DeepSeek, Huawei and Nvidia Remain at the Center of the AI Chip Race

The global artificial intelligence race is increasingly becoming a race for computing power, and three names remain at the center of that competition: DeepSeek, Huawei and Nvidia.

In September 2026, the contest is no longer simply about which company can build the fastest AI model. It is also about who can design, manufacture, supply and operate the chips required to train and run those models at enormous scale.

Nvidia continues to generate extraordinary demand for its data-center hardware. Huawei is accelerating its domestic AI-chip roadmap as Chinese companies seek alternatives to U.S. technology. DeepSeek, meanwhile, is increasingly connected to China's broader effort to build AI systems around domestic hardware while reportedly exploring chips of its own.

The developments illustrate how AI infrastructure has become a critical part of the technology industry, alongside models, software, networking and data centers. For broader context, the Complete Guide to Emerging Technology and Innovation explores how advances across these areas are reshaping the technology landscape.

Nvidia Still Dominates the Global AI Infrastructure Story

Nvidia remains one of the most important suppliers in the global AI-computing market.

The company's latest financial results underline the scale of demand. Nvidia reported $96.2 billion in revenue for its fiscal second quarter ended July 26, 2026, up 106% from the same quarter a year earlier. Data Center revenue reached $89 billion, an increase of 117% year over year.

Nvidia expects fiscal third-quarter revenue of approximately $108 billion, plus or minus 2%. The company's forecast does not assume any Data Center compute revenue from China.

The company is also moving into its next major hardware generation. Nvidia says its Vera Rubin platform is entering full production, with systems being deployed by major cloud and infrastructure partners.

That gives Nvidia a substantial position in the global AI infrastructure buildout.

But the Chinese market presents a very different challenge.

Huawei Is Building a Domestic Alternative

Huawei has emerged as a major competitor in China's AI-chip market as U.S. export restrictions have made access to the most advanced Nvidia hardware more complicated.

In September, Huawei said demand for its AI computing equipment in China was greater than its production capacity. Rotating chairman Eric Xu said the company could not currently produce enough equipment even to meet domestic demand, limiting its ability to expand internationally.

Huawei is also accelerating its chip roadmap.

The company plans to launch the Ascend 960DT in the first quarter of 2027, earlier than its previous timetable. A related Ascend 960PR processor is planned for the third quarter of 2027. Huawei has also outlined longer-term plans for the Ascend 970 and 980 generations.

Rather than relying only on improvements in individual chips, Huawei is also focusing on connecting large numbers of processors into computing systems.

Its Peerium architecture is designed to connect very large numbers of processors, allowing system-level performance to compensate for some limitations at the individual-chip level.

That approach is important because AI performance increasingly depends on entire computing systems rather than a single accelerator.

DeepSeek Is Becoming a Hardware Story Too

DeepSeek initially became globally known for its AI models, but its role in the chip race has expanded.

In July, Reuters reported that DeepSeek was developing its own AI chip, with the project focused on inference rather than training. The reported objective was to reduce reliance on both Nvidia and Huawei hardware.

The development is significant because inference is the stage at which an already-trained AI model generates responses or performs tasks for users.

As AI usage expands, inference can require enormous computing capacity. A company that can optimize its models and hardware together may have opportunities to reduce costs or improve efficiency.

DeepSeek's reported hardware ambitions therefore suggest that the company is not simply competing at the model layer.

DeepSeek Is Also Turning to Huawei

At the same time, DeepSeek appears to be working closely with Huawei's hardware ecosystem.

Bloomberg reported in early September that DeepSeek planned to deploy at least 160,000 Huawei Ascend 950DT accelerators at a data center being developed in Inner Mongolia. If completed, the installation would represent one of the largest known clusters of Huawei AI chips.

The reported plan comes with an important qualification.

DeepSeek reportedly does not currently plan to use the 950DT chips for training its models, even though Huawei designed the processor for demanding AI workloads. DeepSeek has previously tried Huawei hardware for training but has continued to rely on Nvidia accelerators for that part of the process.

That distinction shows how complicated the transition away from Nvidia can be.

A company can deploy domestic chips for inference or other workloads while still depending on Nvidia for the most demanding training operations.

The Training Versus Inference Divide Matters

AI chips do not perform a single uniform job.

Training involves teaching a model by processing enormous amounts of data and adjusting its parameters. Inference involves using an already-trained model to produce outputs.

Both require substantial computing power, but their workloads and optimization requirements can differ.

This distinction helps explain why DeepSeek could simultaneously explore its own inference chip, plan a major Huawei deployment and continue using Nvidia hardware for training.

The AI-chip race is therefore not necessarily about replacing one processor with another overnight.

It can involve gradually moving different workloads onto different hardware platforms.

Nvidia's Software Advantage Remains Important

Hardware specifications alone do not determine the usefulness of an AI accelerator.

Software is equally important.

Nvidia's CUDA ecosystem has become deeply embedded in AI development, with developers, researchers and companies using its libraries and software tools to build and optimize applications.

Huawei is attempting to close that software gap as Chinese developers increasingly work with domestic accelerators. Huawei's leadership has argued that newer training tools are reducing Nvidia's historical software advantage, although Nvidia continues to have a significant established ecosystem.

This is one reason the chip race is difficult to reduce to simple comparisons of computing performance.

Developers also care about compatibility, programming tools, libraries, networking, memory, system integration and the ability to move existing workloads between machines.

Export Controls Have Changed the Competitive Landscape

The Nvidia-Huawei competition cannot be separated from U.S. restrictions on advanced semiconductor exports to China.

The United States has maintained export controls affecting advanced computing chips and semiconductor technology, with rules evolving over time.

In January 2026, the Bureau of Industry and Security said license applications for Nvidia's H200, AMD's MI325X and similar chips destined for China would be reviewed on a case-by-case basis if specified security requirements were satisfied.

Nvidia's own regulatory filings also describe the changing restrictions. The company says licensing requirements have affected products including the H20 and that a February 2026 license allowed small amounts of H200 products to be shipped to specified Chinese customers, subject to conditions.

The result is a market in which hardware availability is influenced not only by engineering and manufacturing capacity but also by government policy.

Huawei Faces Its Own Production Constraints

Domestic alternatives do not automatically solve supply problems.

Huawei has said that demand for its AI computing products already exceeds its ability to produce them.

That means Chinese AI companies can face a different type of bottleneck: having access to a domestic accelerator does not guarantee that enough units can be manufactured, packaged, supplied and deployed.

Advanced AI infrastructure requires more than processors.

Memory, networking equipment, advanced packaging, power systems and data-center construction all have to scale together.

This is why the semiconductor competition increasingly resembles an infrastructure competition.

The Race Is Moving From Chips to Complete AI Systems

The industry is gradually shifting away from thinking about AI accelerators as isolated components.

Modern AI infrastructure can involve thousands or even hundreds of thousands of processors connected through high-speed networking and supported by specialized memory and storage systems.

Huawei's SuperPod strategy reflects this shift, while Nvidia is also selling increasingly integrated platforms combining GPUs, CPUs, networking and software.

Nvidia's Vera Rubin platform, for example, incorporates multiple components designed to operate as an integrated AI computing system rather than simply as standalone graphics processors.

The ability to build and operate these large systems may ultimately be as important as the performance of an individual accelerator.

Other Chipmakers Are Trying to Capture Part of the Growth

The market is not limited to Nvidia and Huawei.

AMD continues to expand its AI accelerator business, while companies such as Broadcom, Marvell and major cloud providers are developing or supporting alternative approaches to AI computing.

Nvidia's enormous growth has also created opportunities for semiconductor companies that supply networking, custom silicon, memory and other components.

That makes the broader AI-chip industry more diverse than a simple two-company contest.

For example, the question of whether semiconductor demand is still accelerating extends beyond Nvidia itself, which is why Marvell Earnings Could Show Whether AI Chip Demand Is Still Accelerating is relevant to the wider infrastructure story.

AI Labs Are Looking at Custom Chips

AI developers are also becoming more interested in controlling the hardware underneath their models.

The logic is straightforward: if an AI laboratory can design or customize hardware around its specific workloads, it may be able to optimize performance, energy consumption and operating costs.

That does not necessarily mean every AI company will manufacture chips independently.

Some may work with semiconductor designers, cloud providers or specialized chip-development companies instead.

Anthropic's exploration of custom hardware illustrates this broader industry trend. The company's reported interest in alternatives to conventional Nvidia infrastructure is discussed in Anthropic Explores Custom AI Chips as Labs Seek Alternatives to Nvidia.

Nvidia's Guidance Is Becoming an Important Industry Signal

Because Nvidia sits at the center of so much AI infrastructure spending, its financial outlook provides a useful indication of how aggressively companies are continuing to invest in computing capacity.

Nvidia's August 2026 results showed just how strong that investment remains, with quarterly Data Center revenue reaching $89 billion. The company also projected $108 billion in total revenue for its next quarter.

The company's outlook is therefore closely watched by chip manufacturers, cloud providers, AI developers and investors.

The broader significance is explored in Nvidia Guidance May Matter More Than Earnings for the AI Industry.

China Is Building Around Its Own AI Ecosystem

The Huawei-DeepSeek relationship is part of a wider push by Chinese technology companies to develop more domestic alternatives.

Alibaba, for example, announced a new AI accelerator called the Zhenwu V900 on September 22, 2026, alongside plans for a much larger AI model. The company said the chip delivers triple the performance of its predecessor and is expected to enter mass production in early 2027.

That development shows that Huawei is not the only Chinese company working on AI silicon.

Chinese technology firms increasingly have incentives to develop models, chips, cloud infrastructure and software that work together.

The result could be a more vertically integrated domestic AI ecosystem.

The Global AI Chip Race Is Becoming More Fragmented

For years, the AI hardware story could largely be described through Nvidia's rapid expansion.

That picture is changing.

Nvidia remains deeply important to global AI infrastructure, with extraordinary revenue growth and continued demand for its newest systems.

But China's restrictions on access to advanced U.S. chips have also accelerated efforts to develop domestic alternatives.

Huawei is advancing its Ascend roadmap. DeepSeek is exploring its own inference hardware while reportedly preparing a massive Huawei-based deployment. Alibaba is developing new accelerators. Other semiconductor companies are targeting parts of the same expanding market.

The result is an AI-chip race increasingly defined by performance, availability, software ecosystems, manufacturing capacity, energy efficiency and geopolitical access.

What Comes Next for AI Hardware

The next phase of the competition is unlikely to be determined by one benchmark alone.

The companies that can combine powerful processors with efficient software, high-speed networking, reliable memory supply and massive data-center infrastructure will have important advantages in deploying AI at scale.

For Nvidia, maintaining its software ecosystem and accelerating new hardware generations will remain central.

For Huawei and other Chinese chipmakers, expanding production and developer adoption will be critical.

For DeepSeek and other AI labs, improving model efficiency and gaining greater control over inference and training costs could become increasingly important.

And for the wider technology industry, the AI-chip race means that semiconductor design is no longer simply a hardware story. It is becoming a defining part of how AI models are developed, deployed and scaled around the world.