US Firms Test China’s DeepSeek as Silicon Valley AI Costs Climb

US firms are testing China’s DeepSeek as the cost of Silicon Valley AI tools continues to climb, with DeepSeek topping Ramp’s June list of “trending software vendors.”

“Some companies have been so eager to use AI that they’ve gone overboard in a usage binge called ‘tokenmaxxing’ Global firms are starting to rethink their embrace of artificial intelligence as the costs of using the technology skyrocket amid shifting industry priorities and soaring demand for agentic AI”

Asia Financial

TechRepublic reported that Ara Kharazian, lead economist at Ramp Economics Lab, said US firms appeared to be making direct payments to DeepSeek rather than only running its open-source models on their own infrastructure.

Asia Financial

Kharazian warned that “some are willing to use cheaper, Chinese models, sending US data back and forth from China-hosted servers,” raising questions about savings, data residency, and risk.

The same TechRepublic report said DeepSeek’s corporate adoption on Ramp rose to 0.3% in January 2025 before falling back to 0.1%, and that by April 2026 DeepSeek’s adoption rate was still 0.1%.

By comparison, TechRepublic said Anthropic and OpenAI led the Ramp AI Index at 34.4% and 32.3%, respectively, in April.

Asia Financial and the South China Morning Post both described a shift as AI spending rises while productivity gains lag, with Uber’s chief operating officer saying the spending showed “no noticeable increase in productivity.”

Asia Financial said companies are rethinking AI use as costs skyrocket amid “tokenmaxxing,” where a usage binge can make “the cost of tokens exceed the cost of the employee within a month or two of use.”

South China Morning Post

The Asia Financial report tied the problem to agentic AI, saying agents “actually do things — book appointments, write code, manage files,” and that one task can spin up dozens of agents at once.

South China Morning Post similarly framed the pricing shift as AI companies charging “rock-bottom prices to hook customers after ChatGPT burst onto the scene,” before costs rose across the board.

Both outlets pointed to the same driver: AI agents are expensive to run because charges are measured in tokens.

WION said Google is promoting its Gemini 3.5 Flash as a more affordable option for businesses that want AI capabilities without the high costs tied to advanced systems.

WION reported that Google CEO Sundar Pichai said that if major Google Cloud customers shifted “around 80 per cent of their AI workloads to a mix of Gemini 3.5 Flash and other AI models,” they could collectively save “more than $1 billion annually.”

WION also said Pichai argued that organizations do not always need the most powerful AI model for every task and can reduce costs by using a combination of models based on specific needs.

Asia Financial described a parallel cost-control approach, saying some companies are switching to free, open-source AI models that anyone can download, while others move to smaller, more specialized models.

Asia Financial added that companies can break big AI tasks into smaller steps and “handing each piece to the cheapest model that can handle it,” with Adrian Balfour of Envorso saying “the big large monolithic model, it’s $15 per million tokens” but “you can get that down to like five cents.”