A sales rep would normally open the Salesforce screen first, then start preparing for a deal from there. Salesforce itself has begun treating that step as unnecessary. On August 26, 2026, the company announced an expanded partnership with Anthropic called "Claudeforce" and released a plugin that lets users operate its CRM (customer relationship management) system directly from within Claude. In the quarterly earnings released the same day, $2.53 of the $5.90 adjusted earnings per share came from gains on strategic investments, with Salesforce's Anthropic stake at the center of that figure. The decision to step its own app back from the front line and an earnings structure in which unrealized gains on a stake form the pillar of profit are not two separate stories.
37 Skills That Let You Prep for Deals Without Opening Salesforce
The plugin, called "Salesforce in Claude," comes pre-built with 37 sales skills covering things like deal preparation, deal health reviews, and pipeline analysis. Users can query, update, and run processes on CRM data using natural language directly from Claude's interface. In effect, Salesforce has put its entire CRM inside Claude, meaning there's no longer any need to even open the Salesforce app. The design shifts where a sales rep's work begins — from the Salesforce screen to Claude's chat window.
The reverse integration was rolled out at the same time. Under "Claude in Salesforce," Claude becomes one of the models available to Agentforce's reasoning engine, the Atlas Reasoning Engine, and is set as the default in Agentforce Vibes and Agentforce Coworker. Agentforce as a whole isn't becoming exclusively Claude-powered, but at the main entry points where customers first interact with it, Claude becomes the standard. The CRM called from within Claude's own product is also Salesforce — the two companies have set up a structure in which the same functionality is offered through two different entry points.
The connection runs through an MCP (Model Context Protocol) server, APIs, and CLI tools. Once an administrator sets up the connection and authentication once, no additional work is required per user. What matters here is permissions: the scope of data visible from the Claude side directly inherits whatever permissions that user already has within Salesforce. A sales rep can't get Claude to surface deals from another department that they don't have permission to view in Salesforce itself. For heavily regulated industries like finance and healthcare, there's also an option to place the model inside a Trust Boundary defined by Salesforce on Amazon Bedrock.
Rollout is staged. Pilot customers already had access as of August 2026, an open beta is set for September, and additional skills are slated to be added progressively throughout the latter half of 2026. The number 37 is a starting point, not a ceiling. That said, the pre-built design, while sparing adopting companies the work of writing their own prompts, also means Salesforce defines the content of those skills. How far companies will be able to customize the skills to match their own sales processes will only become clear once additional skills begin rolling out in the second half of 2026.
Two $300 Million Figures, From Entirely Different Sources
When talking about money flowing between Salesforce and Anthropic, there are two separate streams. Both are roughly $300 million, but they are opposite in nature. The first is equity investment: since Anthropic's 2023 Series C, Salesforce has invested more than $300 million cumulatively across multiple funding rounds. According to the company's disclosure as of July 31, 2026, the value of this stake has reached approximately $5.1 billion.
The second stream is usage fees. Salesforce plans to spend approximately $300 million on Anthropic API token usage in 2026, and Salesforce CEO Marc Benioff has said most of this is tied to coding work performed by AI development agents. That means the company is paying, in a single year of usage fees, an amount comparable to what it built up over three years of cumulative investment. These two figures are additive — one is not a subset of the other. The statement "Salesforce has put $300 million into Anthropic" means something entirely different depending on which of the two it refers to.
Most of the token spending Benioff cited goes toward internal coding work. Salesforce has already moved its own development pipeline onto Anthropic's models. On top of that, it is now recommending the same model to customers for sales functions as well.
As a result, Salesforce is both a customer and a shareholder of Anthropic. Annual payments in the range of $300 million boost Anthropic's revenue, and the resulting rise in the company's valuation, reflecting that revenue growth, feeds back into Salesforce's roughly $5.1 billion stake. A loop has formed between the two companies in which a portion of what Salesforce pays out comes back as unrealized gains. The meaning of this deal as a technology partnership and its meaning as a capital relationship cannot be read separately.
What Lifted the Earnings Wasn't Sales
Q2 FY2027 revenue came in at $11.35 billion, slightly above the market consensus of $11.32 billion. That's up 11% year over year, remaining performance obligations (RPO) — the equivalent of contracted backlog — stood at $66.3 billion, also up 11%, and AI-related ARR (annual recurring revenue) is approaching roughly $4 billion. Looking purely at the headline figures, this looks like a solid earnings report maintaining double-digit growth.
The issue lies in the breakdown of profit. Adjusted earnings per share came to $5.90. Of that, $2.53 came from gains on holdings in what's called the "strategic investment portfolio." According to SEC filings, gains from strategic investments this quarter totaled $2.613 billion, adding $2.53 to non-GAAP EPS ($2.43 on a GAAP basis).
The same filing also disclosed that the Anthropic stake alone generated $2.7 billion in unrealized gains (pre-tax), suggesting that Anthropic shares account for the majority of the portfolio's total gains. As a share of the total, that $2.53 represents 42.9% — meaning more than 40% of this quarter's per-share profit came from gains and losses generated entirely outside the software business.
Non-GAAP EPS excluding the contribution from strategic investments came to roughly $3.37, up about 16% year over year. The core business is growing. But the core business is not what produced the headline number in this earnings report.
Looking at absolute dollar figures, the structure stands out even more starkly. The $2.7 billion in unrealized gains on Anthropic shares equals 23.8% of the same quarter's revenue of $11.35 billion. An amount equivalent to nearly a quarter of a single quarter's revenue was generated purely through a revaluation of a stake, without selling anything to anyone. Motley Fool covered this earnings report from the angle that while the stock surged, $2.53 of the $5.90 in per-share profit came from investment gains.
The revenue beat itself was small. The $11.35 billion figure exceeded the $11.32 billion consensus by only $30 million, less than a 0.3% margin. Yet the stock moved more than 20%, and the reason lies not in the revenue figure but in how the profit was generated and the substance of the partnership.
CRM shares closed roughly 22.6% to 23% higher on August 27, 2026, the day after the earnings release. Pre-market and after-hours trading had reportedly seen moves of around 11% to 14%, meaning the gain widened further during the regular trading session. Still, that $2.53 figure is one that would move by the same amount if the valuation of the stake itself moved. How Anthropic's valuation is updated next is a variable entirely unrelated to how much Salesforce's sales team actually closes.
Why Salesforce Could Step Its Own App Down From Center Stage
In March 2023, Salesforce announced Einstein GPT, which incorporated OpenAI's technology, under the banner of "AI CRM." Since then, the company has maintained a posture of not committing to a single model provider. On October 14, 2025, it announced a partnership making Anthropic the recommended model for Agentforce, initially focused on deployment in regulated industries such as finance, healthcare, cybersecurity, and life sciences. Around the same time, it also expanded its partnership with OpenAI, rolling out an integration that lets users access Agentforce Sales functionality from within ChatGPT — moving from a late-2025 beta to general availability in the first half of 2026. And then, with Claudeforce in August 2026, the Claude-side integration went further still, adding 37 pre-built sales skills and making Claude the default model for the Atlas Reasoning Engine.
For the company that popularized SaaS-based CRM, users opening that screen every day has always been the underlying premise of its revenue model. Claudeforce removes that premise. With Salesforce itself building a workflow that bypasses its own UI, this decision carries more weight than a simple choice of model vendor. If a sales rep can finish deal prep just by talking to Claude, time spent on the Salesforce screen shrinks. Even so, the company made this move because, faced with a shift in which the first piece of software people open is moving toward chat-based AI, it chose to be inside that shift rather than defend its own app against it.
Agentforce isn't disappearing. Agentforce remains an agent framework that runs inside Salesforce's own product, while Claudeforce is a pathway that calls the CRM from inside another company's chat product. With two entry points now set up for the same functionality, it will be customer usage patterns, not Salesforce, that decides which becomes the primary battleground. In the earnings call, Benioff reportedly spoke about the significance of combining probabilistic intelligence with deterministically operating systems. In this division of labor, Claude handles the probabilistic side and CRM records handle the deterministic side — and in that arrangement, the former stands at the conversational front door.
Anthropic at 40%, OpenAI at 27%: The Enterprise Breakdown
According to a survey compiled by Menlo Ventures at the end of 2025, Anthropic accounted for roughly 40% of enterprise spending on LLM (large language model) APIs. That's up from 12% in 2023, and the same survey put OpenAI at 27% and Google at 21%. In the enterprise market specifically, the ranking of API spending destinations has already flipped. Claudeforce adds an entry point in the form of one of the world's largest CRM platforms to that existing share.
The move from 12% to 40% represents more than a threefold increase in roughly two years. That said, if OpenAI's 27% and Google's 21% from the same survey are combined, they total 48% — the gap with Anthropic alone remains relatively small. Claudeforce is a deal that brings one entry point into enterprise data into that tight contest.
The companies falling behind are also clear. Microsoft remains aligned with the OpenAI camp through Dynamics 365 Copilot but lags in the depth of integration with Salesforce's core CRM functionality. OpenAI itself maintains an integration that lets users access Agentforce Sales functionality within ChatGPT, but Anthropic is the only one that has gone as far as 37 pre-built skills and bidirectional model integration.
An analysis published on Yahoo Finance framed this deal as a signal marking the end of the model-agnostic era — a shift from a stage of simply recommending a particular model to a stage of building skills specifically for that model. Including Google, which is losing ground on the CRM front, the next contest will be whether the remaining players can build integrations of the same depth in other business systems.
For customer companies, this becomes a question of switching costs. The 37 sales skills are built to run on Claude, and if a company later wants to move to a different model, it will have to rebuild that entire set of skills. The more widely these are adopted, the larger that rebuilding effort becomes. Salesforce's decision to align its models with Claude also brings the same kind of lock-in to its customers.
With Pricing and GA Timing Still Undecided, What Japanese Customers Need to Check
The announcement doesn't reveal when Salesforce in Claude will reach general availability (GA). All that's specified is an open beta in September and additional skills rolling out progressively from the second half of 2026 onward — no GA date has been stated. The structure of who gets billed for token costs, however, is already visible. According to Salesforce, customers contract separately with Anthropic for Claude's inference usage, while Salesforce bills separately for headless consumption based on the edition of the Salesforce license the customer holds.
This is a dual billing structure that isn't settled in a single contract. Salesforce's own plan to spend roughly $300 million on tokens in 2026 is separate spending for internal development work, distinct from this customer-facing fee. What remains unclear is precisely the total cost level this dual-contract structure will add up to for adopting companies.
For Japanese companies using these services, the scale of these figures becomes more tangible when converted into yen. At a reference rate of 159.55 yen to the dollar (as of August 28, 2026), Salesforce's approximately $5.1 billion Anthropic stake comes to roughly 813.7 billion yen, and the 2026 token purchase plan of $300 million comes to roughly 47.9 billion yen. The quarterly unrealized gain of $2.7 billion on the Anthropic stake comes to roughly 430.8 billion yen, which accounts for most of the total strategic investment gain of $2.613 billion and drove more than 40% of a single quarter's earnings per share. For domestic companies using Salesforce, especially in finance and manufacturing, and organizations that use Slack daily, the reach of this partnership is broad.
On the data-handling side, two mechanisms are in place: inheritance of existing permissions and placement within a Trust Boundary. The very fact that Salesforce went to the trouble of designing a Bedrock-based deployment option for regulated industries is itself evidence that it anticipated strict scrutiny. Still, how long data passed to Claude is retained, and at what level of granularity companies can keep their own audit logs, will be among the first things Japanese financial institutions check against their internal regulations. Without clarity on this point, companies can't move from pilot to production.
Whether Claudeforce takes root in Japanese companies' operations won't be decided by the number 37. What matters is how the total cost of the dual-contract structure shakes out, when GA actually arrives, and how much audit logging companies can retain on their own end. All three of these will first come into view during the September open beta. Once a company confirms that all three fit within its own internal regulations, its sales reps will move toward a way of working in which deal preparation is completed without ever opening the CRM screen. Meanwhile, the capital relationship with Anthropic will remain in place, and the structure in which the valuation of that stake boosts profit every earnings cycle will carry over into the next quarter as well.