Reading time is short on purpose: the European angle in this material is thinner than the headline topic suggests, and we would rather say so than stretch it. Three things in the record are named and checkable. They are a Canadian insurer's five-year deal with Microsoft, a set of agent-security controls from Box, and a run of Nvidia financial filings. The headline survey numbers are the weakest part of the picture, and we start there.
The headline survey: a strong claim from a weak-scoring source
A report titled "Scaling AI agents with trustworthy data" states that within two years 100% of respondents plan to be using agentic AI, with 69% expecting to use it widely.1 The same report says that in organisations it classes as 'data laggards', AI access to company data falls to 30% or less.1 The topic summary we were given adds that only about 45% of company data is AI-accessible on average and that about half of organisations trust their agents' decisions. We do not hold a verbatim passage for those two figures, so treat them as reported, not confirmed.
Here is the seam. Our measured reliability check on this source found that 0% of the 11 claims we checked held up.1 That does not prove the numbers are wrong. It means we could not confirm them, and a figure like "100% of respondents" should make any reader ask who was surveyed. The dossier does not say, and we will not guess.
The central idea is still reasonable: an AI agent can only be as useful as the data it can reach and trust. We cannot tell you how large the gap is.
The European angle: little here, and what there is
The dossier contains no EU regulation, no EU company with a verified fact attached, and no European market data. The summary mentions Siemens and Nvidia deepening ties, but we hold no detail on that partnership, so we make no claim about it. David Villalon of Maisa AI defines his market as process automation of core business and production tasks at regulated industries, covering back office, operations and finance, work that must be auditable, reproducible and hallucination resistant.2 That is a description of what regulated customers demand. It is not evidence about Europe, and the dossier does not place Maisa in any country.
Covecta, a banking-focused AI company, says it serves corporate and commercial banks, specialist non-bank lenders, building societies, credit unions and private credit firms, currently across the US and the UK, with aspirations to move globally.3 The UK is outside the EU, and no EU country appears in that list. For a European reader the fair reading is that this vertical AI wave is, on this evidence, still aimed elsewhere.
Big companies are buying governance, not just models
The most concrete deal in the record is Manulife and Microsoft. The two announced on 22 July 2026 a renewal and expansion of their relationship. It is a five-year agreement. Manulife will adopt Microsoft's Frontier Suite, deploy Microsoft Agent 365, and expand Microsoft 365 Copilot to more than 30,000 employees.4 Shamus Weiland said: "Our partnership with Microsoft is a critical enabler of Manulife's continued evolution into a truly AI-driven organization."4 The release is titled as a push for AI governance as well as innovation. Note that this is a company press release, so the framing is Manulife's and Microsoft's own.
Box made a similar point on 21 July 2026. It announced agent guardrails, oversight of third-party agent activity, prompt injection detection and access policies based on agent classification.5 Tatsutoshi Murata of Nomura Research Institute said the firm expects Box to provide the administrative features needed to safely leverage AI agents.5 He also said it values Box's multi-vendor support, which lets it switch between AI models.5 Both sources carry a measured reliability score: 57% of 4,954 checked claims from this news feed held up.4,5 That is better than a coin flip, but not enough to take a press release at face value without a second source.
Our reading, which is Via News's analysis and not a sourced fact: the money in these two announcements goes to control, meaning who can see what an agent does and how it is kept in bounds. That fits the data-trust bottleneck above. We cannot show from this dossier that one caused the other.
The money in the substrate: what Nvidia's filings show
Nvidia supplies the hardware layer for much of this. Its filings give us checked numbers, though the dossier carries cash and cost figures but no revenue, so this is a partial picture.
Quarterly cash rose from $5.079 billion in Q1 20246 to $15.234 billion in Q1 2026,7 roughly triple, before easing to $13.237 billion in Q1 2027.8 Annual cost of revenue, the direct cost of producing what Nvidia sells, went from $16.621 billion in FY 20249 to $32.639 billion in FY 202510 and $62.475 billion in FY 2026.11 That is about 3.8 times in two years. A cost line rising that fast generally means a business selling far more, but cost alone does not tell us profit.
Earnings per share are harder to read. They were $11.93 in FY 2024,12 $2.94 in FY 202513 and $4.90 in FY 2026.14 The dossier does not explain the drop between the first two years, so do not compare these three figures as like for like until the share count is checked. We will not guess at the cause.
If you hold a broad index fund, you may own Nvidia indirectly. The dossier gives no weighting, so we cannot say how much.
The entity graph links Nvidia to named customers, including Mount Sinai Health System, the State of Alaska Legislative Affairs Agency and Yum! Brands. It also lists agent-related Nvidia products such as NeMo Guardrails and NeMo Agent Toolkit, and names AMD as a competitor. These are relationships, not revenue figures.
Start-ups are pricing themselves against labour, not software
The vertical players size their markets by what customers spend on people. Ben Thomas of Covecta says its market is tens of thousands of financial institutions, and that "we are not just disrupting their software budget but their labor budget as well."3 Glenn Herzberg of Penguin AI says US healthcare administration runs about a trillion dollars a year, about a quarter of total health spend, and that published estimates put around $570 billion of it in work with no effect on health outcomes.15 That is Penguin's own framing and an estimate it attributes to published sources, not a figure we verified.
The pitch is that automating administrative work is a larger prize than selling software. It also means the buyer's staff are the target. For European firms with strong labour protections, that is a question the dossier does not touch.
Money is also moving. Latitude, a payments platform founded by ex-Stripe crypto team member Cyril Mathew, raised a $35 million Series A on 10 September 2026 for a stablecoin-to-local-currency payments platform.16 CB Insights published a run of founder interviews in late September from Shepherd, Arlo, LARX, Veridox and Binary World, all as part of its enterprise AI coverage.17 Those are interviews, not funding or revenue evidence.
The chips keep coming
The infrastructure build-out is scheduled well ahead. Alibaba's T-Head Zhenwu V900 AI chip is set for commercial release in Q3 2027, and the J900 in Q3 2028.18 Those are scheduled dates, not deliveries, and the dossier gives no performance specifications.
What to watch
- The survey's method. Until the 100% and 45% figures can be traced to a sample, treat them as unconfirmed.
- Any European deployment of the governance tools above. That would be the first real European evidence in this story.
- Nvidia's cash and cost lines. Cash eased between Q1 2026 and Q1 2027 while costs climbed, and we hold no revenue to put against them.
- The 2027 and 2028 chip dates. Slippage would be the signal.
