▶ Explore this week’s Tape — live, sortable, drill-down → Microsoft Is Funding the Next AI Layer With the Last One Enterprise AI stopped being a story about models this week and became a story about invoices. On July second, Microsoft — page one of the Cash Flow Memo — stood up a unit called Frontier Co.: two and a half billion dollars, six thousand people, one job, which is to take AI products the last mile into enterprise deployment.¹ Two days later it moved to cut up to five thousand roles from sales, consulting, and Xbox.² The new layer is being paid for with the headcount of the old one. That is not a hiring plan. It is a company rebuilding itself around the part of AI that actually sends a bill. The model layer got three years of narrative. The deployment layer — the boring work of installing the thing, retraining the seat, wiring it into the workflow — is where the money changes hands, and the market has not repriced for that yet. Microsoft just showed you how the transition lands in a real profit-and-loss statement before it lands in a revenue line: deployment headcount up, legacy headcount down, net headcount roughly flat, and the revenue you are supposedly buying still not visible. The screens know how to price a growth story. They do not know how to price a margin-mix reallocation that hasn’t reached the top line. We have watched Microsoft run this exact play once before. The last decade’s version was the move from packaged software to subscription — the company retooled its sales motion and its cost base around the cloud years ahead of the recurring revenue, wore an ex-growth multiple through the gap, and re-rated hard only once the ARR became legible. Frontier Co. is the same bet, one rung up the stack. The reorg comes first. The revenue is the lagging indicator. The interval between them is exactly where a name gets mispriced. The pricing tells landed the same week, on both ends of the stack. Microsoft made the Copilot Business seat a permanent product at twenty-one dollars a month³ — the output price, now fixed. And AWS, per a Yahoo Finance report, raised GPU instance prices about twenty percent on July first⁴ — the input price, moving up. Read those two together and you are watching enterprise-AI unit economics get discovered in real time: the cost of compute rising, the price of the seat set. The names that compound from here are the ones that can lift the output price faster than the input cost climbs. Microsoft, sitting on the seat, can. Most of the memo, buying the compute, cannot. One honest caveat on the input side, because it is doing a lot of work: the AWS move is one vendor’s list price on specific instances. Azure and Google have not publicly matched, and a list price is not what a committed customer actually pays. Whether it holds through the next wave of capacity — the sixty-five-to-seventy-five-billion-dollars-per-gigawatt data-center builds Hunt, Jason, and Mike walked through on Wednesday⁵ — is the tell for whether this is structural scarcity or a headline. Don’t let one instance-price line stand in for the whole compute market. The cashflow read is in Marcus’s column below — short version, Microsoft is paying about thirty-eight times trailing free cash flow⁶ for a deployment curve that isn’t in the numbers yet, on capex already running close to a hundred billion a year.⁷ Palantir, the pure-play version of the same government-and-enterprise bet, asks a hundred and eight.⁸ What changes the read is retention, not growth. The test on the next print, due late this month, is whether Copilot seat retention and attach start to show up inside Microsoft Cloud growth before the capex compounds past them. If deployment headcount grows and the seats don’t stick, that is the tell that enterprises are buying the org, not the outcome — and the whole bet inverts. Watch the AWS price too: if Azure and Google haven’t matched within a quarter, the twenty-percent move was a one-vendor list-price event, not the scarcity signal it read as. Mark the calendar for both. Wall Street’s consensus on enterprise AI: show us the revenue, then we’ll pay for it. But the org charts and the price tags moved this week, and the revenue line always arrives last. The last time Microsoft reorganized ahead of its own revenue, the people who waited for the number paid up for the wait. The Tape — W2627 Universe of 94 cashflow-memo names, snap dates 2026-07-02 → 2026-07-03. Composite is rank-sum percentile of FCF Yield + NTM Revenue Growth (higher = better balance). Banks and finance-book names shown separately. Telltales Yield — Top 10 From the Cashflow Desk — Marcus Graham The enterprise-AI trade got expensive this week, and the cheapest way to own it is sitting at number four on the board. Salesforce already licenses a seat inside nearly every enterprise now buying AI deployment — and the memo has it at 10.5x trailing free cash flow, a 9.5% FCF yield, against single-digit NTM growth. Palantir asks ~108x for a version of the same government-and-enterprise AI story. That gap is either the market correctly pricing Salesforce as ex-growth, or an unpriced option on Agentforce turning installed seats into AI attach. The table can’t tell you which. The test on the next print is whether agent adoption shows up in current RPO and net seat expansion — or whether single-digit growth is the new ceiling. Telltales Yield — Bottom 10 This Week’s Reporters No universe names reporting in the coming 7 days. Sector Medians Debt / FCF Watch (highest leverage on TTM FCF) Weekly Price Movement Top 5 (week-over-week price) Bottom 5 (week-over-week price) Banks (shown separately — FCF metric not meaningful) Finance-book — FCF not comparable Customer-float / captive-finance / reserve businesses (IBKR broker float, KMX CarMax Auto Finance, PYPL customer funds, CRCL stablecoin reserves). The memo’s operating-FCF method overstates their FCF, so they are held off the ranked leaderboard pending the P&L-waterfall rebuild. Data Gaps 91 of 91 ranked-eligible names ranked. 0 dropped for missing FCF yield or NTM revenue growth; 7 shown separately (banks + finance-book, FCF not comparable). Source: cashflow-memo master_2026-07-03.csv. NTM growth from analyst-estimates consensus. Composite is a percentile rank, not a recommendation. The Issue — This Week's Brief The Cashflow Memo W2627 — Tesla Blew Past Delivery Estimates, Comcast Split Itself into Cash, and Enterprise AI Started Setting Prices Tesla blew past delivery estimates, Comcast split itself into cash, and enterprise AI started setting prices. The Telltales Weekend Update. Ava Cabot and analyst Marcus Graham walk through what happened this week — and what’s coming next — across the companies in the Cash Flow Memo. About 14 minutes. No filler. Download the memo at telltales.us. Hunt, Jason, and Mike are back Wednesday on episode E2628. Chapter markers * Time | Segment * 0:00 | Opening disclaimer * 0:15 | Cold open — throughline + prior-Wed callback * 0:45 | Theme — AI Goes to Work (Microsoft, Palantir) * 4:45 | Deep dive — Two Multiples (Tesla, Comcast) * 8:45 | Rapid-fire (Walmart, Harrow, Eli Lilly) * 11:45 | Close — Consensus Watch + earnings season preview * 12:30 | Closing disclaimer Full transcript Opening disclaimer Ava: The following conversation is intended for informational purposes only. You should always do your own work to determine if an investment is suitable for you. Cold open Ava: You’re listening to the Telltales Weekend Update. I’m Ava Cabot. Marcus: And I’m Marcus Graham — the cashflow desk. Ava: Quick note: the show is produced entirely with AI tools, and both voices you’re hearing are AI-generated. Send feedback through the Substack. Ava: Enterprise AI went from story to price action this week. Microsoft committed $2.5 billion and 6,000 employees to a dedicated AI implementation unit — and simultaneously flagged cuts to up to 5,000 more in legacy functions. Palantir won the U.S. Army’s data backbone contract for its highest-priority modernization program. And AWS quietly raised GPU instance prices 20% on July 1[^amzn-gpu-pricing-20260701] — the first open, on-record pricing signal that compute scarcity is structural, not cyclic. Meanwhile, two companies at opposite ends of the Cash Flow Memo’s valuation spectrum each delivered exactly what they promised. And the market’s reaction told you what it’s currently willing to pay for. Ava: On Wednesday, Hunt, Jason, and Mike worked through the cost of building the AI infrastructure itself — data center builds now running $65–75 billion per gigawatt, memory running at 30–40% of the total buildout cost[^ep-e2627]. This weekend, we pick up on the demand side: who is deploying that infrastructure, at what scale, and what the cash flow math says about what the market believes will pay for it. Theme — AI Goes to Work Ava: On page one of the Cash Flow Memo this week — Microsoft moved. Not in the abstract direction of AI, but in the direction of paying for it organizationally. On July 2, Microsoft committed $2.5 billion and is deploying 6,000 employees into a new entity called Frontier Co. — a dedicated AI implementation unit built specifically to take AI products from development into enterprise deployment[^msft-frontier-ai-20260702]. On July 1, Microsoft also made Copilot Business a permanent SKU at $21 per user per month[^msft-copilot-sku-20260701], the pricing structure that the deployment unit is built around. And on July 3, the company announced it is cutting up to 5,000 employees from sales, consulting, and Xbox[^msft-layoffs-20260703]. Microsoft is funding the AI deployment bet with the headcount from the pre-AI org. That is a choice. Marcus — the cashflow frame. Marcus: Microsoft just put a price tag on the deployment layer — and it’s the capex number that makes this intere
