Breaking news

AWS Launches Trainium3 as It Expands AI Compute Capacity

Redefining AI With Custom Chips

At the recent AWS Re:Invent 2025 conference in Las Vegas, Amazon Web Services (AWS) CEO Matt Garman took center stage to introduce Trainium3, the latest iteration of AWS’ in-house custom chip. Promising a fourfold improvement in compute performance, energy efficiency, and memory bandwidth over previous generations, Trainium3 has already begun to deliver significant results—cutting AI training and inference costs by up to 50% in early tests.

Strategic Diversification In The AI Chip Arena

AWS’ approach is reflective of a broader industry trend where tech giants invest in proprietary chip technology. Just as Google has attracted attention with its tensor processing units (TPUs) co-designed with Broadcom and used to power Gemini 3, AWS is solidifying its position by offering enhanced compute solutions. Meanwhile, Meta Platforms has reportedly explored the benefits of TPUs in addition to Nvidia’s dominant graphics processing units (GPUs), which continue to serve as the industry standard for versatile AI workloads.

Expanding Cloud Capacity Through Hybrid Solutions

In a strategic move to address intensifying competition from Microsoft’s Azure and Alphabet’s Google Cloud, AWS also announced the launch of AWS Factories. This on-premise AI platform leverages both Trainium accelerators and Nvidia GPUs—providing customers with access to Nvidia’s accelerated computing platform and full-stack AI software. The dual approach emphasizes scalability and operational flexibility, key factors as the market adapts to surging AI demand.

Capacity Building to Meet Escalating Demand

Beyond technological breakthroughs, AWS is aggressively ramping up its computing capacity. After addressing a period of supply constraints, AWS is now scaling its infrastructure to meet ever-growing demand. According to recent Q3 disclosures and reaffirmed by Amazon CEO Andy Jassy, the company is on track to add more than 12 gigawatts of compute by year-end 2027. Analysts have estimated that each additional gigawatt could translate to roughly $3 billion in annual cloud revenue, potentially boosting AWS revenue by 14% in 2026 and 22% in 2027.

Investors Focused on the Future

Although AWS’ fresh chip developments have garnered attention, the pivotal focus for investors remains on capacity expansion. Wall Street is betting that overcoming current supply limitations will transform a capacity headwind into a substantial revenue tailwind. With unparalleled logistical capabilities, AWS is uniquely positioned to navigate this complex expansion, as evidenced by its recent performance milestones and strategic investments.

The rollout of Trainium3 and the evolution of AWS’ hybrid AI infrastructure signal a broader commitment to staying at the forefront of cloud innovation. These developments not only reinforce AWS’ leadership in the cloud market but also underscore its critical role in powering the next generation of AI-driven enterprises.

China’s Humanoid Robot Boom Faces A Bigger Question: Can These Machines Make Money?

Unitree’s $9 Billion Bet On The Future Of Robotics

China’s humanoid robotics industry is attracting huge investor interest, but as Unitree Robotics prepares for its public debut, questions are growing over whether its robots can move beyond impressive acrobatics and become commercially viable tools.

The Hangzhou-based startup priced its IPO at 150.8 yuan ($22.4) per share, raising $900 million and valuing the company at 61 billion yuan, or about $9 billion. The offering attracted record retail demand on Shanghai’s STAR Market, with the online tranche oversubscribed more than 5,000 times and a winning rate of just 0.018%. Strategic investors included AI startup DeepSeek.

A Unitree-linked pre-IPO perpetual contract was trading at roughly four times the IPO price on Friday, highlighting the speculative interest surrounding the company.

Unitree is known for robots capable of kung fu kicks, backflips and recovering from falls. Yet analysts question whether the technology is ready for large-scale commercial use. “For these humanoid robots, to be honest, they’re fascinating. They can dance and all that, but I’ve never seen them doing any real housework,” said Hao Hong, managing partner of Lotus Asset Management.

In its prospectus, Unitree warned that mass adoption could take longer than expected because robotic hands are still not precise or durable enough for sustained use.

From Acrobatic Robots To Commercial Machines

Even advanced humanoid robots can currently perform only a limited number of tasks and typically operate for a few hours before recharging, according to Dominik Pross, an equity analyst at VP Bank. Most models run for up to four hours, while robots also need to be trained for individual tasks.

“Robots have to be specifically trained for each and every task entrusted to them, even the simplest,” Pross said.

More robotics listings are expected, with Unitree rivals AgiBot and Leju Robotics seeking listings in Hong Kong and Shenzhen. LimX Dynamics founder Will Zhang said last month that “listing is a must.”

China’s Cost Advantage

China’s manufacturing scale has helped it establish a leading position in robotics. Wood Mackenzie expects the global humanoid robot fleet to surpass 10 million units by 2035, while China already accounts for more than 70% of global industrial robot installations and nearly 90% of humanoids deployed last year.

Average humanoid robot prices fell 93% between 2020 and 2025 to $58,000. Unitree’s flagship G1 costs $16,000, while SemiAnalysis estimates that the company has cut the price of its G1 EDU model by more than 45% to $27,300, while maintaining a 67% gross margin.

Falling prices and government support are attracting investment, but analysts say it will take time to prove that humanoid robots can generate strong returns. Unitree’s revenue more than quadrupled last year, although adjusted first-quarter profit fell more than 52% as research and development and marketing spending increased. Nearly three-quarters of its humanoid revenue in the first nine months of 2025 came from research and education, highlighting the gap between demonstrations and widespread commercial use.

“Unlike many early-stage robotics companies, the Unitree story is backed by real revenue growth,” said Jeff Ko, chief analyst at CoinEx. Still, he noted that its $9 billion valuation, at more than 200 times last year’s earnings, reflects significant speculative interest.

Geopolitical Risks

Unitree’s IPO momentum has continued despite growing pressure on Chinese robotics companies. The U.S. moved last month to ban imports of foreign-made humanoid and four-legged robots, potentially exposing Unitree, which generated about 13% of its revenue from the U.S. last year.

Access to Nvidia hardware and software is another risk, as Chinese robotics companies rely on the technology to power their systems. “Chinese robot producers are not yet in a position to do without Western components completely,” Pross said.

China’s control over rare earths used in robot actuators and motors could nevertheless give its manufacturers an advantage, according to Bernstein analyst Dien Wang.

The Bigger Robotics Opportunity

The potential market is attracting major players, including Tesla, whose CEO Elon Musk is expanding production plans for Optimus humanoid robots. At the same time, some researchers argue that the future of robotics will not be limited to humanoids: quadruped and purpose-built robots can be cheaper and more reliable for repetitive industrial tasks, while humanoids may be better suited to unpredictable environments.

For Unitree, the challenge is no longer proving that its robots can perform impressive tricks. It is proving that they can do enough useful work to justify a $9 billion valuation.

Aretilaw firm
Uol
The Future Forbes Realty Global Properties
eCredo

Become a Speaker

Become a Speaker

Become a Partner

Subscribe for our weekly newsletter