The move could balance server demand while courting global developers. Could it trigger a pricing war in an already cutthroat AI market?

DeepSeek, a rising Chinese AI startup challenging Western giants like OpenAI and Meta’s Llama, has unveiled an aggressive discount program offering up to 75% off API access for its models during off-peak hours—a tactical play to redistribute server loads while undercutting rivals on cost. The discounts apply daily between 16:30 and 00:30 UTC (12:30 AM to 8:30 AM Beijing time), a window strategically aligned with peak business hours in U.S. time zones. For developers in San Francisco or New York, this means cheaper rates during their workday while DeepSeek’s infrastructure idles overnight.

  • San Francisco (PST): 8:30AM to 4:30PM
  • New York (EST): 11:30AM to 7:30PM

The company’s R1 reasoning model—a rival to OpenAI’s o1—and its V3 general-purpose language model are central to the promotion. The R1 model sees steeper cuts (75%) compared to V3 (50%), reflecting its higher computational demands and DeepSeek’s push to expand adoption of its reasoning-focused tools.

While midnight in Beijing coincides with afternoon hours in North America, DeepSeek’s pricing may turn U.S. developers into bargain hunters for its API services if privacy isn't a concern. For startups or indie developers working with tight budgets—particularly those experimenting with AI-powered apps or chatbots—the discounts could lower barriers to entry significantly.

But there’s more beneath the surface. Like airlines offering cheaper red-eye flights to fill empty seats, DeepSeek hopes to flatten demand spikes straining its servers during peak hours in Asia.

DeepSeek’s R1 has drawn comparisons to OpenAI’s reasoning models for its ability to simulate step-by-step problem-solving before generating answers—a feature that requires more compute power but improves accuracy for tasks like coding or math problems. The V3 model, meanwhile, serves as a versatile workhorse for text generation and analysis.

As demand surges during peak hours—driven partly by China’s rapid AI adoption—DeepSeek has faced server bottlenecks reminiscent of OpenAI’s early scaling struggles. Last month, users reported sporadic outages during daytime .

What Comes Next

For now, DeepSeek seems focused on growth over short-term gains. The company plans is rushing to release an upgraded R2 model, alongside expanding its open-source offerings—a strategy that mirrors Meta’s Llama playbook but with sharper commercial hooks.

“...lowering the cost of AI will expand the market,” said Intel ex-CEO Pat Gelsinger during a recent interview about industry trends (Ars Technica).

The link has been copied!