AI Tokens Become a Core Line Item in Enterprise AI Spending
Salesforce’s plan to spend USD 300 million (approx. RM1.38 billion) on Anthropic AI tokens in a single year is a clear signal that corporate AI adoption has entered a new phase. Tokens are the discrete text units that Anthropic’s models process, and Salesforce will effectively treat this as a consumption-based AI infrastructure budget for its engineering workflows. Instead of AI being a side experiment, token capacity is now purchased much like cloud compute or storage. This marks a shift from occasional proof-of-concept pilots to continuous, large-scale usage embedded in production systems. With AI tools already driving more than 30% productivity gains for Salesforce’s roughly 15,000 engineers, spending heavily on AI token costs is emerging as a strategic alternative to adding more headcount. For technology leaders, this underscores that forecasting token consumption is becoming as important as planning server or database capacity.
From Hiring Engineers to Buying Tokens and Oversight Talent
Salesforce’s leadership has made the trade-off explicit: it will freeze engineering hiring for the first time in 2025, citing productivity boosts of over 30% from AI tools such as Anthropic’s Claude, OpenAI Codex, Cursor, and its own Agentforce platform. Instead of expanding developer headcount, the company is reallocating budgets toward AI token consumption and roles that supervise AI-generated code. Its engineers are increasingly shifting into oversight positions, reviewing and directing what AI systems produce rather than authoring every line themselves. At the same time, Salesforce plans to add 1,000 to 2,000 sales staff to help customers understand and adopt its AI products. This rebalancing suggests future technology teams will be smaller but more focused on governance, system design, and AI prompt strategy, while revenue-facing roles grow to monetise AI capabilities. For CIOs, workforce planning now means calibrating the mix of human engineers, AI agents, and client-facing experts.
AI Infrastructure Budget: From Experimental Spend to Mission-Critical Stack
Salesforce’s AI business unit, Agentforce, has already crossed approximately USD 800 million (approx. RM3.68 billion) in annual recurring revenue, and AI now powers an estimated 30–50% of the company’s overall workload. These numbers indicate that AI is no longer a niche experiment; it has become mission-critical infrastructure comparable to core CRM or collaboration platforms. The company is building dedicated AI infrastructure layers, including Headless 360, an API-first platform with more than 60 MCP tools that give agents like Claude Code direct access to Salesforce’s enterprise stack. Engineering teams are also exploring AI-powered coding integrations in Slack, bringing intelligent automation into everyday collaboration workflows. Meanwhile, AI agents have enabled Salesforce to cut support staff from 9,000 to 5,000, proving tangible operational impact. For enterprises observing this trend, the message is clear: AI infrastructure budgets will increasingly sit alongside primary cloud and application investments, not beneath them as experimental line items.
Optimising AI Token Costs and Redefining Vendor Relationships
Spending heavily on AI tokens forces enterprises to confront a new kind of cloud economics. Salesforce is already designing an “intermediate layer” to route simple tasks to smaller, cheaper models while reserving Anthropic’s Claude for more complex work. This routing strategy is essentially cost optimisation for AI token costs, mirroring how organisations previously learned to tier storage or balance workloads across different cloud services. At the same time, Salesforce’s multi-vendor approach—using Anthropic, OpenAI Codex, Cursor, and internal tools like Agentforce—points to a future where vendor relationships are defined by orchestration rather than single-provider lock-in. Enterprises will need governance frameworks to decide which model handles which workload, how usage is monitored, and how costs are controlled. As AI token consumption becomes a major component of enterprise AI spending, procurement, architecture, and finance teams will all have to collaborate on a unified AI strategy that aligns performance, risk, and budget.