Deepseek Ventures Into AI Chip Development to Cut Hardware Dependence

WHO:

Engadget - DeepSeek, a Chinese company

Techmeme - DeepSeek, a Chinese startup

WHAT:

CNBC - DeepSeek's share of gateway tokens climbed between May and June 2026 on Vercel, indicating increased usage.

CNBC - Z.ai's GLM 5.2 model was released in June 2026 and saw daily token volume grow about 27x and the number of customers using it grow about 80x in its first full week after launch.

CNBC - Open-source Chinese models can be "60% to 90% cheaper" than leading models from Anthropic and OpenAI.

CNBC - Lindy CEO Flo Crivello stated, "We did it, and you could see that cost curve go down, like, crash to the ground," regarding the switch to DeepSeek's models.

TechRadar - ByteDance is developing its own AI chip to reduce Nvidia dependence, targeting mass production by late 2027.

CNBC - The share of tokens used by U.S. companies on Chinese AI models via OpenRouter has sat above 30% each week since February 8, with that figure rising as high as 46%.

CNBC - Open-source Chinese models can be "60% to 90% cheaper" than leading models from Anthropic and OpenAI.

CNBC - The performance of Chinese AI models is estimated to be "six to nine months" behind top U.S. rivals, with new open-source models proving capable for most tasks.

TechRadar - ByteDance is targeting mass production of its own AI chip by late 2027.

Ars Technica - DeepSeek's plan to develop its own AI chips is aimed at reducing dependency on Nvidia and Huawei.

Ars Technica - DeepSeek is reportedly developing its own artificial intelligence chip amid a technological arms race, aiming to reduce dependency on Nvidia and Huawei hardware.

TechRadar - ByteDance is developing its own AI chip to reduce Nvidia dependence, targeting mass production by late 2027.

CNBC - OpenAI announced it would limit the rollout of a new set of models at the request of the U.S. government.

CNBC - Export controls on Anthropic's Mythos and Fable models were lifted in June 2026 after a standoff between the Trump administration and the company.

CNBC - Kyle Chan from the Brookings Institution stated, "Chinese AI models are particularly attractive to American companies now as AI costs skyrocket."

CNBC - Cien Solon, CEO of LaunchLemonade, noted that GLM 5.2 is now in the top five models on their platform, indicating its growing acceptance.

CNBC - DeepSeek's GLM 5.2 model was released in June 2026 and saw the fastest adoption of any model tracked by Vercel in 2026.

CNBC - DeepSeek's share of gateway tokens climbed between May and June 2026 on Vercel, indicating increased usage.

CNBC - Z.ai's GLM 5.2 model was released in June 2026 and also saw rapid adoption, with daily token volume growing about 27x and the number of customers using it growing about 80x in its first full week after launch.

IJR - DeepSeek is reportedly developing its own artificial intelligence chip amid a technological arms race.

CNBC - Recent model releases from Chinese companies, including DeepSeek and Z.ai, are seen as highly competitive compared to leading frontier systems from Anthropic and OpenAI.

CNBC - The share of tokens used by U.S. companies on Chinese AI models via OpenRouter has been above 30% each week since February 8, with that figure rising as high as 46%.

CNBC - The average share of tokens used by U.S. companies on Chinese AI models via OpenRouter was 11% over the previous 12 months, falling to 4.5% in the first half of 2025.

CNBC - DeepSeek's GLM 5.2 model was released in April 2026 and saw the fastest adoption of any model tracked by Vercel in 2026.

CNBC - Lindy CEO Flo Crivello stated, "We did it, and you could see that cost curve go down, like, crash to the ground," regarding the switch to DeepSeek's models.

CNBC - DeepSeek's AI models are gaining traction among U.S. companies, with the share of tokens used by U.S. companies on Chinese AI models via OpenRouter exceeding 30% each week since February 8, reaching as high as 46%.

CNBC - Open-source Chinese models can be "60% to 90% cheaper" than leading models from Anthropic and OpenAI.

CNBC - AI startup Lindy moved 100% of its traffic from Anthropic's Claude models to DeepSeek, saving millions of dollars within months.

CNBC - DeepSeek's GLM 5.2 model saw a daily token volume growth of about 27x and an 80x increase in the number of customers using it in its first full week after launch.

Engadget - DeepSeek is developing its own AI chips.

Techmeme - DeepSeek is in the early stages of developing its own AI chip designed for inference.

Techmeme - The development aims to reduce reliance on Nvidia and Huawei hardware.

WHERE:

Engadget - China

WHY:

CNBC - The rise of Chinese open-source and open-weight models comes as the U.S. administration increasingly looks to regulate its most powerful AI models and considers how to halt the rapid adoption of alternatives from overseas.

CNBC - U.S. companies are becoming more cost-conscious in AI adoption due to rising costs associated with using advanced models from American companies.

CNBC - The rise of Chinese open-source and open-weight models is occurring as the U.S. administration looks to regulate its most powerful AI models and halt the rapid adoption of alternatives from overseas.

CNBC - U.S. companies are becoming more cost-conscious in AI adoption due to rising costs associated with using advanced models from American companies.

Techmeme - The motivation behind DeepSeek's chip development is to cut hardware dependence on Nvidia and Huawei.

HOW:

CNBC - The performance of Chinese AI models is estimated to be "six to nine months" behind top U.S. rivals, with new open-source models proving capable for most tasks.

Techmeme - The development is described as being in the early stages.

(UPDATED July 8, 2026 10:39 UTC)

DeepSeek's share of gateway tokens saw a notable increase between May and June 2026, reflecting a rise in usage on Vercel. Meanwhile, Z.ai's GLM 5.2 model, which launched in June 2026, experienced a staggering growth in daily token volume by about 27 times and an 80 times increase in customers during its first full week. This shift towards more affordable open-source Chinese models, which can be significantly cheaper than those from Anthropic and OpenAI, comes as the U.S. administration is considering regulations on its most powerful AI models.

(UPDATED July 8, 2026 09:43 UTC)

ByteDance is working on its own AI chip to lessen reliance on Nvidia, with plans for mass production by late 2027. Meanwhile, U.S. companies have been increasingly utilizing Chinese AI models through OpenRouter, with the share of tokens exceeding 30% weekly since February 8, and peaking at 46%. This shift is largely driven by the significant cost savings offered by open-source Chinese models, which can be 60% to 90% cheaper than those from leading U.S. firms, despite their performance lagging by six to nine months.

(UPDATED July 8, 2026 08:48 UTC)

ByteDance is planning to begin mass production of its own AI chip by late 2027, a significant move in the tech industry. Meanwhile, DeepSeek is also developing its own AI chips to lessen its reliance on Nvidia and Huawei. This shift comes as the U.S. administration seeks to regulate its most powerful AI models and curb the rapid adoption of foreign alternatives.

(UPDATED July 8, 2026 07:47 UTC)

DeepSeek is reportedly working on its own artificial intelligence chip as part of a technological arms race, with the goal of decreasing reliance on Nvidia and Huawei hardware. Meanwhile, ByteDance is also developing its own AI chip, aiming for mass production by late 2027 to lessen its dependence on Nvidia. These developments highlight a growing trend among tech companies to create in-house solutions for their AI needs.

(UPDATED July 8, 2026 05:52 UTC)

OpenAI has announced it will limit the rollout of a new set of models at the request of the U.S. government. Meanwhile, export controls on Anthropic's Mythos and Fable models were lifted in June 2026 following a standoff between the Trump administration and the company. Additionally, Kyle Chan from the Brookings Institution highlighted that Chinese AI models are becoming increasingly appealing to American companies as costs rise, despite their performance being estimated to be six to nine months behind top U.S. rivals.

(UPDATED July 8, 2026 02:50 UTC)

DeepSeek's GLM 5.2 model, released in June 2026, has achieved the fastest adoption rate of any model tracked by Vercel this year, while its share of gateway tokens also increased significantly between May and June. Similarly, Z.ai's GLM 5.2 model, also launched in June, experienced remarkable growth with daily token volume soaring about 27 times and customer numbers rising about 80 times in its first full week. This surge in Chinese open-source and open-weight models comes as the U.S. administration considers regulations on its most powerful AI models, even as the performance of these Chinese models is estimated to lag six to nine months behind their top U.S. counterparts.

(UPDATED July 7, 2026 21:24 UTC)

DeepSeek is reportedly developing its own artificial intelligence chip, joining a competitive landscape that includes recent model releases from Chinese companies like Z.ai, which are challenging leading systems from Anthropic and OpenAI. Meanwhile, U.S. companies have been increasingly utilizing Chinese AI models through OpenRouter, with the share of tokens surpassing 30% weekly since February 8, even peaking at 46%. This surge in adoption comes as the U.S. administration is considering regulations on its most powerful AI models to address the rapid rise of alternatives from abroad.

(UPDATED July 7, 2026 19:35 UTC)

In a significant shift, the average share of tokens used by U.S. companies on Chinese AI models via OpenRouter dropped from 11% to 4.5% in the first half of 2025. Meanwhile, DeepSeek's GLM 5.2 model, which was released in April 2026, has experienced the fastest adoption of any model tracked by Vercel this year. Lindy CEO Flo Crivello remarked on the successful transition to DeepSeek's models, stating, "We did it, and you could see that cost curve go down, like, crash to the ground," highlighting the impact of this change.

(UPDATED July 7, 2026 17:42 UTC)

DeepSeek's AI models are rapidly gaining popularity among U.S. companies, with the share of tokens used on Chinese AI models via OpenRouter surpassing 30% each week since February 8, and even reaching as high as 46%. This shift comes as companies look to cut costs, as open-source Chinese models can be significantly cheaper, estimated to be "60% to 90% cheaper" than leading models from Anthropic and OpenAI. In fact, AI startup Lindy has completely transitioned its traffic from Anthropic's Claude models to DeepSeek, saving millions of dollars in just a few months, while DeepSeek's GLM 5.2 model experienced a staggering 27-fold growth in daily token volume and an 80-fold increase in customers during its first week after launch.

(UPDATED July 7, 2026 13:44 UTC)

DeepSeek, a Chinese startup, is in the early stages of developing its own AI chips designed specifically for inference. This initiative aims to reduce reliance on hardware from major players like Nvidia and Huawei. The company is focused on creating a more independent technology landscape within China.