Four Chinese domestic GPU makers (Enflame, MetaX, Moore Threads, Biren Technology) reached combined market cap of 613.2 billion yuan following capital-market debuts, intensifying domestic competition.

On September 11, 2026, Enflame Technology made its debut on the Shanghai Stock Exchange’s STAR Market, surging 179.22% on its first day to reach a market capitalization of 170.8 billion yuan. With Moore Threads, MetaX, and Enflame now listed on the STAR Market, and Biren Technology trading on the Hong Kong Stock Exchange, China’s domestic GPU “Four Little Dragons” have officially converged in the capital markets. Based on closing prices on September 11, their combined market capitalization stands at 613.2 billion yuan (with Biren’s valuation converted using prevailing exchange rates). This marks a milestone for China’s computing power industry. Yet behind the milestone lies a set of data demanding closer scrutiny: Moore Threads, the first to list, has seen its stock price retreat more than 60% from its peak. While all four companies reported robust revenue growth in the first half of the year, none achieved profitability on a non-GAAP basis. Listing is not the finish line. Now that the question of “whether domestic GPUs exist” has been answered, the market’s next inquiry is “who will truly emerge victorious.”

The Four Little Dragons have taken four distinctly different paths to the capital markets. Enflame Technology represents the extreme case of deep strategic binding. Tencent serves as both Enflame’s largest shareholder and its primary customer. Between 2024 and 2025, Tencent’s share of Enflame’s revenue doubled from 37.77% to 83.79%, while the company’s gross margin contracted from 40.78% to 31.78% (a decline of 9 percentage points)—a timeline that provides the clearest evidence of rising concentration and diminishing bargaining power. In the first half of 2026, Enflame’s gross margin stood at just 30.37%, slightly more than half of MetaX’s 57.2%. H1 revenue reached 1.12 billion yuan, up 279.08% year-over-year, yet the net loss attributable to shareholders widened to 632 million yuan (up 3.57% YoY). Operating cash flow stood at -878 million yuan, and the asset-liability ratio reached 78.70%.

However, a positive metric often overlooked is Enflame’s contract liabilities of 6.665 billion yuan as of June 30, a staggering 18.26-fold increase from year-end 2025. Tencent is using advance payments to lock in production capacity for Enflame, providing exceptionally high order visibility. Yet contract liabilities are debts; the 6.665 billion yuan in prepayments inflates the asset-liability ratio. The company itself noted in its loss explanation that interest expenses accrued on these prepayments contributed to the widening deficit. Advance payments prove order visibility but simultaneously erode current-period profits, and because these funds are tied to Tencent’s dedicated scenarios, they cannot be easily replicated as generic market orders. Enflame’s prospectus forecasted a H1 2026 net loss between 608 million and 577 million yuan; the actual figure of -632 million yuan missed the lower bound and failed to narrow. Consequently, the Q1–Q3 forecast (-860 million to -700 million yuan) should be viewed with caution.

MetaX pursues a diversified layout strategy. Its top five customer concentration sits at approximately 61%, the lowest among the four. In H1 2026, MetaX generated 1.324 billion yuan in revenue with a gross margin of 57.2%, up 1.1 percentage points year-over-year. Net profit attributable to shareholders reached 612 million yuan (including 661 million yuan in non-recurring gains), leaving a non-GAAP loss of 49 million yuan. Notably, MetaX achieved a single-quarter non-GAAP profit of 54 million yuan in Q2. Investors should note that single-quarter profitability may be influenced by the timing of project deliveries and acceptances, and may not represent a sustainable steady state. MetaX also features a unique valuation adjustment mechanism: controlling shareholder Chen Weiliang and concerted parties commit to extending the lock-up period by 12 months if the company fails to achieve profitability or sees its non-GAAP net profit drop by more than 50% compared to the pre-IPO baseline in any given year from 2026 to 2028. There is currently ambiguity regarding whether “unprofitable” refers to GAAP or non-GAAP metrics; under the GAAP lens, H1 2026’s +612 million yuan likely meets the threshold, whereas the non-GAAP metric presents a different picture.

Moore Threads boasts the largest revenue scale at 1.736 billion yuan. Its gross margin declined from 69.14% in the prior-year period to 56.95% in H1 2026, dropping sharply to 49.26% in Q2 alone. This is not simply a case of products failing to generate margins; in Q2, Moore Threads recognized revenue from a 660 million yuan Kuai’e AI computing cluster order, accounting for nearly 70% of quarterly revenue. While chip cards themselves carry a gross margin of roughly 70%, cluster projects involve outsourced integration, server assembly, and system debugging, which compress margins. This reflects a strategic trade-off: selling chips for profit while selling systems to demonstrate capability. However, heavily absorbing low-margin system integration work consumes cash flow, ties up delivery resources, and squeezes resources for core chip R&D iterations—the inherent cost of this strategic pivot. Moore Threads recorded a non-GAAP net loss of 151 million yuan in H1 2026, narrowing year-over-year. Non-recurring gains totaled 139 million yuan (primarily government subsidies). Operating cash flow saw an outflow of 2.169 billion yuan, concentrated in Q1 (-1.487 billion yuan) but narrowing to approximately -682 million yuan in Q2, signaling an improving trend. Accounts receivable stood at 880 million yuan, up 102.89% year-over-year, warranting close monitoring of collection quality. Nevertheless, the company’s ammunition reserves remain ample: its A-share IPO raised 8 billion yuan, with approximately 5.6 billion yuan still held in dedicated accounts as of June. The short-term risk of “burning through runway” is unfounded.

Biren Technology experienced the most explosive revenue growth. H1 2026 revenue reached 1.236 billion yuan, a 1,997.56% year-over-year increase. However, with last year’s base at merely ~59 million yuan, a near-20x surge reflects a low starting point rather than a ceiling. Biren posted a loss of 377 million yuan during the period (IFRS basis; adjusted loss was 337 million yuan, a 38.9% reduction), while its gross margin improved to 42.7%, up 10.8 percentage points year-over-year—the strongest improvement among the four. It maintains a focus on cloud-based large-scale training chips, having completed certification with key internet enterprise clients. Inventory stands at 1.21 billion yuan, with prepayments reaching 1.534 billion yuan (up 230% to secure production capacity).

*Note on accounting comparability:* Moore Threads figures reflect non-GAAP net profit attributable to shareholders; MetaX presents both GAAP and non-GAAP metrics; Biren’s figures follow HKEX IFRS periodic loss standards (adjusted loss is non-IFRS); Enflame uses net profit attributable to shareholders. Cross-market comparisons require careful attention to these methodological differences.

In 2022, Nvidia commanded approximately 95% of China’s AI chip market. By 2026, that figure had plummeted to roughly 8%—a concession publicly acknowledged by Jensen Huang in May, who admitted yielding the Chinese AI chip market to Huawei. Domestic AI accelerator share has surpassed 60%, with Huawei’s Ascend platform capturing an estimated 50% to 62%. In other words, domestic substitution is no longer a theoretical opportunity; it is a reality that has already unfolded for the most part. Under the old framework, the fading of scarcity premiums would signal the falsification of substitution narratives. The actual data tells a different story: the scarcity premium has faded because substitution is largely complete, and the Four Little Dragons are not the primary beneficiaries. The pressing question has shifted to: once the cake is divided, what slice do you get?

Morgan Stanley’s 2026 forecasts indicate that within the domestic AI accelerator pie, Ascend holds roughly 62% and Cambricon accounts for about 14%, resulting in a CR2 of 76% and CR4 of 86%. When expanded to the full market including overseas vendors, Nvidia’s share falls to ~8% and AMD to ~12%. The Four Little Dragons must now compete alongside T-Head and Kunlunxin for the residual share. For context, Cambricon reported H1 revenue of 5.996 billion yuan and net profit of 2.311 billion yuan as of September 11; Hygon Information’s market capitalization stood at 536.666 billion yuan on the same date. These represent the true heavyweight tier of China’s AI chip sector.

As the Four Little Dragons vie for residual market share, understanding the magnitude of that remainder and identifying which companies can successfully navigate it requires a direct reference point: Cambricon. Consider a market-cap-aligned comparison: The Four Little Dragons’ combined market capitalization of 613.2 billion yuan corresponds to H1 revenue of 5.416 billion yuan and a simple-summed loss of 409 million yuan (using GAAP/periodic loss metrics; strict non-GAAP adjustments would reveal a significantly larger deficit). Cambricon, by contrast, reported H1 revenue of 5.996 billion yuan and net profit of 2.311 billion yuan on September 11. Cambricon’s half-year net profit exceeds half of the Four Little Dragons’ combined H1 revenue. Furthermore, while the Four Little Dragons’ H1 revenue grew at a rate of approximately 174.8%, Cambricon’s growth stood at roughly 108.13%—a mere 1.6x difference that hardly explains the chasm between profitability and losses.

Cambricon’s half-year revenue alone surpasses the combined half-year revenue of the Four Little Dragons on a comparable basis. The core diagnostic here is not operational inefficiency, but scale disparity. MetaX, despite boasting the highest gross margin, demonstrates that the issue is not unit economics but volume. MetaX’s 57.2% gross margin already exceeds Cambricon’s 55.25%. MetaX’s losses stem not from unprofitable products, but from 525 million yuan in R&D costs spread across 1.324 billion yuan in revenue (39.65%), whereas Cambricon spreads 702 million yuan in R&D across 5.996 billion yuan in revenue (11.72%). However, Biren’s 42.7% and Enflame’s 30.37% margins remain below the benchmark, indicating that for these two, insufficient scale coexists with product-mix challenges.

Cambricon’s 2024 performance mirrors today’s individual Four Little Dragons: revenue of 1.174 billion yuan and a net loss of -452 million yuan, closely matching the current single-company revenue range of 1.12 billion to 1.736 billion yuan. Within a year, Cambricon scaled to 6.497 billion yuan in revenue and 2.059 billion yuan in profit. This is precisely the option the market is pricing: a wager that the Four Little Dragons can replicate this turnaround-and-scaling trajectory within two years. Regarding absolute R&D expenditure: the two highest R&D spenders among the Four Little Dragons already exceed Cambricon’s per-period R&D outlay, yet deliver only one-fifth of the output. A crucial caveat applies: per-period R&D measures input flow, not accumulated stock. Cambricon’s current revenue cashes out years of cumulative investment. This comparison highlights not inefficiency, but a phase divergence between investment and harvest cycles.

A reverse insight emerges in customer concentration: Cambricon’s largest customer share dropped from 79.15% in 2024 to 26.21% in 2025, with four new clients entering its top five, signaling deliberate diversification. Conversely, Enflame’s Tencent share climbed from 33.34% to 37.77% and finally to 83.79%, moving toward concentration. While one is dispersing risk and the other consolidating it, the risk profiles differ fundamentally: Cambricon’s diversification spans multiple government and computing platform entities, whereas Enflame’s concentration revolves around a single entity that is both a major shareholder and primary customer.

The Four Little Dragons’ listing-day performances once epitomized capital market euphoria: MetaX surged 692.95%, Moore Threads jumped 425.46%, and Enflame climbed 179.22%. Euphoria, however, gives way to recalibration. Moore Threads’ stock has retreated from a peak of 941 yuan to an intraday low of approximately 345 yuan on September 11 (closing at 357.85 yuan), marking a drawdown exceeding 60%. MetaX fell from 1,033 yuan to 489 yuan, adjusting downward by more than 50%.

On September 7, Moore Threads faced its first major lock-up unlock. 25.77 million shares entered circulation, instantly expanding the free float by 85.27%. Trading halted at the 20% daily limit down, with the top five selling seats on the Dragon Tiger List all belonging to institutional desks, collectively offloading 1.462 billion yuan. Institutional liquidation at this juncture is logical: even at the limit-down price, early-stage IPO investors retain paper profits exceeding 260%. September’s unlock is merely the opening act. On December 7, Moore Threads faces another ~186 million share unlock (pre-IPO old shares), representing 39.55% of total equity—seven times the September volume. MetaX encounters a 13.966 million share unlock on September 17, expanding its float by 75.4% (lower than Moore Threads’ 85.27%), but against a much smaller absolute float (only 18.53 million shares pre-unlock), compounded by seven consecutive days of declines totaling over 20% and fragile market sentiment. The real pressure stems from a thin float and weakened psychology. Looking further ahead, MetaX faces a massive 68.27% unlock in March 2028.

Yet unlocks do not equate to guaranteed downside. A-share regulations strictly govern block trades, auction ratios, blackout windows, and pre-disclosure requirements; legacy shareholders may also choose to hold. Nevertheless, Pre-IPO original shareholders operate with minimal cost bases and extended investment horizons, creating inherently strong exit demands. Even via block transfers, acquiring institutions will face future sell-side pressure, meaning mid-to-long-term overhead supply is merely deferred, not eliminated. Two additional supply-side variables warrant attention: Moore Threads approved an H-share issuance in August 2026, targeting up to 10% of post-issue equity with a 15% greenshoe option. MetaX is also planning a Hong Kong listing. While these provide financing ammunition, they also introduce 12-to-24-month equity dilution and incremental supply headwinds.

A fundamental technological schism divides the Four Little Dragons. Moore Threads, MetaX, and Biren employ self-developed software stacks that maintain varying degrees of compatibility with the CUDA ecosystem (Moore Threads additionally targets graphics rendering, claiming full-function GPU capabilities). Enflame pursues a fully non-CUDA-compatible DSA (Domain-Specific Architecture) route. CUDA compatibility remains the critical lever for short-term commercial deployment, as the vast majority of enterprise clients already possess CUDA-native codebases. Without compatibility, securing scaled orders is nearly impossible. However, API/ecosystem compatibility does not equate to architectural independence. Long-term uncertainties surrounding patents and ecosystem lock-in persist. The compatibility strategy is fundamentally a commercial calculation, not a declaration of full ecosystem independence.

The DSA route carries pronounced vulnerabilities: independent operators and programming models force external clients to massively rewrite operators and rebuild applications, imposing migration costs far higher than CUDA-compatible alternatives. Enflame’s ability to navigate this path relies heavily on Tencent’s joint adaptation scenarios. Translating this co-validated product capability into broader, independent client adoption is the pivotal challenge in transitioning from dependency to symbiosis. Supply chain constraints also vary significantly. Tight HBM (High Bandwidth Memory) supply exerts a vastly different impact on Biren’s ultra-large-scale training cards compared to inference-focused manufacturers. HBM price escalations and foundry quoting volatility directly compress gross margins. These invisible barriers constitute the “hidden reefs” the Four Little Dragons must navigate beyond their financial statements.

The convergence of China’s domestic GPU “Four Little Dragons” in the capital markets represents a phase-specific answer sheet for this generation of Chinese chip entrepreneurs. Yet capital markets possess finite patience. Moore Threads’ 60% drawdown, the impending December unlock, MetaX’s Q2 non-GAAP turnaround amid overall H1 losses, and Enflame’s widening deficits paired with exploding contract liabilities all convey a singular message: listing merely secures a seat at the table. The actual competition to determine hierarchy has only just begun. Market pricing implicitly embeds an option: a wager that the Four Little Dragons can replicate Cambricon’s 2024-to-2025 trajectory of turning profitable while scaling volumes. The exercise conditions for this option require stitching revenue, gross margin, cash flow, and net profit into a sustainable chain—a premise that assumes product viability, ecosystem resilience, and customer structure do not deteriorate. Volume expansion, after all, does not automatically guarantee commensurate improvements in gross margins or profitability.