Nvidia Lines Up Wall Street Finance, and OpenAI Moves Into Microsoft 365
Updated: 6 days ago
Title: Nvidia Lines Up Wall Street Finance, and OpenAI Moves Into Microsoft 365
Date: 11 August 2026
Type: Blog
Author: SAASiQ (contact@saasiq.ai)
Word count: 1180 words
Reading time: 5 min
Published: 11-08-2026
Nvidia has signed memorandums of understanding with six of the largest asset managers and banks to set up financing platforms for its customers, with the aim of mobilising more than $500 billion of third-party capital over time. Closer to most organisations' day-to-day, OpenAI-operated models are now switched on by default inside Microsoft 365 Copilot, and OpenAI has changed how it sells and funds ChatGPT.
What Nvidia announced
The six are Apollo Global Management, Blackstone, BlackRock, Brookfield Asset Management, Goldman Sachs and KKR. Nvidia announced the agreements on Monday. The platforms are meant to lend to the companies building AI data centres and buying Nvidia hardware, which in practice means the hyperscalers, the frontier AI labs and a growing number of large enterprises.
According to Fortune, the financing is mostly debt, secured against the compute itself and arranged through private offerings and special-purpose entities, with Goldman Sachs as lead bookrunner. Jensen Huang said Nvidia may take up to 25 per cent of an individual opportunity. Larry Fink of BlackRock pointed investors to the credit quality and the yields on offer. Detail on timing and structure is thin, and Nvidia has cautioned that preliminary arrangements of this kind may not lead to definitive agreements.
How the lending works
Using GPUs as collateral puts compute in the same category as the property and infrastructure assets these firms already know how to lend against. Felix Wang of Hedgeye Risk Management summarised the effect to Fortune: the arrangement makes Nvidia's product cheaper without cutting the price of the chips. For a buyer, the change is in how a build is paid for, with a large upfront capital cost spread over a loan term.
The loan is only as sound as the collateral. Nvidia moved to an annual release cycle for its data centre chips in 2024, so hardware bought this year will be two generations old before a multi-year loan against it is repaid. Lenders will have to allow for that in how they value the chips, and in the rates they charge.
For most organisations none of this is a direct purchase. It reaches them through the cloud providers and software vendors whose data centres it pays for, and through the prices those providers charge for AI capacity over the next few years.
OpenAI inside Microsoft 365
The story most likely to need action this week is a setting in the Microsoft 365 admin centre. Microsoft added OpenAI to its list of subprocessors on 23 June and made OpenAI-operated models available inside Microsoft 365 Copilot and Copilot Studio from 9 July. From 24 July those models were switched on for all users at eligible commercial customers, unless an administrator had already chosen No users. Microsoft has also renamed the product: Microsoft 365 Copilot is now Microsoft Copilot.
The distinction that matters is who runs the model. Until now the GPT models inside Copilot were operated by Microsoft through Azure OpenAI. The new option routes requests to models that OpenAI runs on its own infrastructure, as a Microsoft subprocessor. Microsoft says its Product Terms and Data Protection Addendum still apply, with listed exclusions, and that these models sit inside the EU Data Boundary except where its documentation says otherwise.
The exclusions are specific. OpenAI-operated models are currently outside Microsoft's in-country processing commitments, which matters for any UK organisation that bought Microsoft 365 partly on the strength of UK data residency. They are not available in government or sovereign clouds. And Microsoft lists compliance reports that do not yet cover them, including SOC 1 Type 2 and the PCI DSS attestation. Microsoft says OpenAI offers zero data retention for the Responses API it uses, subject to OpenAI's own limitations.
Administrators can switch the models off, or limit them to named users or security groups, under Copilot, Settings, AI providers operating as Microsoft subprocessors. It needs the AI Administrator or Global Administrator role. Microsoft notes that some features and models are only available with the setting enabled, so switching it off has a cost in functionality.
Microsoft's year-end numbers
Microsoft closed its financial year to June with more than 30 million paid Copilot seats, and said net seat additions more than doubled quarter on quarter. Azure revenue grew 43 per cent. GitHub Copilot passed 50 million users. Those figures are the background to Microsoft adding more models, and more suppliers, behind Copilot.
OpenAI's pricing, ads and security programme
OpenAI introduced premium seats for ChatGPT Business at $125 per user per month, or $100 a month on annual billing. A premium seat carries five times the usage of a standard seat and removes the five-hour usage window that applies to agentic tools such as Codex. Standard seats stay at $25 a month, and workspace owners can mix the two and reassign seats as needs change. OpenAI said an expanded seat option had been the most requested feature from Business customers.
On 11 August advertising in ChatGPT went live in the UK, Mexico, Brazil, Japan and South Korea. Ads appear only to logged-in adults on the Free and Go tiers. Plus, Pro, Business, Enterprise and Education accounts carry no ads. OpenAI says ads do not influence ChatGPT's answers and that conversations are kept private from advertisers. The test began in the US in February.
OpenAI also split Daybreak, its programme for security defenders, into two tiers. Daybreak Blue gives verified defenders access to GPT-5.6 Sol for defensive work. Daybreak Red sits behind identity verification, monitoring, approved-use restrictions and legal attestations, and gives access to GPT-5.6-Cyber, a model trained for vulnerability research and exploit validation. OpenAI reports that GPT-5.6-Cyber completed 95 per cent of advanced security requests, against 1.5 per cent for the generally available Sol model, and that it found a high-severity vulnerability in V8, Chrome's JavaScript engine, which Google has since fixed as CVE-2026-15903.
For buyers
The Microsoft 365 setting comes first. If the organisation has not looked at it, OpenAI-operated models have been available to its users since 24 July. Whether that is acceptable depends on the organisation's data residency commitments and its data protection impact assessments for Copilot, and either way the decision should be recorded.
The ads change touches the same question from the other side. Staff using free personal ChatGPT accounts for work in the UK will now see advertising alongside their answers. That does not change what data they are putting into a consumer tool, but it is a reasonable prompt to check whether the organisation's AI use policy covers personal accounts at all.
The premium seat turns heavy agent use into a per-person cost rather than a usage ceiling. At five times the price of a standard seat, the sensible first step is to check the workspace usage data for which users are actually hitting the five-hour window, and assign premium seats to those people only. OpenAI is offering $100 of workspace credit per premium seat, capped at five seats, to the first 10,000 eligible customers who sign up by 20 August.
The OpenAI subprocessor setting is in the Microsoft 365 admin centre under Copilot, Settings, AI providers operating as Microsoft subprocessors.
SAASiQ - Intelligent Solutions for SaaS ©
