Microsoft Corporation (NASDAQ: MSFT) is demanding a more rigorous accounting of AI infrastructure, asking how much useful output each dollar and watt actually produces.
The company calls this objective “useful yield,” a standard that carries significant implications for NVIDIA Corporation (NASDAQ: NVDA), which has a substantial stake in the answer.
At SEMICON Taiwan, outlined 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.
For investors, the implication reaches well beyond a simple benchmark contest between competing chip architectures or data center designs.
Microsoft needs infrastructure economics that can support profitable services, while NVIDIA needs its customers to find its systems valuable enough to justify continued and growing spending.
Microsoft’s efficiency approach spans memory, networking, power, software and fleet operations, with its Azure Maia design featuring an integrated network interface and a two-tier scale-up network.
That architecture illustrates how a cloud operator can optimize hardware directly around its own workloads, creating a genuine competitive test for outside suppliers like NVIDIA.
Custom designs give large buyers an alternative route to efficiency, potentially limiting the economics available to third-party suppliers on suitable workloads, though Microsoft’s announcement does not disclose any replacement of NVIDIA hardware or reduction in purchases.
NVIDIA’s August results also reported Vera Rubin racks running at Microsoft Azure, underscoring that the two companies remain deeply commercially intertwined despite Microsoft’s push toward custom silicon.
Better efficiency from Microsoft’s investments could also expand NVIDIA’s long-term opportunity, as lower computing costs may make more applications commercially viable and prompt customers to deploy additional capacity rather than simply cut infrastructure budgets.
NVIDIA’s August 26 results illustrate the scale of business exposed to that shifting equation, with second-quarter fiscal 2027 Data Center revenue reaching approximately $89 billion.
Strong sales establish current demand, but they do not determine how future efficiency gains will ultimately be divided between hardware suppliers, cloud operators and end customers.
Microsoft faces its own version of this risk, since more tokens per dollar are only valuable if customers want the output and pay enough for services to cover infrastructure and operating costs.
Lower technical costs do not automatically translate into higher shareholder returns if competitive pressure simply forces cloud operators to pass the savings directly to buyers.
Insider Monkey’s tracked fund data counted 273 Microsoft holders in Q2 2026, down from 282 in Q1, while NVIDIA holders rose to 285 from 275, with Ken Fisher’s firm reporting positions in both companies.
The August 14 short-interest snapshot showed 0.92% of Microsoft’s float and 1.23% of NVIDIA’s float sold short, reflecting relatively modest bearish positioning against both names.
For Microsoft, useful yield becomes genuinely valuable only when it supports profitable customer workloads that translate into durable, growing service revenues.
For NVIDIA, the core test is whether its system-level advantages continue to earn attractive economics as large buyers like Microsoft aggressively optimize their hardware fleets.
The two companies can benefit together from AI infrastructure growth, but neither wins simply because more hardware is installed across global data centers.
