Salesforce Borrowed $25B to Shrink Itself, and Owner Cash Still Covers the Bill

Summary

Salesforce ($CRM) runs the customer record for a large share of the world's sales and service teams, and over the past two years it has rebuilt that franchise around agents. In fiscal Q2 2027, the quarter ended July 31, 2026, revenue was $11.35B, up 10.83%, and current remaining performance obligation (cRPO), the contracted revenue due in the next twelve months, was $33.5B, up 14% ^[1].

Agentforce and Data 360 annual recurring revenue reached nearly $3.9B, up over 210% ^[1]. Those are the headline facts.

The facts that matter more for our view sit lower in the filings.

First, about 4.45 points of the quarter's growth came from Informatica, which closed in November 2025. Organic growth was 6.38% (our calculation from $456M of Informatica revenue) ^[1]. Second, Salesforce borrowed heavily to fund a $25B accelerated share repurchase in March 2026. Debt went from $14.44B at January 31 to $39.29B at July 31, and quarterly interest expense went from $67M to $473M ^[2].

Third, the company has also signed cash deals for Contentful (about $1.5B), Fin, the Intercom agent business (about $3.6B), and Braze (about $5.66B of enterprise value) ^[2][4].

Against that, the cash engine is strong. On the trailing twelve months to July 31, 2026, operating cash flow was $15.75B, all capex was $0.6B and stock pay was $3.67B, so owner free cash flow was $11.49B, or 26.15% of revenue

Thesis

Our thesis is that the record still sets the price. Superintelligence makes it cheaper to build a sales screen, but it does not make it cheaper to replace twenty years of customer history, permissions, workflows and integrations that a large company has wired into Salesforce.

Agents need that record to act, and Salesforce is pricing the agent work separately on top of the seats. That is why the data and integration group grew 20.28% in Q2 while the apps group grew 7.65% ^[1].

The second half of the thesis is about capital. Management used debt to buy back about 14.66% of the weighted basic share count in a year (821M shares in Q2 against 962M a year earlier) ^[1].

A debt-funded buyback is leverage, not a return of cash, so we treat it as a risk to be paid for. Our point is that owner cash can pay for it: $11.49B a year of owner free cash flow against roughly $1.89B a year of run-rate interest (our calculation from Q2 interest expense) ^[1][2].

The failing test is the one the whole industry fails. A customer pays $0.10 for a standard agent action whether or not the action closes a deal or resolves a case

Product: the clouds and Agentforce

Salesforce now discloses its subscription revenue in two groups ^[2]. The first, Agentforce Apps, contains Agentforce Sales, Agentforce Service, Agentforce Marketing, Agentforce Commerce, Agentforce Apps Flex Credits and Slack.

It brought in $7.19B in Q2, up 7.65%. The second, Data 360, Headless Platform and Other, contains Data 360, Data 360 and Platform Flex Credits, the Headless Platform, Informatica, Agentforce MuleSoft, Agentforce Tableau and other items.

It brought in $3.62B, up 20.28% ^[1][2]. The renaming matters less than the regrouping: management now wants investors to judge it as an apps company plus a data company.

Sales is the original product. It holds accounts, contacts, opportunities and forecasts, and it sells per user per month in editions.

Service holds cases, knowledge and field service and is the largest home for customer-facing agents, because a support question is the easiest job to hand to software. Marketing and Commerce cover campaigns, journeys and storefronts.

Each of these clouds earns subscription revenue billed mostly annually in advance, which is why unearned revenue swings through the year and why the first quarter carries most of the collections.

Agentforce is the layer that lets software act inside those clouds rather than just display data. An agent can update a record, summarize a case, answer a product question or run a flow ^[5].

Salesforce counts this work in its own unit, the Agentic Work Unit. It delivered 3.2B AWUs in Q2, up 97% from Q1, and 7.0B to date across Agentforce and Slack

Product: data, integration and the acquired pieces

Data 360, formerly Data Cloud, is the part of the company growing fastest. It unifies customer data from Salesforce and outside systems so agents can act on a full profile.

In Q2 it ingested 104T records, up 355%, and 82T of those came through Zero Copy, up 731% ^[1]. Zero Copy means Salesforce reads the data where it already sits, for example in a customer's warehouse, rather than copying it in.

That lowers the customer's cost and switching friction, and it makes Salesforce the place where the permissions and the actions live even when the data does not.

MuleSoft sells integration: connectors and an API layer that link Salesforce to the rest of a company's systems. It is priced on capacity and usage and sits in the faster group.

Informatica, bought in a deal that closed in November 2025, adds data quality, cataloging and master data management, the plumbing that decides which record is the true one. It contributed $456M of revenue in Q2, and management guided to slightly above 4 points of contribution in Q3 ^[1].

Tableau sells analytics seats and is now branded Agentforce Tableau. We see it as a mature product whose main job is to keep the data layer sticky.

Slack is the collaboration surface and, increasingly, the place where employees talk to agents. It posted its fastest growth in net new annual order value since the acquisition, as Slackbot users grew over 150% from Q1 ^[1].

Slack is sold per seat, with Slackbot capacity drawn from Flex Credits ^[5].

Pricing and Flex Credits

Salesforce now sells three ways at once. The first is the classic seat: a monthly price per user by edition.

The second is the premium seat with agents bundled in, such as Agentforce One Edition. The third is Flex Credits, a prepaid pool that customers draw down as agents work ^[5].

Flex Credits cost $500 per 100,000 credits. A standard or custom agent action consumes 20 credits, so $0.10, and a voice action consumes 30 credits, so $0.15 ^[5][6].

Prompts consume between 2 and 16 credits depending on size, and data processing in Data 360 has its own tiered multipliers ^[6]. Credits must be used before the order end date, with no rollover ^[6].

Salesforce also lists an Agentforce user license at $5 per user per month, a conversation price of $2 and a flat-fee access option at $125 per user per month ^[5]. Enterprise Edition customers and above can get 100,000 credits free through Salesforce Foundations, a trial path into paid consumption ^[5].

The economics favor Salesforce in three ways. Credits are paid up front, which adds to unearned revenue and cRPO before any work is done.

Unused credits expire, so breakage is profit. And the multipliers can be changed: the rate card states that usage types, tiers and multipliers may be updated from time to time

Operations

Salesforce had 83,334 employees at January 31, 2026, up from 76,453 a year earlier, partly from acquisitions ^[3][13]. It has not updated headcount since.

Layoffs have come in smaller waves this year: close to 1,000 roles across teams in early 2026, then 86 roles in California and 133 more in California and Washington effective October 5 ^[9][10]. These numbers overlap and should not be added together.

The direction is a reshaping rather than a shrinking: the company is cutting support and product roles while it hires around agents. Restructuring expense was $94M in Q2 against $4M a year earlier ^[1].

Costs tell the same story. In Q2, research and development was $1.69B, up 13.91%; sales and marketing was $3.86B, up 12.08%; general and administrative was $0.73B, down slightly ^[1].

Sales and marketing still takes about 34% of revenue, the largest cost line by far. On the trailing twelve months, sales and marketing spending of $15.1B produced about $1.11B of added quarterly revenue year on year, of which only about $0.65B was organic (our calculation) ^[1][12].

That is low sales efficiency for a company of this scale, and it is the clearest place where agents could cut cost: Salesforce is selling agents that do sales and service work, and it has the largest sales force in enterprise software to test them on.

The buyback, the debt and dilution

In March 2026 Salesforce entered accelerated share repurchase agreements for $25.0B, paid up front, and received about 103 million shares at the start, about 80% of the shares expected ^[2]. Final settlement is due in the fiscal third quarter, which ends October 31 ^[2].

It also bought 11 million shares in the open market in Q1 ^[2]. About $22.9B of authorization remains ^[2].

The money was borrowed. In March the company issued nine series of senior notes and a $6.0B term loan, raising $24.84B net ^[2].

Debt principal now totals $39.5B, with coupons between 4.24% and 6.70% on the new paper ^[2]. The two Informatica credit facilities, $6.0B at January 31, were repaid ^[2].

Cash fell to $8.31B, with $3.09B of marketable securities that we exclude from net debt under our method ^[2]. Quarterly interest expense rose from $67M to $473M ^[1].

The maturities are front-loaded. About $21.5B, 54.43% of principal, comes due by 2031, including the term loan

Financial update

Q2 fiscal 2027 revenue was $11.35B, up 10.83%, with subscription and support of $10.82B, up 11.66%, and professional services of $0.53B, down 3.85% ^[1]. GAAP net income was $3.53B against $1.89B, but $2.61B of that came from gains on strategic investments ^[1].

Social Capital ties most of the gain to Salesforce Ventures' stake in Anthropic ^[7]. We strip it out: it is a mark on a private holding, not operating income.

Remaining performance obligation was $66.3B, up 11%, and cRPO $33.5B, up 14% ^[1]. cRPO growing faster than revenue is the best leading signal in the release, because it means more contracted revenue is lined up for the next twelve months.

Part of it is Informatica and part is prepaid Flex Credits, so it overstates organic demand somewhat. Q3 guidance calls for cRPO growth of about 14%, excluding Contentful and Fin ^[1].

Operating cash flow in Q2 was $1.27B and capex $0.17B, so free cash flow was $1.10B ^[2]. Q2 is the seasonally light quarter because most invoices are collected in Q1. Owner free cash flow for the quarter was about $0.19B after $0.91B of stock pay.

On the trailing twelve months, the numbers are the ones above: $15.75B of operating cash flow and $11.49B of owner free cash flow ^[2]

Valuation

Method: basic cover-page shares (823M as of August 20, 2026), cash and equivalents only ($8.31B), debt at carrying value ($39.29B) plus $5.66B for the cash Braze deal, and owner free cash flow equal to operating cash flow minus all capex minus stock pay ^[2][4]. Today's enterprise value is $227.26B, against $226.67B at the October 29 close.

Enterprise value divided by owner free cash flow is 19.78x.

Owner free cash flow has been positive for six straight fiscal years, from $1.90B in fiscal 2021 to $10.89B in fiscal 2026, so a base value is allowed under our standard ^[12]. The model grows revenue for five years, moves the owner free cash flow margin, and applies an exit multiple.

Base: 9% growth, margin up 3 points, 25x. Bear: 5% growth, no margin gain, 18x.

Bull: 13% growth, margin up 6 points, 30x. The multiple is year-five value divided by today's enterprise value.

The 9% base growth includes acquired revenue from Informatica and, later, Braze. Organic growth was 6.38% in Q2. At 6% growth the base case falls to 1.89x, still above the 1.5x line.

The reverse cash flow model, at a 9% discount rate and 3% terminal growth, says the current enterprise value implies 6.3% annual owner cash growth for five years; at 8% it implies 2.09% and at 10% it implies 10.08% (our calculation). Organic growth roughly matches what the price asks for.

Superintelligence angle and leadership

We rate the superintelligence effect as mixed. The tailwind is that agents need a governed record and a permission system to act safely, and Salesforce owns both for customer work.

The 97% quarter-on-quarter growth in agent work units and the 731% growth in Zero Copy ingestion are early proof that customers are routing agent work through Salesforce rather than around it ^[1]. The threat is that the same models make the screens less valuable.

Social Capital's research notes that Salesforce and Anthropic launched a plugin that lets a seller work inside Claude without opening a Salesforce screen, and that Claude became the default model behind Agentforce and Slack ^[7]. That is Salesforce choosing to hand over the interface to keep the data and permissions.

It is a sound defensive move, but it accepts that seats may be worth less.

Social Capital's broader research on the agentic economy argues that value collects in the harness and data layers and that fewer than 10% of organizations run agents at meaningful scale ^[8]. Both points fit our thesis: the gap between pilots and production is where a governed record earns its keep.

Chamath Palihapitiya has also predicted that traditional per-seat enterprise software will structurally underperform; we treat that as a live risk rather than settled.

Marc Benioff, the co-founder, is Chair and CEO ^[1]. There is a single share class and no controlling holder.

The style is founder-led and acquisitive, with a habit of bold product renaming and big bets, from Slack to Informatica to the current deal run. The capital allocation shift of recent years, from no buyback and no dividend to a $25B debt-funded repurchase plus a dividend, is a break from the company's growth-first past.

Risks

The debt-funded buyback is the first risk. Salesforce moved from a near net-cash position to $39.5B of principal in six months, and it plans to spend more than $10B in cash on three more deals.

Interest now takes about 16% of owner free cash flow. If growth slows or rates rise when the 2028 to 2031 maturities come due, the company will have less room to keep shrinking the share count, and stock pay will start diluting holders again.

Seat compression is the second. Salesforce's own messaging is that agents do work people used to do, and it has cut support headcount on that basis ^[10].

Customers can apply the same logic and buy fewer Sales and Service seats. The apps group grew 7.65% in Q2, and that includes price increases and premium edition upgrades.

If agent credits do not replace the lost seat revenue, the slowest-growing and largest part of the company could turn flat.

Integration is the third. Informatica, Contentful, Fin and Braze all have to be folded into a company that is also changing its pricing model, retraining a large sales force on credits and cutting staff.

Amortization of acquired intangibles already rose to $522M a quarter ^[1]. Overlap is real: Fin's service agent competes with Agentforce Service, and Braze overlaps with Marketing Cloud.

Integration drag tends to show up first in sales productivity, which is already weak.

Gross margin is the fourth. Cost of subscription grew 22.86% in Q2 against revenue growth of 10.83% ^[1].

If inference costs keep rising faster than credit revenue, the margin expansion in our base case will not arrive.

Catalysts and conclusion

The next catalyst is the fiscal Q3 2027 print, expected around December 2, 2026 after the close; the company has not confirmed the date ^[11]. We will watch five things.

Organic growth, once Informatica's roughly 4 points are stripped out, against the 6.38% of Q2. cRPO growth against the guided 14%. The apps group growth, our seat proxy.

Cost of subscription growth against revenue growth. And the final ASR settlement, which sets the share count for fiscal 2028.

We will also look for three signals that would change the call. A move toward per-resolution or per-outcome pricing as a main way Agentforce is sold would flip our failing gate test to a pass.

A second quarter of apps group growth below 7% with no lift in premium edition bookings would point to seat compression arriving faster than credits can replace it. And any new borrowing for buybacks beyond the current authorization would push leverage past what we think owner cash can carry comfortably.

On the other side, a Q3 print that shows owner free cash flow holding near 26% of revenue with interest fully in the base would confirm the core of the thesis.

Other dated events: the Contentful and Fin closings, expected in the fiscal third quarter (unconfirmed until announced) ^[2]; the Braze shareholder vote and regulatory approvals, with closing expected in Salesforce's fiscal 2028 (confirmed in the merger agreement) ^[4]; and the fiscal Q4 print around late February 2027 (unconfirmed), where fiscal 2028 guidance will show whether operating cash flow growth recovers once interest is in the base.

Our conclusion: Salesforce is a mature franchise that has chosen leverage and acquisitions to keep growing per share, and its owner cash is large enough to carry the choice. The base case is 2.17x, the organic-only case 1.89x and the strict-anchor case 1.72x, all above the 1.5x line, and three of four gate tests pass.