DeepSeek Tops US Business Spending Index as Firms Ditch Pricey Silicon Valley AI

For a business running heavy-weight agent workflows or high-volume text generation, those multiples quickly translate to five- and six-figure annual savings. DeepSeek supports a 1-million-token context window and up to 384,000 output tokens, and it has adopted OpenAI and Anthropic API formats to make migration straightforward . The pricing move was not a temporary promotion; DeepSeek is betting it can sustain these rates, forcing every other model provider to either match the floor or lose volume on cost-sensitive workloads .

The Ramp Data: Direct Payments, Not Self-Hosting

Ramp’s June 2026 report analyzed credit-card spending records from more than 50,000 US companies on its platform. It found that DeepSeek had risen to the top of the "Trending" software vendor list, a measure of breakout growth relative to size among first-time business purchases, ahead of vendors such as PheedLoop and Fireworks AI .

The crucial detail is the payment pattern: the firms tracked were paying DeepSeek directly for its official hosted API. They are sending and receiving data through DeepSeek—meaning their prompts and outputs are passing through China-hosted servers—rather than deploying the open-source model on their own infrastructure . This direct API usage is a distinct signal from earlier open-weight adoption spikes, which tend to be driven by tinkerers or small-scale experiments .

Winners and Market Share Still Belong to the Incumbents

The trending index captures new business momentum, not total spend. Anthropic and OpenAI remain the dominant vendors by overall enterprise share, holding 34.4% and 32.3% of the tracked market, respectively . That gap signals that most large-scale production workloads are still running on established US providers, but it also shows that new procurement decisions, especially among mid-market companies and teams with limited AI budgets, are swinging toward the cheaper alternative .

The Risk: What Your Data Costs When the Tokens Are Cheap

DeepSeek’s pricing requires US firms to send data to servers in China, a fact that is drawing sharp warnings from security researchers. The 9to5Mac "Security Bite" report pointed out that any sensitive information employees feed to the model becomes part of the real cost of using the service, regardless of the per-token discount . Compliance teams dealing with regulations such as GDPR, sector-specific data-localization rules, or corporate data-classification policies face hard questions about what can legally be passed to a China-hosted inference endpoint .

So far, the spending data suggests that for many firms the cost differential outweighs the perceived compliance risk—or that the teams swiping corporate cards have not fully internalized where their data is going. Either way, the trend is unlikely to reverse without a major security incident or a regulatory intervention.

What This Means for AI Pricing Going Forward

DeepSeek's strategy is one of volume and commoditization at the API layer. The company’s smaller V4 Flash model is priced even more aggressively, undercutting Anthropic’s Claude Haiku by a factor of 10x to 25x on per-token basis . The economics of this approach squeeze margin out of the inference layer and push value further up the stack toward specialized applications and safety guarantees.

For corporate buyers evaluating their next AI contract in mid-2026, the question is no longer whether DeepSeek offers the lowest raw price—it does, by a wide margin—but whether the operational savings justify routing data through a vendor that is fully subject to Chinese data laws. The Ramp data indicates that for a growing number of US companies, the answer as of June 2026 is yes .