The question Microsoft is now raising about AI infrastructure is more demanding than before: what amount of useful output comes from each dollar and watt spent. NVIDIA has a significant stake in how that question gets answered.
In a September 2 update on SEMICON Taiwan, Microsoft’s Azure hardware lead Rani Borkar said that adding more capacity has to come with better efficiency throughout the whole computing system. The company refers to that goal as useful yield.
Rani Borkar’s Argument at SEMICON Taiwan
Beyond a simple benchmark contest for investors, Microsoft’s success depends on infrastructure economics that sustain profitable services, while NVIDIA’s depends on customers finding its systems valuable enough to keep paying.
Borkar argues that expanding the number of data centers alone will not solve the problem. What matters instead is how efficiently the systems inside those centers operate — how much they produce for each unit of power and money invested.
The Azure Maia Design
The company’s strategy covers memory, networking, power, software and fleet management. The Azure Maia architecture brings together a built-in network connection with a two-tier scale-up network, showing how a cloud provider can tailor hardware to its workloads.
This puts NVIDIA through a competitive trial. When companies build their own custom hardware, they gain an alternative path toward efficiency, which could restrict what outside suppliers can charge for the right kinds of tasks.
Microsoft’s announcement does not, however, disclose a replacement of NVIDIA hardware or a reduction in purchases. NVIDIA’s August results also reported Vera Rubin racks running at Microsoft Azure.
NVIDIA’s $89 Billion Stake
Lower computing costs stemming from improved efficiency could help NVIDIA’s chances by making more applications commercially practical. That would give customers a reason to add extra capacity instead of cutting their infrastructure spending.
That outcome depends on demand expanding enough to absorb the savings.
The August 26 report from NVIDIA reveals the size of the business caught up in that equation: second-quarter fiscal 2027 Data Center revenue came to around $89 billion. Solid sales confirm current demand, though they offer no sign of how future efficiency gains will be split among hardware suppliers, cloud operators and customers.
Microsoft’s Own Risk
The danger Microsoft faces comes from the same source: a higher number of tokens per dollar only helps if people actually want the output and pay enough for services to cover the infrastructure and operating costs.
Reduced technical expenses do not necessarily result in greater gains for shareholders when rivals pass on the cost reductions to customers.
If the market expands, a customer’s savings can help support the supplier. But if efficiency gains merely reduce the total spend, then both Microsoft and NVIDIA suffer.
Tracked Shareholder Positions
Both companies continue to carry significant historical holdings. The firm behind Insider Monkey’s tracked worksheet sample counted 273 Microsoft holders in Q2 2026 versus 282 in Q1, and 285 NVIDIA holders versus 275. Ken Fisher’ has disclosed positions in both.
Those quarter-end holdings predate the keynote.
Key Facts
- Event: SEMICON Taiwan keynote, described in a September 2 Microsoft update
- Speaker: Rani Borkar, Azure hardware chief
- NVIDIA Data Center revenue: approximately $89 billion, second-quarter fiscal 2027 (reported August 26)
- NVIDIA hardware in use: Vera Rubin racks running at Microsoft Azure
- Insider Monkey tracked holders (Q2 2026 vs Q1): Microsoft 273 vs 282; NVIDIA 285 vs 275
Timeline
| Date | Event |
|---|---|
| Q1 2026 | Insider Monkey tracked 282 Microsoft holders and 275 NVIDIA holders |
| Q2 2026 | Insider Monkey tracked 273 Microsoft holders and 285 NVIDIA holders |
| August 26 | NVIDIA reports Q2 fiscal 2027 results; Data Center revenue approximately $89 billion |
| September 2 | Microsoft publishes SEMICON Taiwan update with Borkar’s useful yield argument |
The useful yield test is now the operating standard Microsoft is pushing across its fleet. NVIDIA’s economics will be measured against it, and so will Microsoft’s own returns.
Source: finance.yahoo.com
Get the Notebook.
The day's best stories and every fresh verdict, in plain English, in your inbox by seven. One email a day, no more.