Xpeng's "physical AI" gamble: can it earn a ticket onboard?
```
Recently, XPeng Group’s financial data for the first quarter of 2026 has been officially released. In the first three months of 2026, XPeng Group recorded revenue of 13.03 billion yuan, representing a year-on-year decline of 17.6%. Although the overall gross margin increased by 5.0 percentage points to 20.6% compared to the same period, the net loss still reached 1.78 billion yuan, significantly expanding from a loss of 660 million yuan in the same period of 2025.
The direct reasons for the expanded loss include not only a decrease in the number of deliveries compared to last year’s same period, but also a significant rise in R&D expenses—XPeng Group invested as much as 2.91 billion yuan in R&D during the quarter, with both year-on-year and quarter-on-quarter increases.
During XPeng’s Q1 financial conference call and the podcast released on May 28, XPeng Group chairman He Xiaopeng gave a directional explanation for the increase in R&D investment during the first three months of 2026: the company is fully shifting from a previous “Frankenstein” architecture, which stitched together multiple AI models, to a new “physical AI” system driven by a general large model.
Around this transition, he made a series of commitments with timelines, such as a sharp quarter-on-quarter rise in delivery guidance for Q2; Robotaxi having already rolled off the production line, with pilot operations in Guangzhou starting in the second half of the year; humanoid robot IRON entering mass production in the fourth quarter this year, joining XPeng stores as guides in Q1 2027, and being exported overseas in Q2.
Along with upcoming new models such as L03 and the G9 facelift, XPeng is using a short-term pressured financial report in exchange for access to the commercialization cycle of physical AI.
Whether this is strategic foresight or excessive, premature investment depends on the fulfillment of these commitments over the next four quarters.
01 The Story Behind the Expanded Loss
From the financial report, the core contradiction in XPeng’s Q1 results lies in why the overall loss expanded despite improved per-car gross margin. The answer lies in the expense structure.
From the revenue structure, Q1 total revenue was 13.03 billion yuan, of which automobile sales accounted for about 11 billion yuan, a 23.5% decline from Q1 2025. This drop is mainly due to decreased deliveries in the quarter. Specifically, XPeng delivered 62,682 vehicles in Q1 2026, a 33.3% decrease from 94,008 units in Q1 2025.
However, overall gross margin was maintained at 20.6%, up 5 percentage points from the same period in 2025.
Despite the year-on-year drop in deliveries, the automotive business gross margin rose from 10.5% last year to 12.1%.
This shows that XPeng’s sales mix has improved, with a higher proportion of high-priced models. In terms of profit per vehicle, profitability has also improved. Cost reduction through technology and platform sharing are working, with more profit generated per car sold than a year ago.
Looking at expenses, sales and administrative expenses reached 1.88 billion yuan, down by 3.2% year-on-year and 32.5% quarter-on-quarter. Despite a sharp drop in deliveries and the need for marketing for new models, this expense is still shrinking, indicating management’s cost control.
Another profit-eating factor is the rise in R&D investment. Q1 R&D expenses were 2.91 billion yuan, up over 46.8% from 1.98 billion yuan in the same period last year, and also higher quarter-on-quarter compared to Q4 2025.
The incremental funding mainly went toward smart driving R&D, development of humanoid robot IRON, and software-hardware adaptation for the Robotaxi fleet. He Xiaopeng also stated in the May 28 interview that the company is in a critical investment phase of migrating from “AI Frankenstein” to a unified physical AI architecture. Such structural costs are unavoidable.
So, where could a financial turnaround in the short term come from? It will depend on the cadence of forthcoming new models.
XPeng’s financial call guidance for Q2 points to a sharp quarter-on-quarter rise in deliveries—about 100,000 to 106,000 units.
This expectation is supported by the intensive launch of multiple new cars. The GX model was released in May, and according to the Zhongtai Auto team’s research, its orders exceeded expectations, with nearly 25,000 units ordered within 12 hours, and monthly sales are expected to stabilize around 7,000 units, far above the market’s prior expectation of only 2,000 to 3,000 units.
The more crucial volume model L03, positioned as a pure electric SUV under 150,000 yuan, will be released in Q3.
This model shares much of its underlying architecture with the M03, which has long topped the 100,000–150,000 yuan pure electric sedan sales chart. Market expects monthly sales to break through 15,000 units.
Additionally, the G9 facelift and L05 models are scheduled for sequential launch and delivery in the second half, with seven “dual-energy” models planned for the year.
If sales expectations for L03 and GX are realized as expected, XPeng will enter a steep ramp-up period for delivery volumes in Q3.
The scale effect will dilute manufacturing costs, and together with positive gross margin trends for existing models, automotive business gross margin may continue upward within the year.
Moreover, services and other income—mainly including technology licensing and smart driving subscriptions—reached 2.03 billion yuan in Q1. With the VLA distilled version gradually being deployed in cars this year and technical solutions supplied to more automakers, this high-margin income item will become a key buffer for overall profitability.
The new car cycle starting in Q2 will be the first window testing whether XPeng’s investments can be converted into scale and profit.
02 The “Physical AI” Trilogy
Beyond the financials, He Xiaopeng is betting XPeng’s future on three things: exporting smart driving technology, operating Robotaxi, and mass-producing humanoid robots.
In the May 28 financial call, he clarified three core business lines: advanced intelligent driving, Robotaxi, and humanoid robot IRON. All point to a qualitative leap from "digital AI" to "physical AI" capable of interacting with the real world.
Intelligent driving remains the foundation.
He Xiaopeng believes the VLA 2.0 capability has reached the absolute top domestic level, and this year will begin to deploy distilled model versions in vehicles, with plans for global application of VLA 2.0 by 2027.
However, Huawei’s Qiankun ADS is also spreading rapidly across AITO and Luxeed vehicles, while Tesla’s FSD is accelerating local testing in China. Whether XPeng can maintain its smart driving advantage depends on the speed of technical iteration and cost.
With a unified large model to understand, plan, and control vehicles, XPeng is bidding farewell to the rule-driven “Frankenstein” approach. Prior to this, XPeng reported that VLA2.0’s first month saw over 50% of driving mileage utilize its smart driving system, which is key proof for the correctness of the technical route.
This shows user dependence and trust in the new smart driving system is crossing a chasm, which also lays the commercial foundation for XPeng's technology licensing to other carmakers and expansion of service income.
The “mass production countdown” for Robotaxi and humanoid robot IRON pushes XPeng’s valuation cap from automaker toward a broader physical AI service provider.
For Robotaxi, leveraging its closed-loop smart driving tech, vehicles have already rolled off the production line, with pilot operations set to start in Guangzhou in the second half. He Xiaopeng further revealed that future Robotaxi will introduce lower-priced models and head overseas.
The advanced humanoid robot IRON is a bolder step.
Its commercialization path is unusually clear and pragmatic: mass production in Q4 this year, entering the company’s stores as sales guides in Q1 2027.
This “use-first” strategy can accumulate data and iterate technology in real-life scenarios, while exploring business value at relatively controlled costs.
Meanwhile, humanoid robot IRON is also planned for overseas export, expected in Q2 2027.
XPeng is telling an “one brain drives many bodies” story, with its core AI ability as the brain simultaneously powering smart cars, Robotaxi fleets, and humanoid robots. If the market accepts this narrative, XPeng’s valuation anchor could shift from the carmaker’s price-to-sales ratio to a tech company logic. But at present, income from smart driving tech licensing is volatile, Robotaxi and robots are still in the investment stage, and no single line can independently contribute profit.
What the market needs to see is tangible scale deployment and fulfillment in financial data, not just a perfect technical architecture or business plan.
Can IRON truly replace human staff in stores? What will be the unit economics of Robotaxi when it scales up? Until these questions are answered, “physical AI” remains just a narrative.
For now, XPeng has already secured the ticket to the next track. The race has only just begun.
Risk Warning and DisclaimerThe market involves risk; investment requires caution. This article does not constitute personal investment advice, nor does it take into account the unique investment objectives, financial situation, or needs of any individual user. Users should consider whether any opinions, views, or conclusions herein fit their specific circumstances. Investments made based on this content are at your own risk. ```