The agent story in enterprise AI has changed. The question used to be whether companies would adopt agents. Now it is whether their data and their trust in the software can keep up. A report titled "Scaling AI agents with trustworthy data" says that within two years 100% of respondents plan to be using agentic AI, and 69% expect to use it widely.1 The same material, as summarised in our research notes, says AI can reach only about 45% of company data and only around half of organizations trust their agents' decisions.1
A caution on that source. Our measured reliability check on the MIT Technology Review item found that 0% of 11 checked claims held up.1 The 45% and 'around half' figures come from our research summary rather than a verbatim claim, so we present them as the report's reported findings, not as verified data. The direction is clear. The exact percentages are not confirmed.
The data gap is widest where it matters most
The report singles out organizations it categorizes as 'data laggards'. In those companies, AI access to company data falls to 30% or less.1 For practitioners this is the point. An agent that cannot see most of an organization's records is limited by what it can read, not by how good the model is. Every vendor selling agents is therefore also selling a data-plumbing and trust problem. That is why our analysis concludes that demand is running into legacy-data and trust constraints, which favors data-infrastructure and governance vendors.
Incumbents are buying governance, not just models
Two July 2026 announcements show the same shape. Manulife and Microsoft announced 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.2 Shamus Weiland described the partnership as "a critical enabler of Manulife's continued evolution into a truly AI-driven organization."2 The release headline puts governance first: "Manulife Expands Partnership with Microsoft to Accelerate Enterprise AI Governance and Innovation."2
Box announced new controls for AI agents working with enterprise content. They include agent guardrails, oversight of third-party agent activity, prompt injection detection, and access policies based on agent classification.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."3 The dossier also says NRI values Box's multi-vendor support, which lets it switch between AI models, and expects Box's protective layer to keep critical content protected as AI use grows.3
Both announcements come from company press releases. Our measured reliability check puts the wire service that carried them at 57% of 4,954 checked claims holding up.2,3 They are evidence of what vendors and customers say they are doing, not independent proof of results. Siemens and NVIDIA are also named in our research as pursuing a governance-led partnership, but the dossier contains no detail on its terms.
Startups are defining markets by labor spend
The startups in CB Insights' Enterprise AI coverage have a notable habit in common. They size their markets by what customers spend on people, not on software.
Penguin AI's Glenn Herzberg says the company defines its market as administrative labor spend rather than the healthcare IT software budget. He argues that "US Healthcare Administration runs about a trillion dollars a year, about a quarter of the total health spend," and that "published estimates put around $570 billion of that in work that has no effect on health outcomes."4 That is a company's own market-sizing argument. We have not verified the estimates he cites.
Maisa AI's David Villalon defines his market as "process automation of core business and production tasks at regulated industries."5 The dossier adds that these are back-office, operations and finance tasks that must be auditable, reproducible and hallucination resistant.5 Those requirements are essentially the governance problem seen from the vendor's side.
Covecta's Ben Thomas says the company's addressable 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. It serves banks, non-bank lenders, building societies, credit unions and private credit organizations, currently in the US and UK.6
Via News's analysis is that all three pitches depend on the data-access problem above. An agent that takes on a bank's loan workflow, or a hospital's administrative work, can only do it with access to the underlying records.
The compute side: what Nvidia's filings show
Underneath the software story is the hardware supplier. Our verified, SEC-sourced data shows Nvidia's cost of revenue rising from $16.621 billion in fiscal 20247 to $32.639 billion in fiscal 20257 and $62.475 billion in fiscal 2026.7 That is nearly four times the fiscal 2024 level in two years. The quarterly series shows the same pattern. First-quarter cost of revenue was $2.544 billion in Q1 2024,7 $5.638 billion in Q1 2025,7 $17.394 billion in Q1 20267 and $20.458 billion in Q1 2027.7 That is about eight times the Q1 2024 figure. Cost of revenue measures what it costs to produce what is sold, so it is a rough proxy for volume, not a statement about profit.
Cash tells a steadier story. It was $7.28 billion in fiscal 2024,8 $8.589 billion in fiscal 20258 and $10.605 billion in fiscal 2026.8 Quarterly cash peaked at $15.234 billion in Q1 20268 and was $13.237 billion in Q1 2027.8 Earnings per share show a seam we cannot explain from the dossier: $11.93 in fiscal 2024,9 $2.94 in fiscal 20259 and $4.90 in fiscal 2026.9 The fall from the first figure to the second is large. The dossier does not say why, so we do not draw conclusions from the comparison. The fiscal 2025 to fiscal 2026 rise stands on its own.
Do you already own this? The dossier gives no index weightings, so we cannot say how much of any fund Nvidia represents. Check your own fund's holdings.
The entity graph also shows Nvidia present in the governance layer, not just the chip layer. It develops NeMo Guardrails, the NeMo Agent Toolkit and NeMo microservices.10 Mount Sinai Health System, the State of Alaska Legislative Affairs Agency and Yum! Brands are recorded as customers, and Advanced Micro Devices is recorded as a competitor.10
What to watch
- Whether the data-access figures hold. The 45% and half-trust numbers come from a source whose claims did not verify. Look for a second source.
- Governance as a product line. Manulife's Agent 365 deployment and Box's new agent controls are early tests of whether customers pay for oversight as a separate layer.
- Labor-budget claims. Penguin, Maisa and Covecta each assert a labor-spend market. Watch for customer evidence of budgets actually shifting.
- Nvidia's next quarterly filing. Cost of revenue and cash are the two lines this piece tracks.