Actionable insights — This Week in Visuals (14 Q2 prints)

1. When AI capex arrives, read the free-cash-flow swing — and price the "we can rent it out" hedge

1. On any company ramping AI infrastructure, ignore the capex growth rate on its own and go to the free-cash-flow line: did operating cash cover it, or did FCF flip negative in a single quarter?
2. Then interrogate the standard defence — "excess capacity can be rented out externally." Ask whether the company actually owns a cloud/rental channel at scale, and whether that channel earns a return or merely recovers cost.
3. Offset the bill against the distribution that could monetize it (users, installed base), and demand a concrete usage metric rather than a roadmap.

• A single-quarter capex step-change big enough to invert FCF, paired with a rental-optionality argument and no monetization number yet — the point where an AI story becomes a financing story.

2. Strip the hyperscalers out of the order book to test whether a cycle is broad

1. For any AI-infrastructure supplier, find the growth rate excluding the handful of giant cloud customers. A cycle that survives that subtraction is an industry refresh; one that doesn't is customer concentration.
2. Separate orders from revenue and ask the conversion question: how much of the backlog converted this year, and what does the forward revenue target imply about conversion speed?
3. Check the margin cost of the mix — hardware-heavy AI revenue usually dilutes gross margin, so a demand win can still compress the model.

• A backlog that grows faster than the guided revenue conversion — the gap, not the orders, is what the market prices.

3. When customers beg for faster delivery, the bottleneck has moved — and so has the risk

1. Look for supply-side tells in the language: multi-quarter customer forecasts, customers "pushing" for faster delivery, factory-footprint expansion, output-doubling targets. These say capacity, not demand, is binding.
2. Treat that as a high-quality problem — but immediately check what the stock has already done. A supply-constrained leader that has doubled is priced for the constraint easing in its favour.
3. Locate the demand driver inside the mix (which end-market is pulling) so you know what would have to break for the cycle to end.

• Record results plus raised guidance met with a flat-to-down stock: the tell that the constraint story is fully discounted and the next move depends on surprise, not on strength.

4. Compare revenue growth against volume growth — the gap is the take-rate story

1. For any marketplace or payments business, put revenue growth beside volume growth (GMV, TPV, processed volume). Revenue growing faster means monetization is improving; slower means the take rate is compressing.
2. Decompose the direction: rising take rate → higher fees, advertising, value-added services. Falling take rate → mix shifting to cheaper products, or large customers hitting volume tiers with lower pricing.
3. Then ask whether the volume gained is worth the rate given up — check whether total gross profit expectations are still rising, and whether operating leverage improves below the gross-profit line.

• Volume accelerating while gross profit per unit of volume falls for a third consecutive quarter — the point at which "we are gaining wallet share" has to start showing up as absolute gross-profit dollars.

5. Judge a lender on cost of credit and risk-adjusted margin, never on loan-book growth

1. Whenever a fintech grows its loan book fast, go straight to three numbers: provisions (Y/Y), 90+ day delinquencies (vs a year ago), and risk-adjusted net interest margin — the margin after expected losses.
2. A healthy grower shows the portfolio expanding while cost of credit falls and risk-adjusted NIM rises. A deteriorating one shows the book doubling while provisions and late-stage delinquencies jump.
3. Discount management's "specific troubled borrowers / weaker vintages" explanation until at least two subsequent cohorts season cleanly — vintage problems are only anomalies in hindsight.
4. Check early-stage vs late-stage delinquency separately: early-stage improving while 90+ day rises means the new book is better than the old one.

• Loan-book growth and delinquency rate roughly doubling together — the classic sequence in which a lender discovers it grew faster than its underwriting.

6. Separate a guide cut that management chose from one the market forced

1. When guidance is lowered, find which channel or segment shrank. If the company withheld supply from a channel on purpose, read the cut as a mix decision, not lost demand.
2. Corroborate with the margin line: deliberate discipline shows up as record gross margin and raised margin guidance alongside the lower revenue guide.
3. Watch the healthy channel's trajectory (DTC, first-party, subscription) — if it is compounding while the sacrificed channel shrinks, the mix is improving even as the total slows.
4. Set a review date: distinguish temporary inventory discipline from a durable slowdown by whether the sacrificed channel stabilizes within two to three quarters.

• A lowered revenue guide printed alongside a raised gross-margin guide — the signature of a company trading volume for price integrity rather than losing the customer.

7. When a growth pivot changes the margin structure, re-underwrite the business, not the quarter

1. When a new product line drives the growth, check what it does to gross margin — and whether management describes the new level as temporary or permanent. "We expect that margin profile to persist" is a re-rating statement.
2. Ask what kind of company the mix shift is producing (a high-margin software-like model becoming a distribution/logistics model), then apply the multiple appropriate to the new business.
3. Cross-check cash: revenue accelerating while free cash flow turns negative means the growth is being funded, not self-financing.

• Raised revenue guidance paired with narrowed EBITDA guidance — the arithmetic of buying growth at a structurally lower margin.

8. Size an AI revenue lever against the deceleration it has to offset

1. Credit AI adoption only where it appears as recurring revenue — AI ARR, its share of net-new ARR, consumption above included credits — not as usage anecdotes.
2. Then put that lever beside the forward guide. If the company beats and still leaves full-year guidance untouched while guiding the next quarter materially lower, the lever is not yet big enough to offset the core.
3. Read restructuring (large workforce cuts to fund the pivot) as confirmation of urgency, not of success.

• A beat that does not move the annual guide: management is signalling the second half is where the deceleration lands.

9. Track a cash-hoarder by direction of the pile, not its size

1. For a capital-allocation story, ignore the absolute cash balance (always "enormous") and measure the change plus the three uses: net equity purchases, buybacks, and whole-company acquisitions.
2. Net buyer vs net seller of equities is the cleanest single signal — especially the first flip after a long stretch on one side.
3. Then reset the question from timing to returns: once capital is moving, the risk shifts from "will they deploy?" to "at what return, and under whose judgement?"

• A first-in-years flip to net buying alongside accelerating buybacks and an outright acquisition — three simultaneous uses of cash is a policy change, not a trade.

10. Test whether an acquisition widens the platform or just buys revenue

1. For each deal, ask what it lets the acquirer sell to the existing customer that it could not sell before — loyalty, billing, logistics, a new merchant class.
2. Split the raised guidance into organic and acquired, and check whether management admits the underlying business is also running ahead of plan.
3. Price the dilution explicitly: most platform-broadening deals cost margin points in year one, so the test is whether the widened surface earns them back.

• A guidance raise the company itself attributes partly to the acquisition — the honest split is the tell that the organic number is real too.