Two high-stakes gambles in China's domestic chip industry: Cambricon bets on certainty, Moore Threads buys time

Cambricon and Moore Threads, one is located in Zhizhen Building, the other in Wangjing International R&D Park.

The two office areas are in Haidian and Chaoyang respectively. The gap between them is not only a spatial separation, but also a threshold from "loss-making" to "profitability".

In the second quarter, Cambricon's net profit attributable to shareholders was 2.311 billion yuan, while Moore Threads' net loss attributable to shareholders was 11.56 million yuan.

Cambricon has crossed this threshold, while Moore Threads is standing right at the door.

Taking profitability as the boundary, the semi-annual reports tell two different origins, two sets of countermeasures to face challenges in the same industry, as well as the phased results within a relatively short development cycle.

Two "Origins" of the Rising Star in the Spotlight

The gene of a company is written in the resume of its founder.

Chen Tianshi's resume is almost a vivid interpretation of the term "academic genius". He was admitted to the Special Class for Youngsters of University of Science and Technology of China at the age of 16, obtained his doctor's degree in computer science at 25, and became the youngest full researcher and doctoral supervisor at the Institute of Computing Technology, Chinese Academy of Sciences at 31.

The academic team led by him and his elder brother Chen Yunji (who also graduated from the Special Class for Youngsters of University of Science and Technology of China) had published dozens of top papers in the field of processor architecture before founding Cambricon in 2016.

Zhang Jianzhong's resume is that of a business elite. He joined NVIDIA in 2006 and rose from an ordinary employee to Global Vice President and General Manager of China Region.

During his 14 years of tenure, NVIDIA's GPU market share in China climbed from less than 50% to over 80%. In 2020, the 54-year-old resigned from the position of NVIDIA Global Vice President and founded Moore Threads in Beijing.

The owner of one office is from the academic school, moving from the laboratory to the market, while the owner of the other is from the industrial school, transitioning from a professional manager to an entrepreneur.

These two different origins determine two completely distinct underlying logics.

Cambricon, as the academic school, first believes in "technical barriers". The company spent almost all of its first six years doing one thing: independently developing the MLU instruction set.

This is an extremely time-consuming underlying project: the instruction set is the "language" of chips, which determines the efficiency that software can achieve on the chips.

This self-developed system has built Cambricon an intellectual property moat with a total of 2901 patent applications and 1834 authorized patents, but it also means that the company was almost "burning cash" in the first six years, and its accumulated loss exceeded 4 billion yuan when it went public in 2020.

Moore Threads, as the industrial school, first believes in "demand understanding". Zhang Jianzhong clearly knows what customers in the Chinese market need, including not only AI computing power, but also graphics rendering, physical simulation, and video codec.

Therefore, Moore Threads has positioned itself as a "full-function GPU" from the very beginning, aiming to develop platform-based products instead of making single-point breakthroughs, so as to secure customers first. Most of its core team members have NVIDIA backgrounds, and the company has launched five generations of chip architectures (Suti → Chunxiao → Quyuan → Pinghu → Huagang) in five years, catching up with accumulation at a fast pace.

As of June 2026, the company has filed a total of 2167 patent applications, with 788 patents granted.

The steady accumulation of the academic school and the fast attack of the industrial school present completely different efficiency indicators on the two financial statements: Cambricon has 1007 R&D personnel, and each person contributed 5.95 million yuan of revenue in the first half of the year.

Moore Threads has 1019 R&D personnel, with per capita revenue contribution of 1.7 million yuan, which is only 28.6% of that of Cambricon.

This set of figures does not reflect the gap in capabilities, but the difference in development stages. Cambricon's R&D has entered the "harvest period", while Moore Threads is still in the "key tackling period". The Huashan and Lushan chips under the "Huagang" architecture are still under development, and the scissors gap of input-output ratio has not yet been reversed.

The choices of the two enterprises have been different from the very beginning.

Two Solutions to the Same Test on the Entity List

In December 2022, Cambricon was added to the Entity List. Ten months later, in October 2023, Moore Threads also received the same administrative order.

This 10-month time difference forms a hidden watershed. Cambricon has been on the list for 10 more months, experiencing the whole process from supply chain shock to reconstruction, and then to active defense nowadays.

However, Moore Threads was added to the list later, and is still in the R&D investment period, with huge dependence on external funds.

Cambricon's answer is written in the balance sheet of its 2026 semi-annual report: inventory is 8.248 billion yuan, and prepaid expenses are 2.914 billion yuan. The two items add up to 11.16 billion yuan, accounting for 61% of the total assets. The raw materials in the inventory are 5.749 billion yuan, nearly doubling compared with the beginning of the year, most of which are wafers.

Under the restrictions of the Entity List, production capacity means orders. Cambricon's choice is to use cash to buy certainty, and lock in the production capacity for the next 12 to 18 months in advance, even if it needs to occupy more than 10 billion yuan of funds. The price is that the net cash flow from operating activities is 311 million yuan, down 66% year on year.

On the afternoon of August 12, Cambricon held a half-year performance briefing.

Investors' questions directly targeted the 8.248 billion yuan inventory and delivery rhythm. Chen Tianshi did not give a positive answer, only stating that "the company will continue to expand the market and control relevant risks".

On that day, the secondary market once rose by more than 5% during intraday trading, but suddenly dived at the close. The profit of 2.3 billion yuan was completely covered by the financial narrative of 8.248 billion yuan of inventory, and the market's concerns about inventory outweighed the recognition of profitability.

Essentially, this is an offensive bet that believes the certainty of future orders is strong enough; but the management's silence in response to inquiries has not dispelled the market's doubts.

Moore Threads' answer is reflected in another figure: its loss in the second quarter expanded to 40.92 million yuan, and its long-term loans reached 2.825 billion yuan, up 90% compared with the beginning of the year.

Developing "full-function GPU" consumes more capital for Moore Threads, and the certainty of its orders is not as good as that of Cambricon. Therefore, it needs funds to maintain operational security.

In the first half of the year, Moore Threads borrowed 1.821 billion yuan from banks, repaid 1.268 billion yuan, with a net borrowing of about 550 million yuan; the net outflow of operating cash flow was 2.169 billion yuan, and the gap was larger than that of Cambricon. Cambricon uses historical cash to lock in goods, while Moore Threads uses newly borrowed debts.

On the evening of August 9, when releasing its semi-annual report, Moore Threads announced that it planned to issue H shares and list on the main board of the Hong Kong Stock Exchange, only eight months after its listing on the Sci-Tech Innovation Board.

The reason is to deepen the international strategic layout, attract global R&D and management talents, improve governance level and core competitiveness, and complete the issuance at an appropriate time within 24 months from the date of approval at the general meeting of shareholders.

Facing the same test question, Cambricon has strong certainty of future orders, so it uses huge funds to lock in production capacity. Moore Threads is worried about uncertainty, so it retains flexibility and accumulates strength by borrowing. One believes in "certainty" so it locks in goods, the other worries about "uncertainty" so it borrows money.

Two Bets, Two Directions

Both companies are placing a big bet.

In July 2026, Cambricon launched an equity incentive plan worth about 5.7 billion yuan, covering 945 employees, more than 80% of its total employees.

The performance assessment targets are quite aggressive: the 2026 revenue shall not be less than 13.5 billion yuan, and the cumulative revenue from 2026 to 2028 shall not be less than 100 billion yuan; some senior executives also need to meet additional net profit assessment requirements.

This is a "military order": either achieve 100 billion yuan of revenue in three years, or the equity incentive will be invalid. At this stage, Cambricon's annual revenue is still less than 10 billion yuan, which is equivalent to betting the whole company's future on the certainty of domestic substitution, believing that the market is large enough and the demand is strong enough, and the scale can continue to expand as long as the production capacity keeps up.

Moore Threads places its bet in another direction: sprinting with the help of industry boom.

By the end of the second quarter, the monetary funds on Moore Threads' books dropped from 9.175 billion yuan at the beginning of the year to 6.492 billion yuan. Instead of locking in the 100-billion-yuan revenue target through equity incentive like Cambricon, it chose to open a second financing channel via H shares when its cash flow has not turned positive and its book is not yet profitable.

This is a high-stakes bet: believing that more capital can help it run faster on the full-function GPU track, and believing that Moore Threads' products can fully replace NVIDIA's products.

Both bets have their own costs.

The cost for Cambricon is that its balance sheet is "relatively heavy". Once the market demand fluctuates or the technical route changes, the 8.248 billion yuan of inventory may turn from "assets" into "burdens". The company has accrued 397 million yuan of inventory impairment loss in the first half of the year, and this part of the "cost" has begun to appear.

Moore Threads faces greater volatility. Its gross profit margin dropped from 78% to 57% in the first half of the year, and the cost growth rate of 245% far exceeded the revenue growth rate of 147%. The quality control pressure brought by mass production and the capital pressure to continue expanding R&D all put Moore Threads on the edge of risk.

Industry research institutions predict that in China's high-end AI chip market in 2026, the share of domestic solutions will be close to 90%, of which Huawei is expected to account for 62%, Cambricon about 14%, and the remaining share is jointly occupied by manufacturers such as Moore Threads, MiiX and Biren. This is a large enough market to accommodate multiple parallel technical routes.

But at the current stage, Cambricon's "academic school" route has made breakthroughs in specific tracks, which confirms that starting from underlying technologies, the full closed loop from laboratory to commercialization can be achieved.

Moore Threads' "industrial school" route still needs time. The early stage of pursuing large and full functions requires more time, more technical accumulation and more resource allocation. Moore Threads needs to cross that threshold before its cash flow is exhausted.

The steady accumulation of the academic school and the fast attack of the industrial school are both contributing to the future of domestic AI chips. This is a path that must be taken. Judging from the current situation, Cambricon's route seems to be a more realistic choice.

This article is from the WeChat Official Account "Shudu She", author: Hou Yu, published with authorization from 36Kr.