto team member, raised a $35M Series A on September 10, 2026 for stablecoin-to-local-currency payments.9 CB Insights also published interviews on September 22 and 24 with executives at Veridox, Binary World, Shepherd, Arlo and LARX.10 Those are interviews, not evidence of traction, and the dossier contains no revenue or customer data for any of these startups.
The compute layer: Nvidia's checked numbers
The one area where our evidence is checked against SEC filings is Nvidia, the company behind the NeMo Agent Toolkit, NeMo Guardrails and the Nemotron 3 Super model, according to our entity graph.11 The same graph lists Mount Sinai Health System, the State of Alaska Legislative Affairs Agency and Yum! Brands as Nvidia customers, and AMD as a competitor.11
Nvidia's cost of revenue, which is what it spends to deliver its products, was $16.621 billion in fiscal 2024, $32.639 billion in fiscal 2025 and $62.475 billion in fiscal 2026.12 That is roughly 3.8 times the fiscal 2024 level in two years. For scale, $62.475 billion is only the cost side of one company's income statement, before any profit. Cost of revenue is a volume signal, so it indicates how much hardware and related product is being shipped, but it says nothing on its own about profit.
Earnings per share reads less cleanly: $11.93 in fiscal 2024, $2.94 in fiscal 2025 and $4.90 in fiscal 2026.13 The dossier does not explain the swing, so we do not interpret it. The figures are checked as reported, but the three years should not be compared directly without knowing how the share count changed. Nvidia's year-end cash was $7.28 billion in fiscal 2024, $8.589 billion in fiscal 2025 and $10.605 billion in fiscal 2026.14 The latest quarter in the dossier, Q1 fiscal 2027, shows $13.237 billion, below the $15.234 billion at the end of Q1 fiscal 2026.14 Quarter-to-quarter cash moves for many reasons, and the dossier does not say why.
Chip supply is also being planned well ahead. Alibaba's T-Head Zhenwu V900 AI chip is scheduled for commercial release in Q3 2027.15 A scheduled release is a plan, not a shipment.
What we make of it
Via News's analysis of the data suggests the story is shifting from model capability to data readiness and governance. The evidence for this is the pairing of an unverified survey finding, which says data access is the bottleneck, with two verified-as-announced deals that center on oversight. Startups in finance and healthcare are pricing themselves against the labor they replace. That is a bigger market than software budgets, and a bolder claim to make.
What to watch
- Whether independent surveys reproduce the 45% AI-accessible-data figure, given that the original source has no verified claims so far.
- Whether Nvidia's cost of revenue keeps climbing in the quarters after Q1 fiscal 2027, and what drives the fall in quarter-end cash.
- Whether vertical agent companies such as Covecta, Penguin AI and Maisa publish customer or revenue evidence to back their labor-budget sizing.
- Whether Manulife's 30,000-employee Copilot expansion produces any reported results.
- Whether the Zhenwu V900 launches on its Q3 2027 schedule.