Speculation that DeepSeek would announce its new model during the Chinese New Year did not materialize, with the Chinese company revealing the DeepSeek V4 only in April, last Friday (24). However, the company returned to the spotlight with the news that the new model has been adjusted to (also Chinese) chip technology.
With the launch, the market’s expectation is that the movement will usher in a new wave of challenges to North American technological sovereignty. By adapting its model to Huawei’s infrastructure, it signals a more independent path from American giants — especially from Nvidia, currently dominant in supplying AI chips.
In practice, this could accelerate a reorganization of the sector, with China seeking to reduce its external dependence and strengthen its own ecosystem, capable of competing on equal footing in the global race for artificial intelligence.
Continues after advertising
China was already seeking to reduce its external dependence
It is worth remembering that DeepSeek’s new partner, Huawei, was at the center of one of the biggest technological tensions in recent years, when it became the target of sanctions imposed by the United States during Donald Trump’s government, starting in 2019. At the time, the company was included in a list of commercial restrictions under allegations of risks to national security — accusations that the company has always denied —, which limited its access to American technologies and suppliers.
One of the most visible impacts of this dispute was felt directly by consumers: Huawei’s latest smartphones began to be sold without Google services, such as Play Store, Gmail and YouTube. The measure strongly affected the brand’s global competitiveness in the mobile segment and marked a turning point in the company’s strategy, which began to invest even more in its own alternatives and in an independent ecosystem.
Without access to Google services and with restrictions on the use of advanced chips, including Kirin processors manufactured by Taiwan Semiconductor Manufacturing Company (TSMC), the company quickly lost ground in the global smartphone market. The period between 2021 and 2022 was marked by a significant drop in sales and revenue, ending a cycle in which Huawei came to compete for global leadership in the sector.
Faced with this scenario, the company’s response was to accelerate its technological independence strategy. Huawei started to invest heavily in Research and Development, with around CN¥ 161.5 billion (US$23.2 billion, at the time) invested in 2022, and made progress in creating its own ecosystem, with emphasis on the HarmonyOS operating system and cloud solutions.
At the same time, the company diversified its operations into areas less dependent on American technology, such as network infrastructure, software and even intelligent automotive solutions. This repositioning began to bear fruit: between 2024 and 2025, Huawei returned to growth, driven by launches such as the Mate 60 line (still without Google services), developed with chips produced internally in partnership with Semiconductor Manufacturing International Corporation (SMIC), the largest semiconductor (chip) foundry in mainland China.
In this context of resumption and strengthening of its own ecosystem, experts and entrepreneurs are already beginning to point to a deeper change in the global dynamics of technology.
Continues after advertising
What’s new in DeepSeek V4?
In addition to the partnership with Huawei, DeepSeek V4 arrives on the market as a relevant leap in scale and efficiency. The startup introduced two versions, V4 Pro and V4 Flash, each with unique features that meet different needs.
To understand better, the V4 Pro is the most robust version. It is designed to deliver the maximum performance possible, focusing on more complex tasks such as advanced reasoning, heavy programming and applications that require greater precision. It is the model that competes directly with the most advanced systems on the market, costing US$ 1.74 per million entry.
V4 Flash is the lightest and most economical version (priced at US$0.14 per million input tokens). It has been optimized to be faster and cheaper to run, whilst maintaining a good level of performance. The proposal here is to meet broader use cases, the famous “cost-benefit”.
Continues after advertising
Among the main new features is a new architecture, called Hybrid Attention Architecture, which improves the model’s ability to maintain context in long conversations. This adds up to a context window of up to 1 million tokens, allowing you to process entire codebases or large documents in a single command.
Furthermore, V4 uses the mixture-of-experts technique, activating only part of its parameters in each task. In theory, this translates into lower prices compared to higher-end models from American companies.
Even though it still lags a few months behind the most advanced models on the market in terms of capacity, the DeepSeek V4 increases global competitive pressure. The company claims that it can rival systems from giants such as OpenAI, Google and Anthropic in specific benchmarks, while continuing to focus on reducing costs and expanding access.
Continues after advertising
Content produced by .
