“`html
“It’s noticeably tamer this year. Last year, there was a real sense of imaginative exploration. This year, everyone’s playing it safe, focusing on what’s perceived as stable ground.” That’s the blunt assessment from a key figure at a leading tech firm, speaking candidly from the floor of the World Artificial Intelligence Conference (WAIC).
By noon on July 26th, two hours before the doors officially opened, the South Gate of the Shanghai World Expo Exhibition & Convention Center was already teeming. A steady drizzle couldn’t dampen the enthusiasm, as crowds, umbrellas aloft and luggage in tow, converged from all directions, including numerous international guests who had flown in from afar.
Their mission? To participate in what organizers touted as the “largest WAIC ever.”
Mingling with the raindrops were fervent discussions about the hottest topics in AI today: large language models (LLMs) and embodied intelligence.
However, unlike last year, when LLMs reigned supreme, embodied intelligence—specifically, humanoid robots—stole the show, becoming the undisputed attraction for attendees.
The clearest indicator of this surge in interest? The inflated ticket prices, soaring to over ￥1,000 on the secondary market.
Sources indicate that WAIC tickets sold out well in advance. Two days before the event, resale prices on platforms like Xianyu had reached ￥3,000, a stark contrast to the official price of ￥200 just a month earlier. The frantic demand underscores the high expectations surrounding the conference.
This “largest WAIC ever” lived up to the hype. Official figures reveal that the exhibition area exceeded 70,000 square meters for the first time, featuring over 800 exhibitors, with more than half hailing from outside Shanghai and internationally. Over 3,000 exhibits were on display.
The growth in embodied intelligence was particularly striking: up from a mere 18 companies crammed onto one floor last year, this year saw 80 related companies “occupying” the entire second floor of the Shanghai World Expo Exhibition & Convention Center. The B1 level was also completely filled with exhibitors of varying disciplines.
The second floor was essentially transformed into a “robotics stage.” Exhibitors took turns showcasing their creations, with humanoid robots walking, grasping, and executing tasks with a level of fluidity that, given the industry’s nascent stage, could be described as “almost perfect.”
But this curated “perfection” also created a telling contrast: while robots in the real world are still far from widespread practical use, and glitches are commonplace, WAIC’s exhibition floor was remarkably free of any robotic “mishaps.”
This deliberate “perfection” clashes subtly with the undercurrent of “playing it safe” and the perceived “lack of imagination” voiced by industry insiders.
“Who do you think has the best shot at becoming the Jensen Huang of robotics?” The question, tinged with industry anxiety and anticipation, echoed from the crowds gathered outside the exhibition hall, hoping to get a glimpse inside.
01 The “Safe Bet” Robots
The epicenter of the conference was undoubtedly the second floor, and at its heart stood Unitree Robotics.
Unitree’s booth resembled a grand “stage show,” surrounded by layers of onlookers. Despite multiple attempts, this reporter couldn’t penetrate the crowd to witness the robot “battles” firsthand, instead observing a sea of raised phones – everyone eager to capture a moment of this robotic extravaganza.
Not far from Unitree, the Zhiyuan Robotics booth, occupying the largest space in the embodied intelligence area, also drew massive crowds.
Every half hour, their Expedition A2 robot took center stage, performing drumming or dance routines that consistently elicited gasps of amazement.
Unitree and Zhiyuan stood out as the two most eye-catching highlights of WAIC.
Beyond being crowd-pleasers, they are darlings of the capital markets: one valued at over ￥10 billion, the other at ￥15 billion, fueling persistent IPO rumors.
Adding to the intrigue is their proclaimed speed of commercialization. Just a month before WAIC, Unitree founder Wang Xingxing revealed at the Davos Forum that the company’s annual revenue had reached “over ￥1 billion.”
Meanwhile, Zhiyuan partner Yao Maoqing told local media, “Zhiyuan can also achieve ￥1 billion in revenue this year. In fact, we have made significant progress in commercialization in the first half of the year.”
A revenue target of ￥1 billion would be astronomical for a nascent humanoid robotics sector.
While it remains to be seen how these two robotics companies will achieve this ￥1 billion target, it’s clear that they have established a considerable lead in terms of visibility and resources over their competitors.
Looking around the expansive second-floor hall, only a handful of robots could confidently demonstrate dynamic capabilities like Unitree and Zhiyuan.
Most exhibitors remained confined to traditional scenarios such as supermarkets and logistics, showcasing basic grabbing and moving actions – maneuvers less prone to “failure”.
Some even had robots lying passively on sofas, not even plugged in, simply to join the hype and show they weren’t absent.
And for the vast majority of these robot companies, commercial progress remains difficult.
Sources indicate that some manufacturers have shipped only a handful of units, with individual prices even below ￥100,000.
This contrasts sharply with their eight-figure funding rounds. Perhaps choosing the most conservative demonstrations is a way to avoid “embarrassing” themselves in front of a global audience and potential investors.
It’s also been revealed that, aside from scheduled performances, many robots will only put on a show when potential clients visit the booth.
The primary objective is to display their prowess and technical capabilities to investors and customers, further emphasizing their stability and sophistication.
This year, many investors focused on embodied intelligence emphasized, “One of the most crucial metrics for the embodied intelligence track is fundraising ability.”
And so, we’ve witnessed a surge in funding rounds.
According to market data, the number of funding events in the embodied intelligence sector in the first five months of this year exceeded the total for the entire year of 2024, with funding amounts surpassing ￥23 billion, exceeding the total for 2024 as well. Many sizable multi-million dollar funding rounds have been observed.
However, this has sparked some highly critical commentary.
Prior to the WAIC conference, one industry insider stated, “The reason embodied intelligence companies are constantly raising funds this year is because if they don’t seize the opportunity now, they’ll soon enter a cooling-off period in the coming years, making it difficult to secure any more money.”
This perfectly illustrates the dramatic scene unfolding at the WAIC.
In this reporter’s view, these “playing-it-safe” embodied intelligence companies are genuinely afraid that if the bubble bursts, their narratives will fall apart, making it nearly impossible to secure further funding.
02 The No-Frills Large Language Models
In stark contrast to the displays and performances on the second floor, the focus on the first floor was noticeably more practical, centered on Large Language Models.
However, it must be noted that embodied intellgence has eclipsed the LLM’s buzz and enthusiasm from the previous year.
The goal of “¥1 billion” in revenue seems to be trending this year. Not only within the domain of embodied intelligence, but within the realm of large language models, StepX’s founder Jiang Daxin echoed the very same sentiment as Wang Xingxing.
One day prior to the World AI Conference, close the World Expo Center at the West Bund MGM hotel, StepX held a major press day. Not only did they officially announce the opening of their multimodal Step3 for open-source, but declared their ambitions to reach “￥1 billion in revenue this year.”
Along with this statement came another key announcement that a Shanghai state-owned capital investment firm would be investing in the project.
With all the buzz and activity, StepX was poised to generate much interest during the following day. Within the first-floor exhibit hall, StepX successfully attracted a large audience and maintained a down-to-earth attitude.
And they didn’t hesitate to bring a automobile exhibit and a singer to attract attention.
StepX collaborated with Canxing Culture to create the “China AI Good Voice” interactive zone, featuring the famous red instructor seats and attracting plenty of social media influencers.
Another key focus of Shanghai is on MiniMax. This may be one of the few LLM exhibits that boasts a prominent display space as a “Big Six LLM” company.
In contrast to StepX’s end user strategy, MiniMax is focusing purely on selling API’s and investing heavily in consumer applications. Just weeks before the World AI Conference, rumors were circulating of a planned IPO in Hong Kong.
Moreover, this year’s World AI Conference’s keynote speaker has been replaced from a major brand to a leader in AI large language models. After deep learning pioneer and Turing Award winner Geoferry Hinton, MiniMax’s founder Yan Junjie gave a speech on the future of China AI.
Other major AI large language firms appeared more subdued and even subdued in comparison.
ZhiPu AI and Kimi (Moonshot AI) were present, however, it was apparent the booth spaces have been decreased substantially in respect to the previous year, implying a attitude of “participation-focused”.
Baichuan AI has vanished this year. With obvious indications, the competitive landscape and consolidation of the LLM sector are shown by the floor area of the exhibitor spaces.
As attendees strolled by the stands of former and current well-known firms, an obvious trend became apparent. LLM producers this year have all but ditched competing on parameters and are now focusing on “telling tales of success”. They are now prioritazing on staying alive under strain and mainting relevance.
Several notable events were revealed amongst the Big Six LLM in the first portion of the year:
1) Until the beginning of the World AI Conference, all of the businesses engaged in training had released their individual inference models.
2) The “survivors” are now well-liked and embraced. Examples include Shanghai state-owned groups investing in StepX and MiniMax.
3) Beyond trying to improve or create unique LLM technologies, LLMs need to show how they can be implemented and successfully achieve scale and adoption.
Evidently, AI LLM businesses must prioritize existing and also be able to “build a narrative” and successful usecase that appeals to both customers and investors.
03 What Remains Following Enthusiasm?
Despite the lack of DeepSeek in this year’s World AI Conference, they have surely sparked enthusiasm in China as well as for AI practitioners across the globe.
To a degree, it is a primary factor for that reason that the WAIC of 2025 garnered so much global notice and captured so many overseas fans.
But, as the exhibition center was packed with people and cell phones were taking photos of every seemingly flawless dance performed by robots, a chilling question suddenly surfaced: How much bubble is behind this massive amount of interest?
Even though the embodied intelligence display on the second floor attracted attention and amazement, it was precisely those beautifully orchestrated and almost flawless shows that soured the event from becoming just another trade show.
As though time is repeating itself.
The LLM industry swiftly and dramatically decreased from a tumultuous state and is now only a few unicorns in the last 2–3 years.
Furthermore, considering that, how many may be left from the 80 players vying for the spot in the embodied intelligence sector in terms of attendees?
The prospects are huge, as in stars in the night sky, in the field of embodied intelligence, that are forward-thinking and technologically and futuristically advanced as those sectors must dominate the scene immediately. No one is sure of what can or will happen.
Instead, at the World AI Conference of 2025, it was a different scene. The heads of businesses made lofty statements about reaching “¥1 billion in revenue”, while the others in attendance carefully displaying traditional use cases that were less likely to fail.
The rush to attain guarantees, and promote profits rapidly, is stifling the audacious mindset needed to promote technological breakthroughs.
¥1 billion is good goals as an aim, but there may be a bubble looming in the future.
“`
Original article, Author: Tobias. If you wish to reprint this article, please indicate the source:https://aicnbc.com/5793.html
