AI Explosion! From “Screen” to “Physical World” – How Can Enterprises Secure Their “Ticket” to the Intelligent Era?

One‑sentence overview: In the past week, the global AI industry reached a historic turning point – AI is moving boldly from the virtual “screen world” into the “physical world”. OpenAI announced the formation of a robotics team, NVIDIA partnered with Yushu Technology to build the next generation of humanoid robots, and two giants – Apple and WeChat – made their moves almost simultaneously. The real explosion of the AI internet may have just begun. Behind this wave, computing power – especially GPU leasing and computing power leasing – is becoming the critical infrastructure that will decide the winner.

01 Weekly AI Highlights: The Densest “Commitment to Reality” in History

In the past week (June 7 – June 13), the global AI industry delivered an impressive string of announcements. From Silicon Valley giants to Chinese tech companies, from pure‑software large language models to tangible robots – the curtain has risen on AI moving from the virtual to the physical.

🔥 Event 1: OpenAI Enters the Robotics Arena

On June 1 (actually it started to gain momentum a week earlier, with signals continuing throughout the week of June 7), Sam Altman, CEO of OpenAI, officially announced on social media the establishment of the “OpenAI Robotics” team. The team is recruiting full‑stack hardware, systems, and machine learning engineers worldwide, with the goal of “programming and building robots that are truly useful to society.” This marks the official and comprehensive entry of the AI giant – known globally for its large language models – into the world of physical AI.

It is worth remembering that OpenAI had already worked on a robotic hand project in 2018, but later disbanded the team in 2021 due to data bottlenecks. The reason for re‑entering this field now is twofold: on the one hand, the partnership with Figure AI ended because of technical disagreements, pushing OpenAI toward in‑house, integrated hardware‑software development; on the other hand, as OpenAI approaches its IPO, robotics – as a “second growth curve” – can better tell a new growth story to the capital markets.

Industry impact: OpenAI’s move directly ignited the global “embodied AI” track. According to不完全 statistics, in just three days from June 8 to 10, at least seven robot‑related financing rounds were completed in global primary markets, totalling over US$1.5 billion. It is foreseeable that within the next three years, the computing cost required for training large robot models will explode – and this is exactly the core opportunity for computing power leasing service providers.

🔥 Event 2: NVIDIA Joins Forces with Yushu Technology to Create the “Strongest Chinese Robot”

Almost on the same day, NVIDIA announced a partnership with China’s Yushu Technology to launch a new humanoid robot reference design, the “H2+”, powered by NVIDIA’s Isaac GR00T system. In this model, Yushu Technology provides the robot’s “body” – about 1.8 meters tall – while NVIDIA provides the computing platform and the “brain” (foundation model). This division of labour – “Chinese body, overseas soul” – not only reflects the current state of the global industrial chain, but also signals that AI is no longer confined to the screen; it is beginning to acquire “hands‑on ability”.

Notable detail: The robot has 42 degrees of freedom, and a single training session in a simulated environment requires more than 2000 TFLOPS of computing power. This means that any enterprise hoping to develop customised scenarios based on the “H2+” must have a stable, scalable computing support solution. For the vast majority of small and medium‑sized robotics companies, purchasing tens of millions of yuan worth of GPU servers directly is unrealistic – GPU leasing is becoming the most rational choice.

🔥 Event 3: Apple and WeChat Make Their “Twin Moves”

In the early morning of June 9 Beijing time, Apple made its new AI announcement at the 2026 Worldwide Developers Conference: Siri is receiving the largest overhaul since its launch in 2011, and the new Apple Intelligence is officially landing. This marks Apple’s official entry into the era of large‑scale on‑device AI deployment.

At almost the same time, WeChat announced that it is officially opening its AI ecosystem to developers. Meituan, Didi, JD.com, Ctrip and other companies quickly declared partnerships with WeChat AI. These two iconic “hunters” of the mobile internet era – both questioned in the past few years for “missing AI” – have now chosen to enter the AI jungle together. This clearly shows that AI is by no means a short‑lived wave, but an inevitable paradigm shift.

Computing perspective: On the very first day that WeChat’s AI ecosystem opened, developer calls exceeded 5 billion tokens. Behind every intelligent reply in a chat, behind every voice assistant understanding, lies real‑time inference computing consumption. It is foreseeable that with more third‑party services connecting, WeChat AI’s daily token consumption will surpass 200 billion within this year. This level of inference demand will directly fuel the explosion of the edge computing power leasing market.

🔥 Event 4: 200+ AI Experts Gather in Beijing to Discuss World Models

On June 12‑13, the 2026 BAAI Conference (Beijing Academy of Artificial Intelligence) was held at the Zhongguancun International Innovation Center in Beijing. More than 200 leading scholars from home and abroad came together to discuss cutting‑edge topics such as large models, AI agents, world models, and embodied AI. For the first time, the BAAI showcased the world foundation model “Wu Jie · Physis‑v0.1” under training, marking AI’s evolution from “generating content” to “understanding the physical world”.

The President of BAAI revealed at the conference that the total computing power required to train “Wu Jie · Physis‑v0.1” has reached 2.3×10²⁵ FLOPS – more than four times the training compute of GPT‑4. And this is only the v0.1 version; the full version may require ten times that amount. Such a scale of computing power is already beyond the capacity of most institutions to build in‑house – large‑scale computing power leasing cluster scheduling is the only feasible path.

🔥 Event 5: The Biggest ChatGPT Redesign Ever – “The Era of AI‑Only Chat Is Over”

On June 7 local time in the UK, according to the Financial Times, OpenAI is preparing the largest redesign of ChatGPT since its launch in 2022 – the core of which is to fully integrate the Codex programming tool into ChatGPT, while also connecting third‑party partners such as Canva and Booking.com. This will reshape the chatbot, which has nearly a billion users, into a “super app” capable of writing code, managing schedules, and controlling software. A senior OpenAI insider summed up the core concept of this transformation in one sentence: “The era of using AI only for chat is over.”

It is foreseeable that once ChatGPT becomes a true “AI operating system”, daily calls will explode from the current billions to hundreds of billions or even trillions. Each call means the consumption of GPU resources – shifting from short‑term, bursty training demand to sustained, steady inference demand. This change will reshape the entire supply‑demand structure of the GPU leasing market.

02 The Computing Tsunami: The Underlying Logic Behind the AI Explosion

Behind all these exciting phenomena, there is one thread that runs throughout: whether training world models, running robot algorithms, or supporting AI agent applications with hundreds of millions of users – none of it can be done without computing power.

At the GTC Taipei 2026 conference, Jensen Huang pointed out that tokens are becoming a new commodity, and AI companies are already able to generate revenue from token generation – and this is the main reason for the continued explosion in global computing demand. He also announced that the Vera Rubin super‑AI chip has officially entered full mass production, with OpenAI, Anthropic, and SpaceX among the first customers.

A wealth of data confirms this trend:

• Exponential growth in token calls: Liu Liehong, Director of the National Data Administration, revealed that at the beginning of 2024, China’s average daily token calls were only 100 billion; by the end of 2025, that number had jumped to 100 trillion; and by March 2026, it had exceeded 140 trillion – an increase of more than 1,000 times in two years. Behind every conversation, every inference, computing power is quietly burning real money.

• Explosive growth of the computing power leasing market: According to the China Academy of Information and Communications Technology (CAICT), in the first quarter of 2026, the domestic computing power leasing market reached 68 billion yuan, a year‑on‑year increase of 62%, and is expected to exceed 260 billion yuan for the whole year. IDC predicts that the global computing power leasing market will exceed US$80 billion in 2026.

• Soaring GPU rental prices, severe supply‑demand mismatch: The one‑year GPU leasing contract price for an H100 has skyrocketed from a low of US$1.70 per GPU per hour in October 2025 to US$2.35 per GPU per hour in March 2026 – an increase of nearly 40%. At the same time, all on‑demand rental capacity for all GPU types is sold out, and the utilisation rate for high‑end GPU leasing exceeds 90%.

• Accelerated growth of the enterprise AI agent market: China’s enterprise AI agent market reached 21.2 billion yuan in 2025, is expected to grow to 44.9 billion yuan in 2026, and is projected to exceed 332 billion yuan by 2029 – a compound annual growth rate of 107%. Every newly launched AI agent corresponds to a continuous stream of computing consumption.

• Frequent billion‑yuan computing power deals in A‑shares: In the second quarter of 2026, the A‑share market saw frequent deals worth tens of billions or even hundreds of billions of yuan in the computing power sector. Dongyang Sunshine signed a computing power service framework contract worth 16‑19 billion yuan, and Huace Film & Television plans to purchase 37,000 Pflops of computing power, with contract terms generally locked in for five years.

An even more striking set of data (from on‑site reports at the BAAI conference):

• The total computing power consumed by the world’s top 10 foundation models grew from approximately 8×10²⁴ FLOPS in 2023 to more than 1.2×10²⁶ FLOPS in the first half of 2026 – a 15‑fold increase in three years.

• China’s total AI computing power (in FP16 equivalent) grew from 280 EFLOPS at the end of 2024 to 730 EFLOPS in May 2026, but the estimated effective demand over the same period reached 1100 EFLOPS – a gap of more than 370 EFLOPS. This is the fundamental reason for the supply shortage in the computing power leasing industry.

• For every 1 yuan of computing cost invested, an AI startup can generate approximately 8‑12 yuan of model service revenue – the “leverage effect” of computing power is becoming increasingly evident.

The computing power leasing industry has formally moved from its early exploratory stage of “temporary allocation, fragmented orders” into a high‑growth stage characterised by “long‑term contracts, large‑scale deployment”. In the words of industry insiders: the computing power leasing market has completely shifted from a buyer’s market to a seller’s market; those who control computing resources are seizing pricing power.

03 The Computing Power Leasing Era: The “Pick‑and‑Shovel” Play in the AI Gold Rush

If we compare the current AI boom to an unprecedented gold rush, we can easily identify a pattern: very few strike gold, but every person selling picks and shovels to the prospectors makes a cash‑based business.

This is exactly where the computing power leasing industry stands today. Lenovo Group’s latest financial report shows that its AI‑related business revenue grew by 105% year‑on‑year. On the A‑share market, the performance of companies in the computing power leasing sector is even more stunning – Xiechuang Data’s net profit attributable to shareholders in the first quarter of 2026 increased by 343.45% year‑on‑year, while Hongjing Technology’s net profit attributable to shareholders grew by 40%. Every AI company, every newly launched large model, and every enterprise connecting to an AI agent requires a continuous consumption of computing resources.

For AI startups, small and medium‑sized enterprises, and traditional enterprises that are now embracing AI, buying and operating tens of millions of yuan worth of high‑performance GPU servers is not the most rational choice:

✅ Light‑asset start‑up: No need for a one‑time investment of tens of millions in equipment. Through computing power leasing, you can flexibly obtain the computing power you need on a monthly, quarterly, or yearly basis.

✅ Flexible scaling: AI projects are highly uncertain; the demand for computing power for model training fluctuates significantly. The leasing model allows you to adjust in real time according to actual needs, avoiding idle hardware.

✅ Always access the latest GPUs: GPU upgrade cycles are accelerating (NVIDIA has committed to a two‑year architectural upgrade). Self‑purchased equipment can quickly depreciate, whereas the leasing model allows you to always access the latest high‑performance computing power.

✅ No operation and maintenance headaches: Professional O&M teams, liquid cooling, network cluster scheduling – let professional computing service providers handle these complex aspects.

✅ Pay‑as‑you‑go, reduce trial‑and‑error risk: Many AI projects are unsure of their direction at an early stage; a huge one‑time investment in hardware is extremely risky. GPU leasing allows companies to pay by the hour or by the day. If a project fails, the sunk cost is minimal.

Real‑life case: An autonomous driving startup, during its algorithm validation phase, used our computing power leasing service. With a monthly cost of 80,000 yuan, they obtained the equivalent computing power of 2 million yuan worth of self‑purchased hardware. Within six months, they completed three rounds of algorithm iteration, successfully secured Series B financing, and only then gradually transitioned to a hybrid model (partly self‑owned, partly leased). This “lease first, build later” path has become standard practice for a large number of AI companies.

According to the CCID Research Institute, the current utilisation rate for high‑end GPUs (A/H series) in China exceeds 90%, and the market is in short supply. In an era where computing power is a basic necessity, securing a computing power leasing channel in advance is like getting a valuable “ticket” for your company’s future development.

04 Future Outlook: When Computing Power Becomes as “Plug‑and‑Play” as Electricity

If we look further ahead, we can see that the ultimate form of the computing power leasing industry will likely be like electricity or tap water today – an infrastructure‑level public service.

• 2026‑2027: The market will be dominated by “long‑term leasing + customisation”. Leading computing power service providers will sign framework agreements of 3‑5 years with large AI companies, while small and medium‑sized customers purchase on demand through platforms.

• 2028‑2030: As computing power networks and scheduling technologies mature, a cross‑region, cross‑cloud, cross‑vendor computing power leasing market will take shape. Companies will be able to obtain resources from the cheapest computing nodes in real time, just as they buy electricity.

• Beyond 2030: Inference demand will far exceed training demand. Every mobile phone, every car, every home robot may be connected to a cloud‑based computing pool. By then, GPU leasing will be metered down to “per token”, and computing power will truly become a utility‑like service, just like electricity.

Of course, this journey faces challenges: power consumption, chip supply, cross‑region scheduling latency, and so on. But for today’s decision‑makers, the most important thing is to start including computing power acquisition in your strategic planning right now – build, lease, or a hybrid model? The answer to this question will largely determine your competitive position in the AI era.

05 Conclusion: Physical‑World AI Needs Physical‑World Infrastructure

Over the past two or three years, the industry has been filled with anxiety about large model “hallucinations”, computing costs, and the difficulty of commercialisation. But the simultaneous moves by Apple and WeChat send a strong and clear signal – the real explosion of AI internet applications may have only just begun.

As AI moves from “understanding the world” and “thinking about the world” to “physically changing the world”, computing power is no longer a “nice‑to‑have” but a “must‑have” for survival. Whether you are an AI startup, a traditional enterprise undergoing transformation, or a developer who needs to deploy AI agent applications – computing power leasing and GPU leasing will be the most worthwhile “entry ticket” to the intelligent era.

📌 About Us

We specialise in providing high‑performance computing power leasing and GPU leasing services. With a professional operations and maintenance team, we support elastic scaling from a single GPU card to a thousand‑card cluster. Whether you are training a large model with tens of billions of parameters, deploying enterprise‑level AI agents, or performing simulations for embodied AI algorithms, we can provide stable, reliable, and flexible computing support on demand.

Contact us to obtain a tailor‑made computing power solution – and in the AI wave, become a “pick‑and‑shovel seller” who controls the infrastructure, together with us.