Microsoft Corporation (NASDAQ:MSFT) is asking a more demanding question about AI infrastructure: how much useful output does each dollar and watt actually produce? NVIDIA Corporation (NASDAQ:NVDA) has a substantial stake in the answer.
At SEMICON Taiwan, described in a September 2 Microsoft update, Azure hardware chief Rani Borkar argued that capacity expansion must be accompanied by better efficiency across the entire computing system. The company calls that objective useful yield.
Microsoft Headquarters
For investors, the implication reaches beyond a benchmark contest. Microsoft needs infrastructure economics that support profitable services. NVIDIA needs customers to find its systems valuable enough to justify continued spending.
The customer’s savings can support the supplier
Microsoft’s approach spans memory, networking, power, software and fleet operations. Its Azure Maia design includes an integrated network interface and a two-tier scale-up network, illustrating how a cloud operator can optimize hardware around its workloads.
That creates a competitive test for NVIDIA. Custom designs give large buyers another route to efficiency, potentially limiting the economics available to outside suppliers on suitable workloads. 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.
Better efficiency can also support NVIDIA’s opportunity. If lower computing costs make more applications commercially viable, customers may deploy additional capacity rather than simply reduce their infrastructure budgets. That outcome depends on demand expanding enough to absorb the savings.
NVIDIA’s August 26 results show the scale of the business exposed to that equation: second-quarter fiscal 2027 Data Center revenue reached approximately $89 billion. Strong sales establish current demand; they do not determine how future efficiency gains will be divided between hardware suppliers, cloud operators and customers.
Output must become an economic return
Microsoft’s own risk is equally important. More tokens per dollar are useful only if customers want the output and pay enough for services to cover infrastructure and operating costs. Lower technical costs need not automatically produce higher shareholder returns if competition passes the savings to buyers.
Historical positioning remains substantial. 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’s firm reported positions in both. Those quarter-end holdings predate the keynote.
The August 14 short-interest snapshot showed 0.92% of Microsoft’s float and 1.23% of NVIDIA’s sold short.
For Microsoft Corporation, useful yield becomes valuable when it supports profitable customer workloads. For NVIDIA Corporation, the test is whether system advantages keep earning attractive economics as buyers optimize their fleets. The companies can benefit together, but neither wins merely because more hardware is installed.
READ NEXT: NVIDIA (NVDA): What Foxconn and Super Micro Are Telling Us about the AI Boom and Pony AI Is Scaling Robotaxis Fast—Can the Stock Reach BofA’s $17 Target?
