On October 15th, NVIDIA (NVDA) announced the official delivery of the world’s smallest AI supercomputer, the NVIDIA DGX Spark.
DGX Spark Supercomputer Debuts
To celebrate the global delivery of DGX Spark, Jensen Huang personally visited SpaceX’s Starship base in Texas and delivered one of the first DGX Sparks to SpaceX Chief Engineer Elon Musk.
DGX Spark integrates the entire NVIDIA AI platform (including GPUs, CPUs, networking, CUDA libraries, and the NVIDIA AI software stack) into a compact system suitable for labs or offices. It delivers the powerful performance needed to accelerate agentic AI and physical AI development, transforming desktop computers into AI development platforms. Jensen Huang said, “In 2016, we developed the DGX-1 to provide AI researchers with a dedicated supercomputer. The DGX-1 not only ushered in the era of AI supercomputers, but also revealed the principles of scale that drive modern AI development. Now, with DGX Spark, we’re returning to our original aspiration: putting AI computing in the hands of every developer, thereby unleashing the next wave of technological breakthroughs.”
Global AI’s boom continues
As a leading indicator of global innovation, AI hardware has advanced by leaps and bounds. NVIDIA has already packed 1PFLOPS of computing power, once required by a cluster of machines, into desktop-level devices. Developers with pre-installed NVIDIA hardware can also use DGX Spark to create AI agents and run advanced software stacks locally.
Global AI’s boom continues in 2025, with continued momentum on the application side. Computing power companies are expected to enter a period of rapid growth. Brokerage firms are optimistic about the continued high growth in computing power demand driven by AI applications, and believe that AI applications both domestically and internationally are reaching a turning point in their widespread adoption. Good news from the industry chain is expected to continue to catalyze the market, and we are optimistic about investment opportunities in China’s computing power sector in the second half of this year and beyond.
Related companies drive continued rapid growth in computing power.
OpenAI
OpenAI, the developer of ChatGPT, has ambitions beyond the present. OpenAI recently partnered with Broadcom to produce its first independently developed artificial intelligence (AI) processor.
The two companies stated that OpenAI will design the chip, while Broadcom will begin development and deployment in the second half of 2026, with deployment of the new custom chips to be completed by the end of 2029. These chips will have a computing power of 10 gigawatts, equivalent to the electricity needs of more than 8 million American homes.
OpenAI’s partnership with Broadcom is the latest in a series of large-scale chip investments by tech companies. This collaboration represents a new attempt by tech companies to develop custom AI chips, highlighting the tech industry’s thirst for computing power in the race to build advanced AI systems.
Altman also revealed to employees that the partnership with Broadcom is a key step in building the infrastructure needed to unleash the potential of AI. He envisions building OpenAI into a full-stack AI powerhouse akin to Google, spanning large models, consumer hardware, processors, social platforms, and data centers. Intel (INTC)
On October 15th, Reuters reported that Intel announced a new AI chip for data centers, slated for release next year, marking the company’s renewed push into the AI chip market.
Intel CTO Sachin Kati stated at the Open Compute Summit that the new graphics processing unit (GPU) will be optimized for energy efficiency and support a variety of use cases, including running AI applications or inference. Kati stated that Intel will release new data center AI chips annually going forward, a release cadence consistent with AMD, Nvidia, and several cloud computing companies that develop their own chips.
AMD (AMD)
Chip designer Advanced Micro Devices (AMD) has reportedly announced a partnership with Oracle to deploy approximately 50,000 of its latest AI chip, the MI450, in Oracle’s data centers, marking a deeper integration of the two companies in the field of artificial intelligence computing power.
According to the plan, AMD will deploy this cluster in Oracle’s data centers starting in the third quarter of 2025, with a total computing power equivalent to 200 megawatts of electricity. Both parties stated that the partnership will be further expanded after 2027. Market analysts believe this project will become another significant competitor in the AI computing market after NVIDIA. AMD’s MI450, the most advanced GPU (graphics processing unit) to date, will be installed in the company’s independently developed Helios server rack system. Combined with AMD’s own central processing unit (CPU), it directly competes with NVIDIA’s next-generation “Vera Rubin” series AI chips.
WiMi (WIMI)
Accordingly, Wimi Hologram Cloud Inc. is keeping pace with tech giants, continuing to maintain computing power autonomy through proprietary chip development. Through technical breakthroughs, ecosystem integration, and scenario adaptation, it has built a full-stack proprietary research and development platform covering chip design, computing infrastructure, and multimodal applications. This will break through bottlenecks in the AI chip ecosystem, accelerate the large-scale application of AI in computing power, modeling, and industrial fields, and meet the diverse needs of AI computing power.
Currently, WiMi is focusing on high-end computing chips, industrial multimodal algorithms, and software and hardware adaptation technologies. The company is building a computing infrastructure covering the cloud and edge, supporting the integration of AI chips with diverse architectures to meet the diverse needs of training and inference. Behind this technological breakthrough lies WiMi’s full-stack defense. Its strategy also encompasses quantum computing hardware, an open-source ecosystem, and cross-domain technology integration, aiming to strengthen the foundation of the chip industry.
In summary
In today’s AI landscape, computing power can only improve the efficiency of early AI training. In the “post-training” phase, when AI truly takes shape, computing power demand increases exponentially. Therefore, it’s clear that the AI technology revolution is essentially a computing power revolution. The AI industry’s “arms race” has entered a new phase, and all of humanity is witnessing this unprecedented gamble.
