DeepSeek Is Building Its Own AI Chip — And Nvidia Should Be Worried

DeepSeek just made its boldest move yet. And it has nothing to do with a new AI model.

The Chinese AI startup that rattled Silicon Valley in early 2025 is now quietly designing its own AI chip. Three sources familiar with the effort told Reuters on July 7, 2026. The chip is built for inference — the stage where a trained AI model actually answers your questions — not for training new models.

This is a big deal. Here's exactly what it means, why it matters, and what happens next.

DeepSeek AI Chip — Quick Facts

Detail	Info
Reported By	Reuters — July 7, 2026
Sources	Three people familiar with the effort (anonymous)
Chip Type	Inference chip (not training)
Effort Started	About one year ago (~mid-2025)
Current Stage	Early — discussions with chip design, foundry, and memory firms
Target	Reduce dependence on Nvidia and Huawei chips
Nvidia Stock Reaction	Slipped ~1.6% in pre-market trading after the report
DeepSeek Valuation	$52B–$59B (first-ever funding round of $7B reported)

What Is DeepSeek — And Why Does This Matter?

DeepSeek is a Chinese AI lab based in Hangzhou. In January 2025, it released its R1 model — a highly capable AI that cost a fraction of what American rivals spent to build. It was so efficient and so cheap that it sent US tech stocks into a tailspin overnight. Nvidia alone lost hundreds of billions in market cap in a single day.

Since then, DeepSeek has been seen as the most credible challenger to American AI dominance. It has operated on limited chip access — US export controls have blocked Chinese firms from Nvidia's most advanced processors. DeepSeek has been running its models on Nvidia's older China-specific H800 chips and Huawei's Ascend accelerators.

Now, DeepSeek wants to build its own silicon entirely. That would remove its dependence on both.

Why Is DeepSeek Building an Inference Chip — Not a Training Chip?

This is the most interesting part of the story. DeepSeek is not targeting training — the massive, expensive process of building an AI model from scratch. It is targeting inference — the process of running that model to answer user queries.

Here's why that choice makes sense:

• Inference is the fastest-growing segment of AI computing. As AI applications spread, more compute is spent running models than training them.
• Inference chips can be cheaper and less power-hungry than the high-end GPUs Nvidia sells.
• Inference is more forgiving on manufacturing process nodes — meaning DeepSeek does not need access to the most advanced chip fabs that US export controls restrict.
• DeepSeek runs inference at massive scale for real users. Optimizing that layer first makes financial sense immediately.

DeepSeek is essentially saying — we'll let others fight for training supremacy. We'll own the serving stack. That's where the money is going.

DeepSeek vs Nvidia vs Huawei — The Full Picture

	Nvidia	Huawei Ascend	DeepSeek Chip (Planned)
Primary Use	Training + Inference	Training + Inference	Inference only (so far)
Available in China?	❌ Most advanced chips banned	✅ Yes	✅ Would be fully domestic
Manufacturing	TSMC (advanced nodes)	SMIC (China)	TBD — foundry not named
DeepSeek Uses It?	H800, H100 (training)	Ascend (inference)	In development
Market Position	Global leader	~50% of China's AI chip market	Not yet released

Huawei currently holds roughly half of China's $50 billion domestic AI chip market — largely because Nvidia is banned. A working DeepSeek inference chip would directly challenge Huawei's grip on that market too, not just Nvidia's.

Who Else Is Building Custom AI Chips?

DeepSeek is not alone in this move. The entire AI industry is racing to control its own silicon.

• OpenAI — Unveiled "Jalapeno," its first custom inference chip, built with Broadcom. Announced in June 2026.
• Anthropic — Reportedly in discussions with Samsung over custom AI chips. Also signed a $19 billion, 20-year infrastructure deal with TeraWulf in Kentucky.
• Google — Has its own TPUs (Tensor Processing Units) — the most mature custom AI chip program in the industry.
• Amazon — Has Trainium and Inferentia chips for AWS workloads.
• Alibaba and Baidu — Both building their own AI chips in China, already gaining market share from Huawei.

The pattern is clear. Every major AI lab wants to control its hardware. The era of pure Nvidia dependence is ending — not because Nvidia is weak, but because everyone wants to own more of the stack.

What Are the Challenges for DeepSeek?

The project is ambitious. The obstacles are real. Here is what stands between DeepSeek and a working chip:

1. Manufacturing access is limited
US export controls bar Chinese companies from using the most advanced overseas chip foundries — specifically TSMC's leading-edge nodes. DeepSeek would likely need to work with China's SMIC, which is multiple generations behind TSMC on process technology. That limits what the chip can do and how efficient it will be.

2. High-bandwidth memory is restricted
Inference chips need high-bandwidth memory — a component that separate US rules have restricted China from accessing. Without cutting-edge memory, inference performance suffers significantly.

3. It takes years and heavy capital
Designing a competitive AI chip from scratch typically takes three to five years and hundreds of millions of dollars. DeepSeek started about a year ago. This is not a 2026 product.

4. No foundry partner named yet
Reuters' sources did not name a manufacturing partner, a process node, a timeline to first silicon, or a prototype. The project is at the discussion-and-hiring stage. Progress is real but early.

What This Means for Nvidia

Nvidia's stock slipped 1.6% in pre-market trading after the Reuters report. But analysts are divided on how much this actually threatens Nvidia.

Richard Windsor of Radio Free Mobile was blunt: "Nvidia is at zero in China and staying there. DeepSeek has almost no chance of selling silicon outside of China unless it gets access to leading-edge manufacturing."

That's the key point. DeepSeek's chip — even if successful — would likely be captive to China's market. Nvidia has already lost China. What DeepSeek's chip might do is reduce Huawei's dominance inside China — not Nvidia's global position.

The real risk to Nvidia is not one company's chip. It is the pattern. Every AI lab building its own silicon chips away at Nvidia's future growth ceiling. When enough of the industry's inference workload moves to custom silicon, Nvidia's addressable market shrinks — even if Nvidia remains dominant in training.

DeepSeek's First Outside Funding — $7 Billion

The chip news arrives alongside another milestone. DeepSeek — which had famously rejected outside investment for years — is reportedly raising $7 billion in its first-ever external funding round. The valuation sits between $52 billion and $59 billion.

That funding would give DeepSeek the capital to actually pursue chip development seriously. Hardware is expensive. Model research is cheap by comparison. This funding round signals that DeepSeek is no longer just a research lab — it is becoming an infrastructure company.

What Happens Next — What to Watch

The DeepSeek chip story is early. Here are the signals that will tell us how serious this gets:

• Which foundry gets named as the manufacturing partner — that is where export controls actually bite
• Whether future DeepSeek model releases are tuned specifically for their own chip architecture — a sign that software and hardware are being co-designed
• Whether the $7 billion funding round closes — capital is the fuel for chip development
• Nvidia's China inference revenue in coming earnings reports — if it drops, custom chip pressure is real
• Huawei's Ascend market share inside China — a DeepSeek chip would be Huawei's first serious domestic competitor

Final Verdict — DeepSeek AI Chip

DeepSeek building its own chip is not a surprise. It is the logical next step for any AI company serious about the long game.

The surprise is the timing and the approach. Targeting inference first — not training — is smart. It is the fastest-growing workload, the most cost-sensitive, and the one most accessible to a company with constrained chip manufacturing options.

This is not an immediate threat to Nvidia's global dominance. But it is another brick in the wall being built around China's AI ecosystem. DeepSeek wants to own its models, its serving stack, and now its chips. That is what self-sufficiency looks like at the frontier of AI.

The chip wars just got a new player.

Frequently Asked Questions (FAQ)

Q: Is DeepSeek building its own AI chip?

A: Yes. Reuters reported on July 7, 2026, citing three sources, that DeepSeek is developing a proprietary AI inference chip. The effort started about a year ago and is currently at an early stage.

Q: What kind of chip is DeepSeek building?

A: DeepSeek is building an inference chip — designed to run AI models and answer user queries — not a training chip. Inference is the fastest-growing segment of AI computing.

Q: Will DeepSeek's chip threaten Nvidia?

A: Directly, probably not in the near term. Nvidia has already lost access to China. DeepSeek's chip would primarily challenge Huawei's dominance inside China's $50 billion AI chip market. The broader pattern of AI labs building custom silicon does add long-term pressure on Nvidia's growth ceiling.

Q: When will DeepSeek's chip be ready?

A: No timeline has been disclosed. The project is still at early stages — discussions with chip design and foundry partners. Competitive AI chip development typically takes three to five years.

Q: How much is DeepSeek worth?

A: DeepSeek is reportedly raising $7 billion in its first external funding round at a valuation between $52 billion and $59 billion, reversing years of rejecting outside investment.

Q: Who else is building custom AI chips?

A: OpenAI (Jalapeno with Broadcom), Anthropic (discussions with Samsung), Google (TPUs), Amazon (Trainium), Alibaba, and Baidu are all building or operating custom AI chips.