WEBVTT NOTE The Rundown — nextbig.dev daily audio edition, 2026-07-29 1 00:00:02.600 --> 00:00:10.620 The cloud is monetizing faster than the assistant. It's Wednesday, July twenty-ninth, and Microsoft spent forty-one billion dollars in a quarter while Azure grew forty-three percent. 2 00:00:10.800 --> 00:00:23.370 Quarterly revenue reached ninety billion dollars, up eighteen percent. Microsoft Cloud produced fifty-nine-point-three billion, and management said new Azure capacity was quickly monetized. About two thirds of capital spending went into short-lived CPUs and GPUs. 3 00:00:23.360 --> 00:00:36.530 Operations generated fifty-five-point-four billion dollars, free cash flow held at nineteen-point-six billion after the build, and commercial obligations rose eighty-four percent to six hundred seventy-eight billion. All sequential obligation growth came from customers outside frontier-model companies. 4 00:00:36.530 --> 00:00:50.870 Microsoft three-sixty-five Copilot passed thirty million paid seats after net additions more than doubled. Agent three-sixty-five registered nearly forty million agents in two months, and the assistant is moving toward seat-plus-usage pricing. The next proof is keeping cloud margin near sixty-five percent as those seats become active inference. 5 00:00:51.050 --> 00:01:06.380 Mark Zuckerberg expects billions of personal agents within five years and says more than one million businesses already use Meta's agents. Reality Labs still lost four-point-six billion dollars in the quarter. OpenAI, meanwhile, described a family of devices, multiplying the surfaces where an assistant must earn a repeated job. 6 00:01:06.560 --> 00:01:23.290 A hidden instruction inside Word can pass through Copilot and reproduce in documents generated downstream. The HANDBOOK dot M D benchmark explains the failure: the best of thirty long-context agent configurations obeyed every standing rule in only thirty-six-point-two percent of trials. Policy needs a control outside the model and evidence after the action. 7 00:01:23.460 --> 00:01:38.950 To the tape. We lift the Microsoft watch to high conviction as growth, obligations and free cash flow now support the infrastructure thesis, with short-lived hardware and cloud margin as the risks. We sharpen the Meta watch, where distribution is proven and agent economics are not. The tape is the desk's scorecard, not advice. 8 00:01:39.130 --> 00:01:47.910 Our call: by November fifteenth, Microsoft reports at least forty million paid Copilot seats and cloud gross margin of at least sixty-three percent. Missing either threshold proves us wrong.