AI infrastructure also defines the cost of operating
On August 10, 2026, Nvidia announced partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to create computing infrastructure financing platforms. The stated goal is to mobilize more than US$ 500 billion in third-party capital over time.
The number needs to be read carefully. It isn't a single fund or a direct Nvidia investment of that amount. The final agreements still depend on formalization. The move, however, shows that chips, data centers, energy, networks and credit now influence the capacity available for AI applications.
The AI race also happens outside the models
A company that uses AI for occasional tasks may barely feel these factors. When the technology starts to support customer service, analysis, production or recurring integration, changes in price, availability or vendor can affect the process.
The cost also goes beyond the subscription. Depending on the application, it includes compute consumption, storage, integration, maintenance, data, security, latency and team time. The ability to handle demand peaks can be as relevant as the unit price of a call.
The decision is to plan before scaling
The announcement doesn't mean an SMB should build its own infrastructure. It means the company needs to understand which layer it depends on to keep the process running. A solution can look economical at small volume and lose viability when use grows.
You also need to evaluate vendor dependence, processing location, contracts, continuity and the possibility of migrating. The most technically powerful option isn't necessarily the most suitable if there isn't the financial and operational capacity to sustain it.
The next step is testing the total cost
Before expanding an application, estimate how use changes with volume, peaks and new areas. Ask what capacity will be needed, how the price evolves, whether there is a vendor alternative and how much downtime the operation tolerates.
This planning takes time and may delay an expansion. The cost is lower than finding out, after building AI into a critical routine, that processing became expensive, scarce or dependent on a single infrastructure. Scale starts with a capacity hypothesis, not just a license.
Scale is also an operational decision
Before expanding, estimate demand, contracted capacity, expected availability and the cost of interruptions. Record which components depend on a single vendor and which alternative keeps the process running if the planned capacity isn't available.
Sources
Nvidia, 26 Aug 2026. https://investor.nvidia.com/news/press-release-details/2026/NVIDIA-Announces-Financial-Results-for-Second-Quarter-Fiscal-2027/default.aspx
Banco da Inglaterra, 7 Jul 2026. https://www.bankofengland.co.uk/financial-stability-report/2026/july-2026
Banco da Inglaterra, 7 Jul 2026. https://www.bankofengland.co.uk/financial-policy-committee-record/2026/july-2026