Skip to main content
Daniel J Glover
Back to Blog

Nvidia earnings: plan AI capacity spend

Published
Coverage:
6 min read
Article overview
Written by Daniel J Glover

Practical perspective from an IT leader working across operations, security, automation, and change.

Published 9 September 2026

6 minute read with practical, decision-oriented guidance.

Best suited for

Leaders and operators looking for concise, actionable takeaways.

Topics covered

Retrospective covering 26 August 2026, written on 9 September 2026.

Nvidia earnings are a signal to review AI capacity exposure, not proof that a particular AI project deserves a larger budget. My recommendation is to connect every proposed capacity commitment to measured demand, an accepted-work metric and a credible exit or fallback arrangement.

Nvidia reported second-quarter fiscal 2027 revenue of $96.2 billion on 26 August, including $89.0 billion in Data Center revenue. Total revenue was 106% higher than a year earlier. Those are supplier results for the quarter ended 26 July, not a forecast of the savings a UK business can achieve from AI. Nvidia's official financial results.

Translate the news into the right question

The scale of the announcement can make a local project feel urgent. A finance director may ask whether the business risks falling behind. A technical team may see an argument for reserving compute. A supplier may use the wider demand story to encourage an earlier commitment.

My response would be to ask what has changed in this organisation's own demand. Has a pilot become a service? Are waiting times preventing accepted work? Is there a contractual capacity problem? Has a provider supplied a revised quotation or an availability constraint?

These are decision inputs the organisation can verify. General market momentum is context. Keep that distinction visible in the paper so the board can understand whether the recommendation responds to a measured operational need or an expectation about the market.

The IT budget business-case guide provides the broader structure. The capacity request should explain what additional work the business can complete and why the existing arrangement is insufficient.

Classify the commitment before comparing prices

Different AI services create different obligations. I would first write down what the business is buying: usage of a hosted application, access to a model API, reserved infrastructure or hardware it will operate itself.

Then ask which costs continue when demand disappears. A comparison that places a flexible usage estimate next to a committed infrastructure price without accounting for unused capacity is not a useful buying comparison.

Use this proposed review table:

Buying questionEvidence to request
What work needs capacity?Observed task volumes and service requirements
What is being committed?Minimum spend, term, scope and cancellation conditions
What could remain unused?Low-demand scenario and utilisation assumption
What else is required?Storage, network, support and operational dependencies
What happens if it fails?Tested fallback and named service owner

Ask the supplier to distinguish a guaranteed allocation from an expectation, and a quoted price from a forecast. Retain those answers with the decision. They will matter more during an incident or renewal than the market slide that introduced the proposal.

Measure accepted work and demand shape

For an internal evaluation, choose a business result that has a clear acceptance point. It might be a reviewed report, an approved draft or a completed analysis. These are suggested measures; the organisation should select the one that reflects its own service.

Track unsuccessful attempts and correction effort alongside successful output. A capacity expansion that produces more material for a reviewer to reject may increase activity without relieving the actual bottleneck.

Also examine when work arrives. A service used unpredictably has a different planning problem from one with a stable queue. Record the difference between typical demand and exceptional bursts, and ask the business owner how quickly each type of work really needs to finish.

The AI ROI guide explains how to connect results to value. Here the operational question is whether the proposed capacity is the limiting factor in producing those results.

Test the downside case before committing

I would require a downside scenario in which demand is lower than forecast and the expected productivity improvement arrives later. Use the organisation's own assumptions and quotations, rather than invented industry averages.

Ask what would still be payable, which work could move to another service, and how long it would take to exit. Identify the person authorised to stop or reduce the commitment if the assumptions fail.

Then consider a separate service interruption scenario. The business might need an alternative process, a second provider, local equipment or simply a clear queueing arrangement. Choose the response according to the consequence of delay. Do not add a complex second architecture without a demonstrated need and a team able to maintain it.

This is where the multi-cloud strategy guide is relevant. Provider diversity has to solve a concrete operational problem; it should not become an automatic response to a concentrated market.

Keep supplier results and forecasts distinct

Nvidia's release also provides a third-quarter revenue outlook. An outlook is forward-looking company guidance. It should not be presented as revenue already earned, or converted into a promise about your supplier's future pricing. Nvidia's results and outlook.

In the local business case, apply the same discipline. Label observed demand, supplier commitments and management forecasts separately. If the proposal relies on an unconfirmed future product, discount or delivery window, expose that dependency before approval.

The useful comparison is between options that can actually meet the requirement. A lower theoretical cost on unavailable capacity is not an alternative the business can rely on. Nor should a rapid procurement decision conceal unresolved data, support or exit requirements.

Give finance a decision it can revisit

End the recommendation with the workload, the spending boundary, the acceptance measure, the review date and the trigger for reducing or stopping the service. Keep the technical explanation sufficient for scrutiny without turning the paper into a catalogue of processors.

Assign a service owner who can reconcile the operational evidence with the bill. That person should be able to explain whether increased spend reflects more accepted work, poorer efficiency, a changed requirement or an unused commitment.

The August results make the infrastructure boom tangible. A well-run SME does not need to imitate the purchasing scale behind that boom. It needs a capacity decision grounded in its own work, with assumptions that can be checked and commitments that remain defensible when demand changes.

Frequently Asked Questions

What did Nvidia report on 26 August 2026?

Nvidia reported second-quarter fiscal 2027 revenue of $96.2 billion, including $89.0 billion from its Data Center business. Those results describe the supplier's business. They do not establish the return from an individual customer's AI project or guarantee the availability and price of capacity for a particular workload.

Should an SME reserve GPU capacity because demand is growing?

I would require measured workload demand and a comparison of commitment options before reserving capacity. Check the utilisation assumption, service requirements, cancellation terms and fallback. A strong supplier earnings report can justify reviewing exposure, but it is not enough evidence to sign a long commitment for an unproven internal workload.

What cost measure should an AI business case use?

Use a measure tied to completed, accepted work, alongside the infrastructure bill. Include review, correction, retries, support and idle commitments where they apply. Record how the measure changes under lower demand or poorer quality. That makes the proposal easier to compare with the current process and alternative delivery options.

Share this post

About the author

DG

Daniel J Glover

IT Leader with experience spanning IT management, compliance, development, automation, AI, and project management. I write about technology, leadership, and building better systems.

Continue exploring

Keep building context around this topic

Jump to closely related posts and topic hubs to deepen understanding and discover connected ideas faster.

Browse all articles

Explore topic hubs

Ready to Improve Your IT Operations?

Book a free 30-minute consultation to discuss your IT challenges. No commitment required, just a focused conversation about where you want to be.

Book a consultation

Get Occasional IT Leadership Insights

IT leadership insights, occasionally. No fluff. Unsubscribe any time.

No spam. Unsubscribe any time.