The money is real. The readiness is the open question.
The firmest numbers in this story come from a chip company's accounts, not from a survey. Nvidia's cost of revenue, which is what it spends to make and deliver what it sells, was $16.621 billion in fiscal 20241, $32.639 billion in fiscal 20251 and $62.475 billion in fiscal 20261. Each figure is checked against its SEC filing. That is about 3.8 times the fiscal 2024 level in two years. We can say it is a large and fast-growing cost base. The dossier does not contain Nvidia's sales, so we draw no conclusion about profit.
Meanwhile, a survey reported by MIT Technology Review says that within two years 100% of respondents plan to use agentic AI, with 69% expecting to use it widely.2 “Agentic” AI means software that carries out tasks on its own rather than just answering questions. The same report says that in organisations it classes as “data laggards”, AI can reach 30% or less of company data.2
Put together, demand is strong and the supply of usable data is patchy. That is our reading of the material, not a finding of the survey.
A caution about the survey itself
Trust is the point of this publication, so here is the problem. Our system checks claims against their sources and keeps a track record for each publisher. For this MIT Technology Review piece, none of the 11 claims we checked held up (0%).2 That does not prove the survey is wrong. It does mean we cannot treat its percentages as verified, and you should not either.
The briefing behind this piece also cites two further figures: that AI can access only about 45% of company data, and that only around half of organisations trust their agents' decisions. Neither is among the claims we could check, so we treat both as unconfirmed.
What the verified Nvidia figures can and cannot tell you
Nvidia's year-end cash was $7.28 billion in fiscal 20243, $8.589 billion in fiscal 20253 and $10.605 billion in fiscal 20263. Quarterly cash is lumpier. The first quarter of fiscal 2026 showed $15.234 billion3, roughly double the $7.587 billion a year earlier3. The first quarter of fiscal 2027 showed $13.237 billion3, lower than the year before it. Cash moves for many reasons, and the dossier does not say why it fell, so we do not read it as a signal.
We are also not drawing a trend from earnings per share. The filings give $11.93 for fiscal 20244, $2.94 for fiscal 20254 and $4.90 for fiscal 20264. That sequence is not a like-for-like decline, and the dossier does not explain the break between the first two years. Without that, any story we told from it would be a guess.
The dossier links Nvidia to the agent trend through products and customers. It lists NeMo Agent Toolkit and NeMo Guardrails among Nvidia's products, and Mount Sinai Health System, Yum! Brands and the State of Alaska Legislative Affairs Agency among its customers. It also shows Advanced Micro Devices competing with it.5 Whether you already own Nvidia depends on which funds you hold. The dossier contains no index weightings, so we will not guess.
Big companies are buying governance, not just models
Two announcements in July 2026 show that large organisations talk about control as much as capability. On July 22, 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.6 Manulife's Shamus Weiland said: “Our partnership with Microsoft is a critical enabler of Manulife's continued evolution into a truly AI-driven organization.”6
On July 21, Box announced controls for AI agents working on company content. They include agent guardrails, oversight of third-party agent activity, prompt injection detection and access policies based on agent classification.7 Nomura Research Institute's Tatsutoshi Murata said: “As we rapidly advance our utilization of AI agents, we expect Box—which has consistently led the development of security management capabilities for secure collaboration—to provide the administrative features needed to safely leverage this new era of AI.”7
Both are company announcements, which are promotional by nature. Our measured reliability for this wire source is 57% of 4,954 checked claims holding up.6,7 Read them as statements of intent, not proof of results.
Startups are pricing themselves against payroll
A set of young companies profiled by CB Insights define their markets by the cost of human labour rather than the software budget. That is a bigger pool of money to aim at.
Penguin AI's Glenn Herzberg said: “US Healthcare Administration runs about a trillion dollars a year, about a quarter of the total health spend, and the published estimates put around $570 billion of that in work that has no effect on health outcomes.”8 That is the company's own framing, and the dossier does not name the estimates he refers to.
Covecta's Ben Thomas said: “Covecta's total addressable market (TAM) is tens of thousands of financial institutions globally, and for them we are not just disrupting their software budget but their labor budget as well.”9 He says Covecta serves banks, non-bank lenders, building societies, credit unions and private credit firms, currently in the US and UK.9
Maisa AI's David Villalon said: “I define my market like the market of process automation of core business and production tasks at regulated industries.”10 His description stresses work that must be auditable, reproducible and resistant to AI inventing facts.10 That emphasis suggests these buyers care as much about proof as about speed, which is the same governance theme as the big-company deals.
Investors are paying attention too. Emily Man of Primary, which backs Casap, said her firm was drawn to its founders because they had “experienced the pain points of disputes firsthand at their respective large fintech companies.”11 We use that only as a description of how one investor picks founders. It says nothing about Casap's results.
Via News's analysis
Our analysis of this material suggests that the bottleneck is shifting from the AI model to what surrounds it: whether the company's data is reachable, and whether anyone can prove what an agent did. Three things in the dossier point that way. Large buyers are signing governance-heavy deals. Startups stress auditability. The one survey that quantifies the data gap is exactly the one we cannot verify. This is our interpretation. The dossier does not test it against sales or customer results, and none of the profiled startups' revenue appears in it.
What to watch
- Nvidia's next filings. Watch whether cost of revenue keeps climbing and whether cash recovers from its latest quarterly level.
- Evidence that data access improves. A repeat survey with checkable figures would settle what the current one cannot.
- Results from the governance deals. Manulife's Copilot rollout to more than 30,000 employees is a concrete case to follow.
- Revenue from the labour-budget startups. So far the evidence is their own descriptions of the market, not proof that the budget is shifting.
