The dense disclosure of first-half 2026 financial reports from listed automakers has laid bare the cash reserves and R&D spending of China’s EV startup cohort. A comparison of Li Auto, NIO, XPeng, Leapmotor, and Seres shows significant divergence in both financial strength and R&D strategy: Li Auto (LI) sits firmly atop the cash pile with 87.5 billion yuan (approximately $13.0 billion), while XPeng (XPEV) emerged as the biggest R&D spender among the startups with 5.82 billion yuan (approximately $867.4 million) in research expenses.
On cash reserves, Li Auto leads with 87.5 billion yuan, though that’s down 18.1% year-on-year. China’s Seres (601127.SS) follows with approximately 73.15 billion yuan (approximately $10.9 billion). NIO (NIO) ranks third at 56.7 billion yuan (approximately $8.5 billion), having doubled its cash position from a year earlier. XPeng holds 40.48 billion yuan (approximately $6.0 billion), down 14.9% year-on-year. Leapmotor (9863.HK) has 38.59 billion yuan (approximately $5.8 billion), up 30.5% year-on-year.
On R&D spending, XPeng’s first-half research expenses reached 5.82 billion yuan, up roughly 39% year-on-year, with the highest R&D-to-revenue ratio among the startups. Li Auto spent 5.498 billion yuan (approximately $819.4 million) on R&D, up 3.3% year-on-year. NIO’s R&D expenses came to 4.03 billion yuan (approximately $600.6 million), the only company among the five to post a year-on-year decline. Seres invested 3.734 billion yuan (approximately $556.5 million) in R&D (excluding capitalized development expenditures), up 34.8% year-on-year. Leapmotor’s R&D expenses totaled 2.32 billion yuan (approximately $345.8 million), up 22.8% year-on-year.
Where is the money going?
The allocation of R&D resources reflects starkly different strategic choices at each company.
Li Auto is channeling funds heavily into chips and large AI models. Its self-developed 5nm automotive-grade Mach M100 chip has shipped over 50,000 units and is now deployed in production vehicles, with the Mach VLA large model rolling out in tandem. Li Auto CFO Li Tie said on the Q2 earnings call that full-year 2026 R&D spending is expected to reach 12 billion yuan (approximately $1.8 billion), with AI-related investments accounting for roughly 50%.
XPeng Chairman He Xiaopeng said on the Q2 earnings call that R&D investment is primarily directed toward new model iterations, intelligent driving, and physical AI-related technologies, with over 10 billion yuan (approximately $1.5 billion) in R&D funding planned for 2026. A substantial portion of the company’s first-half R&D spending went to frontier areas including its second-generation VLA intelligent driving model, Turing chips, the IRON humanoid robot, and flying cars.
Seres invested 7.007 billion yuan (approximately $1.0 billion) in R&D in the first half (including capitalized expenditures), with R&D spending exceeding 12% of revenue. The financial report shows the company achieved multiple technological breakthroughs in electronic and electrical architecture, 800V high-voltage platforms, vehicle safety redundancy, and AI engineering.
Leapmotor emphasized in its interim report that the company has consistently adhered to full-stack in-house development of core technologies, with high-value-added core components accounting for 65% of vehicle costs now developed and manufactured in-house.
NIO’s R&D decline has a different explanation. Its Q2 R&D expenses were 2.145 billion yuan (approximately $319.7 million), down 28.7% year-on-year, which the financial report attributed primarily to lower R&D personnel costs from organizational optimization, as well as reduced design and development expenses from different development stages and improved operational efficiency. NIO CFO Qu Yu said on the earnings call that quarterly non-GAAP R&D expenses will be maintained at a baseline of 2.5 billion yuan (approximately $372.6 million) throughout 2026, with dynamic fine-tuning based on project progress.
NIO, XPeng, Li Auto: three divergent paths
Judging by overall Q2 performance, NIO, XPeng, and Li Auto — once collectively known as the “big three” of China’s EV startups — have embarked on three distinctly different trajectories.
NIO delivered its strongest report card in recent years. Q2 total revenue reached 32.14 billion yuan (approximately $4.8 billion), up 69.1% year-on-year; deliveries of smart EVs hit 107,658 units, up 49.4% year-on-year; operating profit came in at 210 million yuan (approximately $31.3 million), marking a third consecutive profitable quarter. Vehicle gross margin jumped to 18.5%, up 8.2 percentage points year-on-year. Net loss narrowed from 4.995 billion yuan (approximately $744.4 million) a year ago to 528 million yuan (approximately $78.7 million).
The root of this turnaround is the “basic operating unit” reform that NIO founder William Li pushed internally. Each unit must establish clear ROI metrics and performance-based reward and penalty systems, with R&D, supply chain, and sales all brought under refined assessment. The multi-brand strategy is also beginning to generate scale effects: the NIO brand delivered 60,945 units, Onvo delivered 29,124 units, and Firefly delivered 17,589 units. The three brands share R&D, supply chain, factories, and the battery-swap system — what was once “one more brand means one more pile of money burned” is now becoming “one more brand means one more share of costs spread.”
XPeng’s financials show a “lopsided” profile. Q2 total revenue was 19.74 billion yuan (approximately $2.9 billion), up 8.0% year-on-year; consolidated gross margin was 20.7%, holding above the 20% threshold for a second consecutive quarter. But a closer look at the structure reveals Q2 deliveries of 103,295 units, up only 0.1% year-on-year; vehicle sales revenue of 17.05 billion yuan (approximately $2.5 billion), up just 1.0% year-on-year. What truly propped up gross margin was 2.7 billion yuan (approximately $402.4 million) in services and other revenue — up a staggering 93.9% year-on-year, with gross margin as high as 75.1%, the bulk of the increment coming from technology development cooperation with Volkswagen.
At the same time, XPeng’s Q2 net loss was 1.34 billion yuan (approximately $199.7 million), widening 179.9% year-on-year; cash on hand fell from 47.66 billion yuan (approximately $7.1 billion) at end-2025 to 40.48 billion yuan in the first half, burning through more than 7 billion yuan (approximately $1.0 billion) in six months. He Xiaopeng declared in his New Year letter: “In the last decade, XPeng was about smart EVs; in this decade, it is about global physical AI.” He plans to make the company the world’s first to achieve same-year mass production of Robotaxi, the IRON humanoid robot, and flying cars in 2026.
Li Auto’s situation is the most complex. Q2 deliveries were 98,330 units, down 11.5% year-on-year; revenue was 25.7 billion yuan (approximately $3.8 billion), down 15.1% year-on-year; operating loss was 2.3 billion yuan (approximately $342.8 million), with a net loss of 1.7 billion yuan (approximately $253.4 million). Vehicle gross margin was just 9.4%, nearly halved from 19.4% a year earlier, though it did recover sequentially from 6.1% in Q1.
Li Auto’s pain stems from multiple factors: China’s new energy passenger vehicle retail sales declined year-on-year in the first half, the extended-range hybrid market showed sluggish growth, and Li Auto is in the midst of a full model-cycle refresh across its L series. Costs of core components such as batteries and memory chips rose, with lithium carbonate prices up 123% year-on-year. Li Auto chose not to pass costs on to consumers but to absorb them internally, directly eroding profit margins.
Facing these headwinds, founder Li Xiang’s response strategy is to accelerate the shift to pure EVs and fully embrace AI. At an all-hands meeting in January 2026, he stated bluntly: “2026 is the last year for any company that wants to become an AI leader to get on board.” He assesses that globally, no more than three companies will simultaneously lay out foundation models, chips, operating systems, and embodied intelligence — and Li Auto intends to be one of them.
R&D efficiency is the real dividing line
The scale of R&D investment matters, but whether capital can be efficiently converted into product competitiveness depends even more on development cycle speed and platform reuse efficiency.
A white paper jointly released by the China Association of Automobile Manufacturers and automotive R&D services provider IAT — “Agent-Empowered Automotive R&D Design White Paper (2026)” — points out that traditional fuel vehicle development cycles typically run 36 to 48 months, while mainstream industry players have compressed this to 18 to 24 months, with some approaching 12 months. The dramatic shortening of development cycles means per-vehicle amortized R&D costs continue to rise, placing higher demands on automakers’ capital reserves and R&D efficiency.
Cui Dongshu, head of the China Automobile Dealers Association’s passenger vehicle committee, told China News Service’s financial news platform that using the formula “cash reserves ÷ annualized R&D,” the R&D buffer periods of different automakers vary significantly across the industry. Some leading automakers have buffer periods of around three years with positive operating cash flow, forming a virtuous cycle of investment and returns; others have buffer periods as long as eight years, indicating extremely strong investment stability; but some companies have buffer periods of only about 3.5 years with negative operating cash flow, facing notable capital-matching pressure.
Cui emphasized that the core variable truly determining R&D sustainability is free cash flow and the rate of capital consumption, not merely book cash balances. When cash is relatively abundant, automakers can lay out exploratory, long-horizon technologies; when cash tightens, R&D focus naturally shifts toward survival-oriented projects with stronger near-term monetization potential.
He believes that effective R&D that can actually translate into owner-perceived value is concentrated in five areas: intelligent driving algorithms and data, battery-motor-electronic control technology, cockpit and OTA, chassis and safety, and self-developed chips. By contrast, investment directions such as exterior design, overseas regulatory compliance, and traditional internal combustion engine iteration offer limited improvement to the user experience in China. To judge whether R&D is “effective,” he proposed three core metrics: commercialization conversion rate, R&D density, and platform reuse rate.
“There are already cases in the industry showing that some automakers achieve profitability with relatively low R&D investment, while others invest more yet remain in loss-making territory — the gap lies not in the scale of investment, but in whether R&D precisely matches user pain points and whether it forms efficient platform reuse,” Cui said.
Beyond the five companies above, data from Zeekr and Xiaomi Auto also merit attention. Zeekr’s first-half sales reached 178,000 units, up 97% year-on-year, generating revenue of approximately 55 billion yuan (approximately $8.2 billion), accounting for 31.7% of Geely Group’s total revenue. However, its financials are consolidated into Geely Automobile Group’s statements, so cash reserves and R&D expenses are not disclosed separately. Xiaomi Auto’s R&D spending is reported at the group level, with cumulative first-half R&D investment exceeding 18.2 billion yuan (approximately $2.7 billion), including 9.2 billion yuan (approximately $1.4 billion) in Q2 alone, up 18.9% year-on-year, primarily directed toward large AI models, embodied intelligence, and automotive intelligence development.
From the cross-comparison of cash reserves and R&D spending, Li Auto supports its 5.5 billion yuan (approximately $819.7 million) R&D expenditure with the largest capital pool, giving it the greatest buffer; XPeng supports the highest R&D expenses with the relatively tightest cash reserves, facing the most pronounced capital-matching pressure; NIO’s cash position has improved significantly after consecutive profitable quarters, and its R&D contraction is more a result of efficiency gains than financial constraints. The pace of new model launches and AI technology commercialization in the second half of 2026 will determine whether these investments translate into substantive changes in market positioning.
