AI & Software
Where AI spending is actually landing
Model releases, enterprise adoption and software margins, read for what they do to revenue — across the labs, the platforms and the software they run on.
2026-11-09
%22%2F%3E%0A%20%20%3Ccircle%20cx%3D%22220%22%20cy%3D%22150%22%20r%3D%22180%22%20fill%3D%22%2322d3ee%22%20fill-opacity%3D%220.12%22%2F%3E%0A%20%20%3Ccircle%20cx%3D%221010%22%20cy%3D%22130%22%20r%3D%22220%22%20fill%3D%22%2338bdf8%22%20fill-opacity%3D%220.08%22%2F%3E%0A%20%20%3Ccircle%20cx%3D%22920%22%20cy%3D%22520%22%20r%3D%22170%22%20fill%3D%22%2322d3ee%22%20fill-opacity%3D%220.06%22%2F%3E%0A%20%20%3Cpath%20d%3D%22M140%20430h920%22%20stroke%3D%22url(%23line)%22%20stroke-width%3D%2212%22%20stroke-linecap%3D%22round%22%2F%3E%0A%20%20%3Cpath%20d%3D%22M220%20360c120-150%20260-220%20380-220s260%2070%20380%20220%22%20fill%3D%22none%22%20stroke%3D%22%2367e8f9%22%20stroke-opacity%3D%220.5%22%20stroke-width%3D%2216%22%20stroke-linecap%3D%22round%22%20stroke-linejoin%3D%22round%22%2F%3E%0A%20%20%3Cpath%20d%3D%22M300%20398h600%22%20stroke%3D%22%2394a3b8%22%20stroke-opacity%3D%220.24%22%20stroke-width%3D%228%22%20stroke-linecap%3D%22round%22%2F%3E%0A%20%20%3Ctext%20x%3D%22220%22%20y%3D%22510%22%20fill%3D%22%23e2e8f0%22%20fill-opacity%3D%220.7%22%20font-family%3D%22Arial%2C%20Helvetica%2C%20sans-serif%22%20font-size%3D%2238%22%20letter-spacing%3D%228%22%3EINSIGHTS%3C%2Ftext%3E%0A%3C%2Fsvg%3E)
Agentic Hiring Is Repricing the Fee Pool, Not Emptying It: Robert Half Earned 5.3% More per Placement in Q3 2026 While ManpowerGroup's Staffing Margin Fell 50 Basis Points
The November 8 recruiting-fee debate was aimed at the wrong line item: AI is deflating the billable hour, not the placement fee. Robert Half raised its average fee per placement 5.3% year over year while placing only 0.9% more candidates, and permanent placement still converts at a 99.8% gross margin, whereas ManpowerGroup grew revenue 4% but saw gross profit fall 1.2% as staffing margin contracted 50 basis points. The rent on those freed hours is migrating upstream into agent software and downstream into direct-hire platforms — Workday, Salesforce, Kanzhun and Recruit Holdings — not into consumer prices. With private-industry wages up just 3.1% and a 29% global hiring outlook, the December Fed decision turns on inflation prints and hiring volumes, not on agency margins.
Amazon's Home-Robot Team Arrives Just as DRAM Prices Quadruple — iRobot's Collapse Shows Who Actually Wins
Amazon has assembled a home-robotics team and is hiring for a \"new-to-world robot product,\" according to a November 8, 2026 CNBC analysis of the company's own job postings. But this is not a category war: the benchmark brand iRobot was cancelled out in a Chapter 11 that handed 100% of its equity to its Chinese contract manufacturer, and Amazon's own first attempt, Astro, was retired in 2024. The binding constraint is the memory bill — DRAM prices have risen more than 400% since the start of 2024 and are still rising 10–15% quarter on quarter, while every robot class from cobots to humanoids needs memory bought at the worst prices in the modern DRAM cycle. Memory makers are already sold out through 2028 and do not need consumer robots; the cost lands on whoever tries to build one.
$2,000 a Kilowatt: Anthropic Paid Peak-of-Cycle for 1.7 GW of Turbines and Inherited the Gas Risk
Anthropic PBC agreed in late September 2026 to pay about $3.4 billion for commitments covering 1.7 GW of heavy-duty gas generators bought from OnePWR Solutions — roughly $2,000 per kilowatt for the machines plus the engineering scope to install them, the same all-in cost analysts put on a new combined-cycle plant. What the lab bought at peak-of-cycle prices was the right to skip the grid, and with it a fuel bill of roughly 0.25 Bcf per day: about $332 million a year at the EIA's 2026 Henry Hub forecast, and $95 million for every $1/MMBtu move. The cash itself went to the turbine order book and the construction chain, not to the grid — Caterpillar enters the fourth quarter with a record $78.1 billion backlog, and the developer that had been holding the slot exited with $3.4 billion. So this is a repricing of who buys power equipment in 2026-27, not a liquidity test of a company holding $131.7 billion of cash and about $895.4 billion of committed spend.
Anthropic's $895 Billion Compute Book Is Financed by Suppliers — While CoreWeave's Debt Yields 16%
Anthropic's amended prospectus was filed on Nov. 2, not Nov. 8 or 9, and it is not a bond deal: it adds a roughly $17 billion third quarter and $895.4 billion of purchase commitments while registering no new public debt. The $10 billion figure belongs to Alphabet's option over Class N shares at the IPO price, and the real debt capacity is roughly $77 billion of bilateral convertible notes from Alphabet, Amazon and Broadcom. That structure has split AI credit into two markets that never meet: CoreWeave's 2032 bonds cleared above 16% yield with five-year CDS near 1,155 basis points the same week Alphabet priced a 16-tranche investment-grade deal from 2.375% to 5.500%. The read-through for investors is that contract structure, not AI demand, now sets the price of AI infrastructure credit, and the risk has migrated onto the supplier guarantees and landlord backstops holding the other side of the buildout.
AppLovin's 43.9% Two-Session Jump Priced the Toll, Not the Quarter: Inference Cost 11 Cents on the Dollar
AppLovin closed 43.9% higher over Nov. 5 and Nov. 6 after reporting third-quarter revenue of $2.18 billion and an 84.3% adjusted EBITDA margin, but the print itself is not what moved the stock. Cost of revenue — the line where rented datacenter inference sits — fell to 11.0% of revenue from 12.4% a year earlier even as management shipped a model transition it calls 'as significant as Axon 2.0,' and 81 cents of every new revenue dollar dropped to operating income. Because AppLovin prices dynamically to campaign goals and books revenue net of publisher payouts, inference is the auction's input cost rather than the toll it charges, so the re-rating paid for pricing power, not for a beat. The cracks are already visible: Q4 guidance embeds an increase in data-center cost, and the constancy of model uplift is now the subject of a securities class action with a Nov. 16 lead-plaintiff deadline.
NVIDIA's Governance Risk Is a Multiples Problem; China's 34% Share Is a Revenue Problem Nobody Owns
The NVIDIA board opened an independent investigation on Nov 8, 2026 into whether founder and CEO Jensen Huang was the target of a foreign intelligence effort, hours after Huang told staff in writing that he had been unfaithful to his wife. The probe lands nine days before a quarter guided to $108.0 billion of revenue and one day before a closed-door Congressional briefing, so what it reprices is the multiple, not the earnings: the outlook already assumes no China data-center compute revenue and data-center Hopper shipments to China ran under 1% of data-center revenue. The number that matters more is one the probe makes riskier: China's share of AI accelerator shipments has fallen to 34% from 55% in 2025, roughly 600,000 H200 chips are sitting in Hong Kong awaiting Beijing's import approval, and the index owns 6.93% of this name with no mechanism to replace a founder.
Sanders' No-Bailout Line Misreads the Ledger: AI's Real Backstop Is a $108.5 Billion Guarantee Book
On Oct. 30, 2026, Sen. Bernie Sanders told Big Tech it “won't be bailed out, period” — a day after Reps. Greg Casar and Maxine Waters introduced a House resolution opposing any federal rescue of AI firms. The promise costs nothing, because the AI buildout was never financed by taxpayers: the $3 trillion Sanders cites is contractual commitments, mostly undrawn leases, and the credit support that actually matters sits in private hands — Nvidia's guarantee book tops out at $108.5B, Broadcom backs the first lien on a $63B financing vehicle, and roughly $500B of AI debt was issued this year alone. What the no-bailout posture removes is the implicit put private lenders and insurers were pricing at, so losses now arrive as marks on retirement portfolios — private credit is 13% of U.S. insurers' cash and invested assets — while the five largest hyperscalers funded $301.8B of the last twelve months' capex with net new debt and $76.0B with net new equity. The trade is therefore not against AI but against the borrowers whose equity has been stripped out, starting with CoreWeave, where $51.6B of debt sits on $25.6B of market value.
One robots.txt Line, Four AI Giants, Zero Dollars: The AI Training Opt-Out Standardizes Refusal, Not Payment
On October 30, 2026, OpenAI, Google, Microsoft and Anthropic said they would honour a single robots.txt token, so a publisher can refuse AI training once instead of crawler by crawler. The text behind it is an individual IETF draft with no formal standing, Anthropic has already implemented it, and Google and Microsoft only promise 2027 dates, which makes this a deployment schedule rather than a rulebook. The token standardizes the refusal and not the payment: it hands the labs a documented compliance record in the publisher litigation while the real gate stays a CDN default, where mixed-use crawlers are 36.6% of Cloudflare's verified crawler traffic and Cloudflare owns the billing relationship. The gap the token cannot close is the shadow supply chain, including a 30-billion-page index that publishers say resold their journalism to five AI labs.
2026-11-08
A Research Firm Read Anthropic's Customer Budgets. The $1.5 Trillion IPO Prospectus Still Doesn't Mention It
On Nov. 7 Anthropic began warning business customers that an outside cybersecurity research firm may have reached their contact, account and spend data — five days after its amended registration statement was filed and roughly a week before the Nasdaq offering can be priced. What leaked was not a chat log but a commercial file: plan tier, usage and spend for a base where more than half the Fortune 500 each spend at least $100,000 a year. Because Anthropic cannot say whose records were viewed, the loss cannot be litigated, so it gets priced through contract indemnity, cyber insurance and the prospectus itself. The investable consequence is a security budget line that moves from agent defence to vendor-access audit.
Anthropic's IPO Buyers Are Talking $1.1 Trillion — Below OpenAI's Just-Closed $1.25 Trillion: The Frontier AI Mark Is Falling From Above
On Nov. 6, prospective Anthropic IPO buyers were discussing $1.2–1.3 trillion for Anthropic, with some weighing as low as $1.1 trillion, the same morning OpenAI closed $11 billion of new capital at a $1.25 trillion valuation. That is the first reported inversion of the frontier-lab mark hierarchy, and it happened while the compute bill was still rising: Anthropic's amended filing shows $895.4 billion of committed spending, $16.96 billion of third-quarter revenue and a $23.29 billion quarterly net loss. The inversion is a financing signal rather than a demand signal, because buyers are repricing a company whose growth slowed from 144% to 47% in a quarter and whose cash cannot cover the next twelve months of contracted compute. For public investors the read-through runs upstream to accelerators and memory, and downstream to the neoclouds financing the buildout, where credit spreads rather than order books are now the transmission channel.
Axon Paid $5.2 Billion for 37% of Skydio — More Than Ondas' Entire Market Cap — and Its Own Shares Fell 18%
Axon Enterprise agreed on Nov. 5, 2026 to invest $1.2 billion for roughly 20% more of Skydio, lifting its fully diluted stake from about 15% to about 37% at a $5.2 billion pre-money valuation and becoming Skydio's exclusive public-safety distributor in the US and Canada. The transaction says the marginal buyer of drone autonomy is no longer a defense prime but a camera-and-evidence company that already owns the procurement relationship with police departments. Skydio's mark exceeds Ondas Holdings's entire $3.1 billion market capitalization at the Nov. 6 close and approaches Kratos Defense's $6.6 billion, while Shield AI's $28.8 billion financing mark is 2.3 times the combined market value of AeroVironment and Kratos Defense. Investors sold the one listed vehicle that owns the channel by 18% on the day it bought it.
Fireworks AI's $40 Billion Ask Never Printed — Now Meta's CFO Has to Make $17.5 Billion Work
Fireworks AI's valuation has not been cut; it has stopped moving — $17.5 billion, set on July 15, 2026, while the company was reported on Oct. 20 to be chasing $30 billion to $40 billion, and peers reprinted higher (Fal closed at $16 billion on Nov. 3). Dave Wehner's Nov. 6 arrival as Meta's former chief strategy officer says the binding constraint is capital structure, not product: 95% of Fireworks' 40 trillion daily tokens now run on customer-specialized models, 25,000 GPUs sit inside customer data centers, and the capacity is leased from landlords such as IREN that earn 86% gross margin. At roughly $0.07 of revenue per million tokens against a $0.05 best-in-class cost that Nvidia publishes, the margin in this stack sits upstream of the layer priced at 17.5x revenue. Read the stale mark and the CFO hire together: the layer is becoming infrastructure, and infrastructure does not keep a software multiple.
Nooks Took $500 Million Without Diluting — AI-App Growth Capital No Longer Comes From Equity Markets
On Nov. 4, 2026, Nooks took a $500 million growth-capital commitment from General Catalyst's Customer Value Fund and launched Node, its first AI personal assistant for sales. The money is non-dilutive, has no maturity date and no covenants, and is repaid only from collections on the customer cohorts it financed — 7.1x everything Nooks has raised as equity in its life. The closest published term sheet, General Catalyst's $1 billion facility with IM8 filed with the SEC, shows the instrument: capped return per cohort, no equity instrument, downside held by the fund. That is why the card rails and banks are the natural next owners of this paper — Visa and Mastercard already credential agents and settle their purchases, and FIS ships a know-your-agent platform to issuing banks — while the equity market is repricing the apps themselves: HubSpot lost 59% of its market value in the year to September 2026 even as trailing revenue grew 20%.
2026-11-07
Airbnb Spent $7 Million on Capex and Authorized $12 Billion of Buybacks While the AI Complex Spent $615 Billion
On Nov. 5, 2026, the Nasdaq 100 fell 6.36% — its worst session since March 2020 — and roughly $2.5 trillion of market value evaporated as investors pressed the case that a $900 billion AI capex wave is running ahead of returns. That night, Airbnb reported revenue up 19% to $4.87 billion, a 39% net margin, and a new $12 billion repurchase authorization. The divergence is structural rather than lucky: Airbnb spent $7 million on capital expenditure over the trailing year because the AI it buys is operating expense rented from hyperscalers that spent $615 billion on the same balance sheets. Capital allocation has become the AI factor, and on the test of who turns AI into cash without capitalizing it, Airbnb behaves like an annuity trading at 21 times trailing free cash flow.
Five9's AI Revenue Grew 91%, Its Legacy Seat Business Grew 7.5%, and the Stock Fell 17.6%
Five9 beat on revenue ($323.3 million, up 13%) and raised full-year revenue guidance by $6 million on Nov. 5, yet the stock fell 17.6% the next trading day because the non-GAAP profit guide was cut from $3.26 to $3.19 per share at the midpoint while gross margin slipped 110 basis points. Strip out the $45 million of AI revenue and the subscription business that still licenses human agents grew about 7.5%, less than half the AI line's pace. The $325 million cash purchase of Level AI (Ujwal, Inc.), at roughly 16 times its $20 million of ARR, marks agent revenue at a price about nine times what the public market now pays for all of Five9's revenue.
Alphabet’s SpaceX Governance Link Just Broke—Don Harrison’s Exit Forces a Reprice of How Investors Should Think About “AI in Orbit” vs. a $94B Stake
SpaceX disclosed that Google executive Don Harrison resigned from its board effective immediately on Nov. 4, 2026, removing a long-running related-party governance bridge between Alphabet’s SpaceX holding and Google’s competing push to put AI compute into orbit. The next market test is whether shareholders still underwrite Alphabet’s “orbital AI” story on the assumption that board-level alignment reduces execution risk—or whether the loss of that bridge raises the probability-weighted cost of capital.
OpenAI Closed $11 Billion at $1.25 Trillion. Anthropic's $1.5 Trillion IPO Is Now the Political One
Private capital funded the buildout; voters set the discount rate. OpenAI closed $11 billion at a $1.25 trillion mark on Nov. 6, three days after voters in 10 states backed restrictions in 40 of 41 local data-center ballot measures, and its CFO said the spending continues regardless. Anthropic, by contrast, must sell its story to the public market at the same moment: its amended S-1 shows $33.2 billion of nine-month revenue against $895.4 billion of mostly non-cancellable compute commitments and a $104 billion net loss driven by fair-value marks on the instruments that paid for that compute. The result is a two-tier frontier stack in which private money carries the buildout without a daily mark while the public market is asked to absorb the political and contractual risk at IPO. For listed suppliers, the read-through is that contracted demand is intact while the equity financing it is losing.
Palo Alto's Second $500 Million Agent-Security Deal Prices a $2.8 Billion Market That Isn't Budgeted Yet
Palo Alto Networks is reported to be paying roughly $500 million in cash, plus milestone payouts, for Clutch Security — the same price it paid for Console in September 2026 — and two days earlier CrowdStrike committed about $495 million to Fiddler AI and Dreadnode. Those two commitments alone equal 53% of Gartner's forecast $2,835 million of worldwide AI-security spending for all of 2026, even though the budget barely exists: Clutch is closing on roughly $10 million of annual recurring revenue, about 50 times sales. The mechanism that matters for shareholders is the currency — Palo Alto's fiscal 2026 Form 10-K shows $24.9 billion of acquisition consideration, 74% of it Palo Alto stock, and fiscal 2027 guidance assumes 844 million to 847 million diluted shares against 764 million a year earlier. Agent security is consolidating ahead of its demand, and Palo Alto shareholders are being asked to fund the gap.
Twelve Leveraged Anthropic ETFs Are Armed Before a Single Share Trades — and the Bid Comes From a Tape That Just Fell 5.5%
ProShares' Nov. 6 launch of a 2x and a -2x Anthropic ETF completes a 12-fund single-stock complex whose own prospectuses state that ANT has not yet commenced its IPO, so the vehicles are registered but not tradeable. Once Anthropic lists, the complex will not offset itself: on an up day the +2x cohort adds exposure while the -2x cohort covers its short, and on a down day both sell, which makes the debut price a reflexive flow artifact before it is a valuation verdict. That matters now because the AI tape those flows would ride has already broken — the Nasdaq fell more than 5.5% on Nov. 5 and SK hynix dropped 10.44% on Nov. 6 — while the print under discussion is roughly 41x trailing sales.
2026-11-06
What to expect
Evidence-first notes with a visible point of view.
This section collects sharp takes on earnings, shareholder meetings, and market structure. Each new piece should make the thesis, the facts, and the implications obvious within the first few screens.
Expect direct analysis, not generic commentary.
Expect the data to be explicit and the argument to be easy to follow.