DeepSeek DSpark Cuts AI Response Time by 85% Without Advanced Nvidia Chips

As demand for AI models continues to grow, companies face a critical problem: computing resources. Data centers need thousands of the most advanced GPUs to run large language models. For Chinese AI companies like DeepSeek, advanced AI chips from Nvidia are largely inaccessible due to US export controls .

Now, the Chinese startup claims it has found a way to make its AI models respond much faster—without needing the most advanced chips .

Introducing DSpark

DeepSeek has unveiled DSpark, a speculative decoding framework for its flagship V4 model family. The company says it can speed up AI responses by as much as 85% . For example, a single GPU that previously handled 100 user queries could process about 185 with DSpark .

The framework is aimed at speeding up AI inference —the time an AI model takes to respond to a query—which is often seen as a major bottleneck in serving AI models .

How It Works

AI models generate text one token at a time, which becomes slow and wasteful when responses are long. DSpark addresses this with speculative decoding :

1. A lightweight draft model quickly proposes responses

2. The main model verifies them in batches rather than generating everything from scratch

If the draft created by the smaller model is correct, the system skips ahead. If it's not, it falls back. According to DeepSeek, most tokens are easy to predict, so the system can often move ahead efficiently .

The framework also uses a semi-autoregressive generation method —instead of generating responses token-by-token, it can produce small chunks of tokens at a time, making the process quicker .

Open-Source and Wider Implications

DeepSeek has open-sourced its DSpark research—a joint effort with Peking University—on GitHub and HuggingFace . The company tested the framework on several open-source models, including Google DeepMind's Gemma and Alibaba's Qwen, suggesting the gains could be applied more widely .

DSpark does not improve a model's general capabilities, but it could help companies get better performance without large additional investment in computing resources .

The Context: Rising Compute Costs

The breakthrough comes at a critical time. AI companies are spending billions of dollars to acquire more compute for data centers. At the same time, companies like Uber and Walmart are restricting AI token usage for employees due to rising costs .

In April 2026, DeepSeek open-sourced V4 Preview, positioning the model family as a cost-effective option for handling 1 million-context inputs. The company said V4-Pro is aimed at higher performance, while V4-Flash is designed as a faster and cheaper option .

The Broader Trend

DeepSeek is not alone in working on improving AI response speed. Earlier this month, Xiaomi's AI team said it had raised the output speed of its MiMo-V2.5-Pro-UltraSpeed model to more than 1,000 tokens per second —among the fastest speeds in the industry .

What This Means for the Industry

For Chinese AI companies operating under US chip sanctions, DSpark represents a potentially significant workaround. By reducing the hardware requirements for fast inference, the framework could:

• Lower the barrier to entry for AI deployment

• Reduce operational costs for AI companies

• Enable faster AI responses on existing hardware

• Make AI more accessible in markets where advanced chips are restricted