Despite a +16% move today, the market may still be under-reacting to $MSFTs results last night (still down on yr). Azure growing +43% and guiding to +45% grabbed the headlines, but MSFT was in its worst performance stretch since the GFC, and lowest valuation in a decade, because the world viewed its flagship Office franchise as an AI loser. That view is now demonstrated to be wrong, as Office showed accelerating growth, and Mgmt. guided to continued acceleration throughout FY27 (a strong signal from a conservative mgmt team). I believe MSFT's ~$100B Commercial Office segment can accelerate from mid-teens growth today to >20% growth rates over the next several quarters, driven by current pace of Copilot adoption and a reasonable assumption around E7 agentic orchestration SKU penetration. Paired w/ mid-40s growth in Azure, MSFT sets up well to drive accelerating overall topline and earnings growth over the next couple years. See below where, adjusted for Fx and rev rec, MSFT Commercial Office is already showing clear acceleration, from 14% growth to 16% growth in the last year. If you add the typical beat margin to mgmt's guide, Office will accelerate to +17% growth in 1Q and then further through FY27 (a strong indication from a conservative team). In the the thread below I'll show how continued execution of Copilot at 4Q levels can lift the segment to >20% growth.

We believe superintelligence will create significant new opportunities for all people and businesses. Meta already serves billions of people at scale and helps hundreds of millions of businesses reach customers. Today we are starting the next major pillar of our business, Meta Enterprise Platform, to help businesses use AI to grow and transform in new ways as well.

😈

A humble attempt to est. the infra required to serve 100M DAU @Muse Rough conclusion is: 1 GW of power to serve 100M DAU in the base case, of which only ~0.1 GW comes from the CPU/VM layer. Depending on the # of reasoning-equivalent model calls one Muse DAU generates per day, 3-4GW is entirely plausible. Maybe that’s why @Meta is rumored to be adding 7-10GW of compute next year. The sandbox layer = sub $1B of CPU content and ~$2B of DRAM content, which is much smaller than many expected. Lot of moving assumptions. Welcome all feedbacks/ pushbacks. ------ Two very different pieces of infrastructure behind Muse. 1. Muse VM / sandbox infrastructure 2 vCPUs, ~8 GB of RAM and ~100 GB of persistent logical storage per user. starkinsider.com/2026/09/met… 2. Muse Spark inference Model inference goes out through Meta's external inference infrastructure. research.meta.ai/blog/securi… ------ 1/ Sandbox infrastructure A. CPU The first mistake is assuming that 100M DAU means 100M VMs are actively consuming compute at the same time. Suppose the average Muse DAU has an agent actively working for two hours per day. 100M users * 2 hours / 24 hours = ~8M average simultaneous active VMs Meta obviously cannot provision only for the daily average. Usage will be concentrated during waking hours and bursty. Assume a 2.5x peak-to-average ratio: 8M * 2.5 = ~20M peak active VMs Then add roughly 20% capacity headroom: ~25M provisioned live VMs. So the base assumption is effectively that Meta needs enough infrastructure to support roughly 25% of DAU being live simultaneously. The next important distinction is between virtual CPU allocation and physical CPU demand. Agent sandboxes are particularly well suited to CPU oversubscription. They spend a lot of time waiting. During those periods, the VM may still be alive, but it is barely using CPU. DeepSeek’s recently published DSec infrastructure provides a useful benchmark. Its production agent sandbox platform runs approximately 30,000 physical CPU cores and 250TB of DRAM across ~160 nodes, with peak concurrency above 380,000 sandboxes. arxiv.org/abs/2609.22978 DSec also demonstrates stable operation at around: 800 microVMs per node. With roughly 188 physical cores per node: 188 physical cores / 800 microVMs = ~0.23 physical cores per live VM. DeepSeek is obviously the King of efficiency. The number for Muse might be at 0.3-0.75 physical cores per live VM, or assume 0.5 physical cores per live VM as the base case. That is equivalent to roughly two simultaneously live Muse VMs per physical CPU core. Using the base assumptions: 25M live VMs * 0.5 physical cores per VM = 12.5M physical CPU cores. On a 256-core CPU: 12.5M cores / 256 cores per CPU = ~50K CPUs; Or on a 192-core CPU that would be 65K CPUs. At the current public pricing, that is ~$800M. B. DRAM CPU can be aggressively oversubscribed because a VM that is waiting may consume almost no CPU. Memory is harder to oversubscribe because a live VM still needs to retain its working state. Muse exposes roughly 8GB of RAM to the user environment, but one observed instance was actually using only around 3GB at the time of measurement. 25M live VMs * 3GB = 75PB of physical DRAM, call it ~75-100PB of physical DRAM feels like a reasonable base range. At the current public pricing, that is ~$2B. C. Sandbox power ~0.1 GW for the entire Muse sandbox / VM layer at 100M DAU. ------ 2/ Inference Muse’s personal computer executes tools and stores state locally, but the actual model runs on separate inference infrastructure. Meta’s Muse architecture Energy per inference event Microsoft’s 2026 study estimates that optimized frontier-scale inference consumes a median of approximately: 0.31Wh per normal query But a long reasoning query with roughly 15x the token count consumes approximately 13x as much energy, or around: 4Wh per long reasoning query The study specifically highlights reasoning and agentic workloads as significantly more energy intensive. microsoft.com/en-us/research… Sensitivity analysis on # reasoning-equivalent events per DAU per day Suppose each active @Muse user generates the equivalent of 50 heavy inference events per day. At 5Wh each: 100M users * 50 events/day * 5Wh = 25GWh/day 25GWh/day / 24 hours = ~1.0GW average power So inference alone could require: ~1-2GW of average power A 3-4GW Muse is entirely plausible. Maybe that’s why @Meta is rumored to be adding 7-10GW of compute next year. ------ The popular framing around Muse is that giving every user 2 vCPUs and 8GB of RAM creates an enormous CPU requirement. But the naive calculation materially exaggerates the CPU requirement because it treats logical VM allocation as dedicated physical infrastructure. The more interesting conclusion is: Consumer agents may be a meaningful new demand driver for CPUs and conventional DRAM, but inference remains the real compute bottleneck. And as agents do more work, run longer trajectories and increasingly spawn other agents, inference demand can scale much faster than the number of users itself. +++ Calling my peer review group: @bubbleboi @damnang2 @Midnight_Captl @FundaAI @fi56622380 . Feedback/ Pushbacks pls :). ++ Better formatted: robonomics.substack.com/p/ag…

TSMC's C.C. Wei told Lip-Bu Tan that he wishes Intel is successful in foundry to serve customers because TSMC doesn't have enough capacity to serve demand. Via @ABitPersonalPod $INTC $TSM

Satya Nadella reveals his coding agent pulls every hyperscaler and neoclouds' SEC filings into a dashboard that refreshes every day for real-time ROIC "And this is the other aspect of it, which is the enterprise context combined with the world's context. In fact, I go to the SEC filings of every cloud provider, hyperscaler, each of these neoclouds. It's in real time." "I have a data runner in Fabric that brings all that data, puts it into a semantic model that then gets read by my coding agent and then surfaces it as a dashboard. And every day it's fresh." "So I have the entirety of every SEC filing that goes out there, plus all of my internal analysis constantly coming together, giving me real-time ROIC by layer."

Now that raw model performance isnt the biggest driver of usage anymore $MSFT has the perfect chance to be the “Muse” of enterprise as they can put together a product with the best harness and integrations. While $META has the strongest distribution with consumers, $MSFT has it with enterprises.

We’re building Copilot as a new OS for work that spans every model, every form factor, and every task. Today, we’re announcing our biggest update to Copilot to date, bringing four things together: · Autopilot: proactive and long-running agent built for the enterprise · Code: build apps with Copilot, hosted inside your company’s tenant · Home: Chat + Cowork together · Office: now fully embedded in Copilot (and Copilot embedded in Office, of course!) Plus, you can invoke Copilot in Teams, and we’re introducing Today, a proactive experience that surfaces the most important information from across M365 without needing to ask for it. The way we work is changing and so are our workflows. This update brings AI into that flow, from answering a question, to building an app, to getting work done on your behalf.

Largely agree. 1/ Agentic commerce's impact on $AMZN Ads rev ($100B rev, or est. 50% of operating profit): real disintermediation risk. GMV: depends. If $AMZN is truly "cheaper, faster, better", an objective agent should send more GMV to $AMZN. But I personally doubt $AMZN’s GMV share grows meaningfully in an agentic world. Just look at the $50B in marketing exp. $AMZN throws out today - large platforms have an "unfair advantage" bc they can afford higher CAC. Long-term value / customer ownership: intrinsic value should go down if $AMZN loses the top of the funnel. You may still get the trx and the buyer’s info, but you lose browsing behavior, retargeting, cross-sell opps, and ultimately the customer relationship. Potential upside: $AMZN can open up its best-in-class logistics backbone to other merchants = new rev stream (which it is already doing a little). The risk feels more skewed to the downside, imho: top of funnel → infrastructure layer. 2/ What happens to ads as an industry Under agentic commerce, ad spend and take rate - which ppl historically think of as distinct business models - will converge. In a fully agentic world, ads as an industry could theoretically "disappear." But that doesn’t mean the ad dollars (~$300B in the US) disappear. Sellers will still need (and want) to pay digital tax on distribution. Whether that comes in the form of ads or a take rate is largely semantic. Take OTAs ($BKNG $EXPE) as an example. Say half of their bookings and traffic are direct, and the other half indirect, mostly through $Google. For the indirect portion, the companies basically break even given the high CPCs they pay $Google. Take rate on the direct portion is ~15%. So even though the cost is paid in the form of ads, the equivalent take rate OTAs pay Google is effectively ~15%. An agent charging 2%, 5%, or eventually 10%+ of GMV is therefore not necessarily introducing a new cost. It may simply be repackaging an existing acquisition cost from CPC into CPA / take rate. And agents @Muse will probably start cheap. 3/ What holds true no matter what (pre- or post-agentic world): A. Owning top-of-funnel customer intent = pricing power. Losing direct traffic means losing much more than the immediate transaction. B. Commerce winning formula (is always) = cheaper, faster, better. ++ A more verbose version: robonomics.substack.com/p/ag…

Free newsletter: I estimate that ~50% of AI chip sales - $200bn to $300bn+ - are sitting in warehouses, with NVIDIA selling hundreds of billions of GPUs years in advance. Data centers take way longer to build than hyperscalers are leading us to believe. wheresyoured.at/wherere-all-…

SCOOP: The @Yankees are nearing a deal that would eventually hand a whopping 16% stake in the venerable franchise to private equity giant Apollo Capital Management. The Steinbrenner family would relinquish a “de minimis” slice of its controlling equity, finalizing a transaction that will give the famed franchise a new, cash-rich partner, the @nypost has learned. nypost.com/2026/09/20/sports…