Data Divers Dialogs · By Christian Schappeit · August 2026
On 26 August Salesforce reported its best quarter in years and the stock closed up almost 23 percent — $2.53 of its $5.90 in adjusted earnings per share came from a stake nobody sold. In the three weeks before that, HubSpot had quietly begun moving judgement calls that used to belong to people into the software.
The first: $33.5 billion, up 14 percent. That is cRPO, the contracted backlog for the next twelve months. We had that number in the calendar here, and it did exactly what it had been told to do, which was nothing interesting. Fourteen percent last quarter, fourteen now, fourteen in the guidance for next.
The second sits higher up the income statement, on a line nobody has looked at in years: Gains (losses) on strategic investments, net. A year ago it said $6 million.
This time it says $2,613 million.
The stock closed the following day up almost 23 percent.
Salesforce Ventures first backed Anthropic in early 2023 with around $50 million and has taken part in rounds since. At the end of May 2026, Anthropic raised $65 billion at a post-money valuation of $965 billion. In the quarterly filing the position carries at roughly $5.1 billion — 45 percent of the entire strategic investment portfolio. In January it was 22 percent.
Nothing was sold. No shares changed hands, no buyer was found, no money arrived. Somebody else raised a round, and the accounting did the rest.
This is entirely legal, entirely ordinary and, under US accounting rules, actually required. It is also by some distance the best trade Salesforce has made this decade.
What is interesting is not that it happened. What is interesting is where the number landed.
Every quarterly report carries a small table setting out what the company strips out of official profit to arrive at "adjusted" profit. This time Salesforce's looked like this: amortisation of purchased intangibles $0.64, stock-based compensation $1.10, restructuring $0.14, tax effects minus $0.27.
Read it again and notice what is missing.
The investment gain is not stripped out. It stays in. Which is how Salesforce came to report that adjusted earnings per share had more than doubled, to $5.90.
Without the stake it would have been $3.37. Analysts expected $3.27. Of the $2.63 surprise, $2.53 came from the portfolio. Ninety-six percent.
"The operating business beat expectations by three percent. The stock rose twenty-two."
How much of that is Anthropic is not in the press release of 26 August. It is in the quarterly filing that reached the SEC the following day: $2.7 billion of unrealised gain from the Anthropic position alone.
That is more than the $2,613 million on the line. The difference is around $296 million of write-downs on other holdings in the same quarter. The rest of the portfolio lost money, and Anthropic covered it.
The price moved on Wednesday evening. The breakdown arrived on Thursday.
The second half of the news was the raised full-year outlook, and that is what got celebrated. In May, Salesforce had guided fiscal 2027 adjusted earnings to $14.06–$14.12 a share. It is now $16.67–$16.71.
That is $2.61 more. The investment gain already booked is $2.53.
The difference — six to eight cents — is the part of the raise that comes from selling software.
Salesforce does not hide this. It is in the footnote to the guidance, in a sentence worth reading twice: projected earnings per share assume "no change to the value of our strategic investment portfolio as it is not possible to forecast future gains and losses". The sentence after it notes that the impact of future gains or losses "could be material".
Which is to say: we have written $2.53 into the full-year number and assumed the underlying value does not move between now and January. On a privately held stake whose value has just shifted by $2.6 billion inside a single quarter.
Agentforce, the AI product the whole story is built around, passed $1.5 billion in ARR, up 240 percent. Together with Data 360 it comes to nearly $3.9 billion, up 210 percent. Those are good numbers. They are annual.
One footnote belongs here because it fits the rest: as of this quarter, Salesforce counts Slackbot and Headless 360 towards Agentforce ARR. So the 240 percent is not comparing quite the same thing with quite the same thing.
The paper gain on the portfolio was $2.6 billion. In three months.
Which means that last quarter Salesforce reported more profit from owning a piece of Anthropic than Agentforce earns in revenue in a year.
Marc Benioff did two things on the same day.
He called the SaaSpocalypse — the argument that AI models make enterprise software unnecessary — nonsense. His words: the "nonsense of this SaaSpocalypse", and it is "time for it to stop".
And he launched Claudeforce. A plugin that lets salespeople query, update and act on their CRM data without ever opening Salesforce. 37 prebuilt sales skills, pilot customers now, open beta in September. Benioff's line for it: "the UI is the AI." Product chief Patrick Stokes added that the interface to SaaS is going through "a pretty significant period of change".
One detail of the arrangement is too good to walk past. Stokes also explained how access gets billed: "headless consumption pricing", where the number of API calls you get depends on which user licence edition you bought. Access to a product built explicitly so that nobody opens the interface is still measured by the number of seats nobody is sitting in.
Then there are two invoices, one from Salesforce and one from Anthropic. From seats to consumption, in practice. Procurement calls it two cost centres, one of which has no ceiling.
None of this is a contradiction. It is a hedge.
If the models do not eat enterprise software, Salesforce carries on selling software. If they do, Salesforce owns a piece of the thing eating it. Both bets are running. Both pay into the same line of profit.
A hedge is simply not a product strategy. And in the number that went across the tickers on Wednesday evening, you cannot see the difference.
In the three weeks before Salesforce's record quarter, HubSpot shipped four things nobody wrote about.
On 24 July, bulk import of existing contracts. On 3 August, an API for the new contract object that has, since June, been created automatically the moment a quote is accepted — proration included, when somebody upgrades or downgrades mid-term. On 4 August, monitoring of sales tax registration obligations, which flags when a company crosses the threshold in a US state at which it has to register.
And on 7 August, generally available: an AI that sorts open receivables. It scores three dimensions — revenue risk, customer risk, invoice risk — and says which one gets chased first. The release notes describe it with unusual precision as "Breeze's first GA milestone in a billing workflow".
Who gets chased first used to be a decision. It was made by somebody in accounts receivable who knew that the customer in Ingolstadt always pays late and always pays, and that with the other one you had better call straight away. Now it is a ranking made of three scores.
None of it moved a share price. Together it moves something other than an interface.
"Salesforce moved the interface. HubSpot is moving the judgement."
Which tax rate applies to this line item. Whether we have become liable to register in Colorado. What the contract is worth after May's downgrade. Who we chase first. Those were judgement calls. Now they are defaults.
A tax rate is not a recommendation. It is a determination with legal consequences. If it is wrong, HubSpot does not get the assessment.
This is where it turns practical for the readers of this publication.
In the projects I actually sit in — pharma, medical devices, insurance — assessing supplier stability is part of the process. There are questionnaires. They ask about certifications, subcontractors, where processing happens, exit scenarios, occasionally audited accounts.
And at some point the vendor's quarterly slide turns up in the business case. "AI revenue up 210 percent." "Adjusted profit doubled." Both sentences are true. The first is a statement about a product. The second is a statement about a securities portfolio.
How much of our vendor's reported profit comes from selling its software?
I have never seen that question on a supplier questionnaire. It takes two minutes to answer — the reconciliation table is in every press release — and nobody asks it, because until now nobody had to. For twenty years, a software vendor's reported profit was roughly what it had earned selling software.
That has stopped being true. And it has stopped being true of more than Salesforce. Every vendor that invested early in a model company now carries the same line in its books.
The answer rarely changes the buying decision. Salesforce does not become unstable because a paper gain shrinks; the operating business stands on its own. It changes something else — what you tell the steering committee about the vendor's financial strength, and how much weight you give a roadmap that came out of the same deck as the doubled profit.
The second question is the uncomfortable one, because it creates work.
Which decisions will this software now make that a person used to make — and who signs them off?
The Salesforce investment gain costs nobody an hour. It sits on a line, it moves a price, and in a validated environment it changes precisely nothing. HubSpot's invoice prioritisation costs work: specification, testing, evidence, and somebody who can explain, when asked, why this receivable came before that one.
One of them got reported. The other one has to be defended.
Two dates for the diary
September 2026 — Claudeforce open beta. The pilot is running, open beta is announced for September, further skills for late 2026. The number worth watching will not be user counts but the second invoice: Salesforce bills API access through the licence edition, Anthropic bills the tokens. Who owns that consumption is unresolved in most companies.
Q3 FY27 — the repeat. The full-year guidance assumes a change of zero on the investment line for the rest of the year. Any number other than zero is therefore, by definition, unguided — in both directions. cRPO is guided at around 14 percent again.
cRPO stands at $33.5 billion. Operating cash flow is guided to grow four to five percent across the full year. That is the business: solid, very large, growing in the low double digits. On Wednesday nobody cared.
What moved the stock was a line that said $6 million a year ago, and about which the company itself writes that it cannot be forecast.
Next quarter it could say six again. Or carry a minus sign.
cRPO will be around fourteen percent.
And somewhere in an accounts receivable department, since 7 August, a ranking decides who gets chased first. That one didn't move a price.
Figures and sources. Salesforce Q2 FY27, quarter ended 31 July 2026, published 26 August 2026: revenue $11.345bn (+11% Y/Y and in constant currency), subscription and support revenue $10.820bn (+12% Y/Y, +11% in CC, including roughly $440m from Informatica), GAAP diluted EPS $4.29 (+119%), non-GAAP $5.90 (+103%), GAAP operating margin 20.5%, non-GAAP 34.1%, cRPO $33.5bn (+14% Y/Y and in CC), RPO $66.3bn (+11%), operating cash flow $1.3bn (+71%), free cash flow $1.1bn (+81%). Agentforce ARR above $1.5bn (+240%), Agentforce and Data 360 together nearly $3.9bn (+210%). The line "Gains (losses) on strategic investments, net": $2,613m against $6m in the year-ago quarter.
FY27 guidance raised to $46.1–46.4bn revenue and $16.67–$16.71 non-GAAP EPS; in May the figures were $45.9–46.2bn and $14.06–$14.12. Operating and free cash flow growth is unchanged at approximately 4–5%. The GAAP-to-non-GAAP reconciliation contains four items — amortisation of purchased intangibles $0.64, stock-based compensation $1.10, restructuring and acquisition-related costs $0.14, tax effects −$0.27 — and no item for investment results. Salesforce itself puts their effect at $2.43 on GAAP and $2.53 on non-GAAP diluted EPS. The guidance footnote reads: "The Company's projected GAAP and non-GAAP diluted net income per share assumes no change to the value of our strategic investment portfolio as it is not possible to forecast future gains and losses. The impact of future gains or losses from the Company's strategic investment portfolio could be material."
The breakdown by holding is not in the press release but in the quarterly report (Form 10-Q, filed 27 August 2026): unrealised gains of $2.7bn for the quarter and $3.0bn for the half year on the Anthropic position; carrying value approximately $5.1bn at 31 July 2026, around 45% of the total strategic investment portfolio against 22% at 31 January 2026. Gross upward adjustments of roughly $2.9bn in the quarter were set against about $296m of impairments and downward adjustments. Also in the filing: Agentforce ARR has included Slackbot and Headless 360 since Q2 FY27, so the prior-year base is not identically defined.
HubSpot, the same weeks. From the release notes: 24 July 2026, Contract Imports in public beta (bulk import of existing contracts by CSV); 3 August, Contracts API in public beta; 4 August, Tax Registration Obligation Monitoring, with alerts on crossing US economic nexus thresholds; 7 August, AI Invoice Prioritization generally available — Breeze scores open invoices on revenue risk, customer risk and invoice risk, described in the release notes as "Breeze's first GA milestone in a billing workflow". The contract object itself entered public beta in June 2026: accepted quotes automatically generate a contract record governing billing and payments, covering upgrades, downgrades, proration and renewals without a new deal. Automated sales tax on quotes — calculated in real time from line item category and buyer address, for the US, Canada and the UK — became generally available in July.
What is established and what is not. Established, because it comes from the filings themselves: every Salesforce figure above. Established from primary source: Anthropic's $65bn Series H at a $965bn post-money valuation, announced 28 May 2026. Derived but checkable: the $3.37 operating EPS, the 96% share of the surprise against a $3.27 consensus, and the six to eight cents of operating contribution to the guidance raise. From reporting rather than company disclosure: the initial roughly $50m investment in early 2023 — it was the first of several investments, the total capital deployed is not disclosed, and no multiple can be derived from it. The closing price on 27 August was around $252 against $205.62, a gain of about 22.6%; depending on the measurement window, figures between 13.5% and 23% are in circulation. The comparison that a vendor can earn more from a stake than from its AI product sets a quarterly gain against annual revenue: it is arithmetically correct and deliberately lopsided, and it says something about the profit line, not about the business value of Agentforce.
For HubSpot, feature names, dates and beta or GA status come from the release notes; the contract object and the Customer Agent invoice and payment functions are additionally covered by independent reporting. The renaming of Commerce Hub to Revenue Hub on 16 June 2026 comes from partner sources rather than a HubSpot announcement. The reading — that what is moving here is judgement rather than interface — is an interpretation of these releases, not a finding. It is an emerging pattern rather than an established one: the individual features are unremarkable on their own, and whether the cadence holds will only be visible over the next two quarters.
Primary sources: Salesforce Q2 FY27 · Form 10-Q · Claudeforce · Anthropic Series H · HubSpot release notes. Reporting: Reuters · Motley Fool · EBC · Benzinga · VentureBeat · Bloomberg · MarTech · Vantage Point.