Microsoft Resells the Frontier

• Google Cloud growing 82% vs Azure's 43% because owning models + TPUs lets them keep margin that resellers pay away—capex runs at 2x revenue yet operating margin expanded to 35.6%
• Microsoft's $678B contracted backlog only grew 25% excluding OpenAI—roughly $220B of $310B added this year traces to one customer funding commitments from capital markets, not profits
• The circular risk: Nvidia guarantees OpenAI's datacenter leases while OpenAI commits to Azure capacity, so the same borrowed dollar appears on multiple balance sheets (S&P downgraded Oracle for similar exposure)
• Microsoft deploys mostly merchant silicon "renting Nvidia's roadmap at Nvidia's margin" while Google sells TPU systems—thinner margins buy less capacity per revenue dollar
• Nvidia's 5-year CDS doubled to 82bps as markets wake up to concentration risk in a financing structure where one borrower underpins half of multiple cloud providers' forward books

The cloud infrastructure race is revealing a fundamental strategic divergence. Google Cloud is growing at 82%—nearly twice Azure's 43% and almost three times AWS's rate—despite being the smallest of the three. The counterintuitive part: Google spends $44.9B quarterly capex against only $24.8B in cloud revenue, yet operating margin expanded from 20.7% to 35.6% as operating income tripled. This aggressive spending works because Google owns both the models and the chips (TPUs), keeping economics that resellers must pay away. Microsoft, in contrast, deploys mostly merchant silicon and resells OpenAI's frontier models, paying Nvidia's margin on every incremental GPU workload and OpenAI's margin on every API call.

The real vulnerability is concentration masked by headline numbers. Microsoft's $678B contracted backlog—the largest of the three clouds—only increased 25% when excluding OpenAI. Of roughly $310B added this fiscal year, some $220B traces to one customer that has committed to buy $250B of Azure capacity. Nearly half of Microsoft's future book rests on a company funding its commitments from capital markets rather than profits. This creates circular risk: Nvidia guarantees OpenAI's datacenter leases and finances its chip purchases, so the same borrowed dollar shows up as contracted backlog on multiple balance sheets. S&P already downgraded Oracle to one notch above junk for the same reason, naming OpenAI a central credit risk against roughly half of its $638B book.

Credit markets are starting to price this concentration. Nvidia's five-year credit default swaps hit a record 82 basis points, doubling from 40 at month-start after reports of the OpenAI lease guarantees. Microsoft's margin structure also constrains capacity deployment—thinner margins buy less infrastructure per revenue dollar, and exposure to one customer argues for restraint even where spending is affordable. CFO Amy Hood frames two-thirds of capex going to CPUs/GPUs rather than concrete as "flexibility," but the likelier reading is capacity you can actually buy from cash flow without underwriting a decade of fixed assets. Nadella emphasizes that "every model is substitutable"—a virtue you highlight when you don't own the frontier and must resell someone else's margin on your own platform.