What the pre-market is doing
Cloudflare trades at 330.50 in the pre-market against a 284.43 close — up 16.2% on its Thursday-evening print. Twilio trades at 226 against 193.20, up 17.0%. Both reported strong beats and both are being paid without hesitation.
That completes a natural experiment the punishment-phase read set up yesterday: Datadog, which beat and raised in the same neighbourhood on Thursday morning, closed down 19.0% at 229.29. Same week, same macro tape, adjacent businesses — opposite outcomes. The rest of the board is holding its breath: futures are flat, and the July payrolls print lands at 14:30 CET with forecasts running from 18,000 to 83,000 jobs.
Under it all sits the bond line this desk has tracked for four sessions. IEF closed Thursday at 92.95, through the two-year trendline, on the session low — the veto is fired but appealable, and the jobs number is the appeal court. The Morning 10 carries the full scoring plan.
Four prints, one verdict: the operations layer is being paid
Strip the reactions away and look at the filings. Cloudflare grew revenue 35.9% year over year to $696m, and growth accelerated from the March quarter. Twilio grew 22.0% to $1.50bn — its strongest growth in years — and beat the earnings bar by 11%.
Datadog grew 36% and raised its full-year outlook, as covered yesterday. And Palantir, which reported Monday, grew 92.8% to $1.94bn with a 47% operating margin — a software company nearly doubling at scale while printing margins most of the industry only models.
That is the bottom line of the week, and it deserves to be stated plainly before any risk discussion: business-model and revenue-wise, the AI-operations trade works. Enterprises adopting AI are paying more for observability, for edge delivery and security, for customer messaging, for deployment platforms. Four different companies, four different pricing models, one direction.
The squeeze: when the model is free, the labs must eat someone’s lunch
So why did one of the four lose a fifth of its value on a beat? Because the market is no longer pricing whether the opex layer grows. It is pricing who else wants that growth.
The mechanism runs through open source. Every capable open-weights release pushes the price of raw model access toward zero, and a frontier lab that cannot monetise the model directly must monetise around it — upstream into infrastructure, or downstream into applications and services. The OpenAI in-house observability overhang named in yesterday’s Pulse is the first concrete sighting: a frontier lab building, for itself, exactly the product it currently buys.
Follow that logic one step further and it stops being a Datadog story. Any vendor whose largest customers are frontier labs is selling to the entities most capable — and now most commercially motivated — of replacing them. The exposure question is not “does AI help this business” but “does this business’s customer list overlap with a lab’s future product list.”
The gradient: Datadog, Cloudflare, Twilio, Palantir
Datadog sits closest to the fault line. The labs are among its largest and fastest-growing accounts, observability is core engineering competence for any lab, and the product’s buyer is the same engineer who could build the replacement. A 36%-growth beat-and-raise sold 19% is the market marking that exposure to market — the print record on the Datadog card now carries the cleanest disruption-risk repricing in the series.
Cloudflare is in between. The labs are customers, and inference serving at the edge is plausibly on their roadmap — but Cloudflare’s asset is a physical network with fifteen years of buildout, not a software feature, and its customer base is the breadth of the internet rather than a handful of labs. The market paid the print 16% anyway: replicating a network is a different proposition from replicating a dashboard.
Twilio is a different animal. Carrier relationships, telecom plumbing and enterprise messaging compliance are nowhere near a frontier lab’s ambitions — and agentic applications plausibly increase message volume rather than replace it. Its 17% pre-market gain prices a business on the receiving end of the disruption, not in its path.
Palantir is the antipode, and the market knows it. Open and on-premise where the labs are closed API businesses, outcome-priced where they are usage-priced, embedded in government and industrial workflows the labs cannot enter — 93% growth priced as the anti-lab. The gradient the tape drew this week runs exactly along customer-list overlap, and it repriced four adjacent names in four different directions to say so.
September carries the same question: the July-quarter cohort
The next tranche of the opex complex closed its quarter at the end of July and reports from late August into early September. Each lands somewhere on the same gradient.
GitLab carries the highest adjacency. Code is the frontier labs’ home turf — agentic coding is their flagship use case — and GitLab’s seat-priced model sits in the blast radius the market mapped onto HubSpot on Wednesday. Its print will be read for seat counts before revenue.
MongoDB sits mid-gradient. The data-of-record layer is sticky and labs do not want to run databases — but AI-native stacks increasingly route around incumbent data platforms, so the risk is erosion by new architecture rather than direct lab competition. Samsara sits furthest away: physical-operations telemetry on fleets and industrial assets is about as far from a lab’s product list as enterprise software gets.
Around them report the consumption-priced security and data names — Snowflake, CrowdStrike, Zscaler, Okta, Elastic. Wednesday’s spillover map already showed the market’s working theory on that class: consumption pricing tied to workloads took none of the seat-model damage. September tests whether that immunity survives contact with actual guides.
C · members block
Into the close
Four checks carry today, in order.
The payrolls print at 14:30 CET against the bond line. IEF closed through the two-year trendline on Thursday; a strong number confirms the break and the hiking path, a weak one reclaims the line and kills the veto a day after it fired. Score it at the close either way — the framework only means something if it can be embarrassed in public.
Cloudflare and Twilio at the close, not the open. The reaction scores lock tonight. If both hold double digits through a payrolls tape, the entry-premium rule from the print-record series graduates from observation to working law.
Datadog’s second session. Day two after a −19% print tells you whether the disruption discount was fully priced in one sentence or is being paid in instalments. Watch whether it can hold Thursday’s 225–229 zone without the tape’s help.
The gradient’s own test. If the market truly priced customer-list overlap this week, Palantir and Twilio should outperform Datadog on any relief move — the discount should stay ordinal even when the whole group rallies. If the order scrambles, the disruption read is noise and we say so tomorrow.