Microsoft backstabbed OpenAI, Nvidia InfiniBand is in danger?

In this era of information explosion, data centers have become the digital backbone that underpins modern society. And when we talk about the network architecture of the data center, InfiniBand and Ethernet are undoubtedly two loud names. But just recently, a technological revolution is taking place, with the combination of Microsoft and OpenAI heralding a new era of networking — the leap from InfiniBand to Ethernet.

InfiniBand, a high-speed networking technology driven by companies such as NVIDIA, was once the darling of high-performance computing with its unmatched transmission speeds and low latency. However, as technology evolves and market demands change, Ethernet, the more ubiquitous and cost-effective networking solution, is starting to show its unique appeal.

Ethernet's counterattack did not happen overnight, and it has gradually won the favor of the data center field due to its advantages in versatility, scalability, and cost-effectiveness. Especially in the "Stargate" plan between Microsoft and OpenAI, the selection of Ethernet is undoubtedly a major blow to Nvidia InfiniBand.

The computing power business behind Microsoft and OpenAI's "own son" - "Stargate".

In tech giant Microsoft's grand blueprint, artificial intelligence occupies a central place. Microsoft's AI ambition lies not only in the innovation of its own products and solutions, but also in the cultivation and support of external innovation forces. Microsoft's investment in OpenAI is a concrete manifestation of this ambition.

OpenAI, as a rising star in the field of artificial intelligence, has attracted global attention with its open research environment and breakthrough research results. The cooperation between Microsoft and OpenAI is not only a financial investment, but also a deep integration of technology and resources. The jointly built super data center will become a milestone in the cooperation between the two parties and carry an important mission to promote the development of AI technology.

According to reports, Microsoft's total investment in OpenAI is as high as tens of billions of dollars. This funding not only provides sufficient impetus for OpenAI's research, but also lays a solid foundation for the "Stargate" super data center jointly built by the two parties. This data center will be a powerful backing for AI research, supporting large-scale data processing and complex algorithm training.

For OpenAI, the computing power demand behind the "Stargate" project is the core of its pursuit. In the world of artificial intelligence, computing power is the fuel that drives research forward. Whether it is the training of deep learning models or the iteration of complex algorithms, powerful computing power is inseparable.

The goal of the Stargate project is to build a supercomputer that can meet OpenAI's computing power needs for years or even decades to come. The supercomputer is expected to feature millions of dedicated server chips, providing unprecedented computing density and processing speed. Such a scale of computing power will enable OpenAI to conduct more in-depth and extensive exploration in the field of AI.

It is reported that the computing power of "Stargate" will far exceed the capacity of existing data centers and achieve exponential improvement. This will not only bring a revolutionary breakthrough to OpenAI's research, but also set a new benchmark for the entire AI industry. In this process, the cooperation between Microsoft and OpenAI will jointly promote the development of AI technology to a higher level.

InfiniBand and Ethernet have reached the threshold of Armageddon

In the network architecture of the data center, the battle between InfiniBand and Ethernet is a battle of speed, cost, and future potential. InfiniBand, driven by companies like NVIDIA, has carved out a niche in high-performance computing (HPC) with its high-speed transmission and low-latency features. However, with the development of technology and changes in market demand, Ethernet is gradually showing a counterattack trend due to its versatility and cost-effectiveness.

The history of InfiniBand technology is a microcosm of the high-performance computing networking field. In its early days, InfiniBand quickly gained recognition in scientific and engineering computing for its high-speed data transmission capabilities designed specifically for HPC. For example, early InfiniBand networks have demonstrated the advantages of high speed and low latency in areas that need to process large amounts of data, such as financial modeling and oil exploration.

With the advancement of technology, InfiniBand has entered a mature stage, and its application has expanded to larger scale cluster computing. Landmark events in this phase include its adoption by several supercomputers, such as the Jaguar supercomputer in the United States, which briefly became the world's fastest supercomputer and whose network architecture is based on InfiniBand technology.

Further, InfiniBand ushered in the launch of the Quantum2 platform in the expansion phase, which is not only a major technological leap for NVIDIA in the InfiniBand space, but also an important innovation in the data center market. The launch of the Quantum2 platform enables the InfiniBand network to reach 400 Gbps and provides more efficient AI application acceleration capabilities. According to Nvidia's official data, the Quantum2 platform's network switch integrates 57 billion transistors on the silicon wafer, which is even more than the 54 billion transistors of Nvidia's own A100 GPU.

For NVIDIA, InfiniBand has not only delivered commercial success, but also strengthened its leadership position in data center solutions. With InfiniBand, NVIDIA is able to provide an end-to-end solution that includes GPUs, networking hardware, and software, building a robust ecosystem. This ecosystem not only provides users with a one-stop solution, but also lays a solid foundation for NVIDIA's long-term growth in the data center market.

However, the high cost and proprietary features of InfiniBand also limit its adoption in a wider market. According to market analysts, InfiniBand is often several times more expensive to deploy than Ethernet, which limits its competitiveness in cost-sensitive markets to some extent.

At the same time, the rapid development of Ethernet technology is gradually narrowing the speed gap with InfiniBand. Ethernet 25G, 50G, and even 100G standards are gradually meeting the demand for high bandwidth in data centers. The popularity of Ethernet is extremely high, and almost all network equipment supports the Ethernet standard, which makes it an unmatched advantage in the deployment and expansion of data centers. In addition, the relatively low cost of Ethernet equipment and cabling and the ease of maintenance are important considerations for organizations with limited budgets.

Microsoft's Stargate program with OpenAI is expected to cost more than $100 billion, a huge investment that reflects the urgent need for high-performance computing power in data centers. In this project, Ethernet was chosen as the network solution based on a comprehensive consideration of cost-effectiveness and future development. This decision is likely to trigger a re-evaluation of network options within the industry, driving further expansion of Ethernet adoption in high-performance computing.

Ethernet's counterattack did not happen overnight. In the process of its development, there are many typical examples and stories. Google's B4 network, for example, is a successful example of large-scale Ethernet deployment. Google's use of Ethernet technology to build a high-speed network with global coverage to support its vast data centers and cloud computing services demonstrates the feasibility and benefits of Ethernet in large-scale deployments.

As technology continues to advance and optimize, Ethernet is expected to play an increasingly important role in data center network architecture. This contest of network solutions is not only a competition of technology, but also a preview of the future development trend of data centers.

"Stargate" stabbed the InfiniBand in the back

The Stargate project is a bold attempt by Microsoft and OpenAI, which not only represents a challenge to the boundaries of existing technologies, but also heralds a major transformation of data center network architecture. In this plan, OpenAI chose to abandon InfiniBand and adopt Ethernet as the network solution for its data center, a decision that hides deep strategic considerations.

Reducing reliance on InfiniBand was one of the important factors in OpenAI's decision. Although InfiniBand has irreplaceable advantages in the field of high-performance computing, its high cost and proprietary technical characteristics make OpenAI's technology choice limited to a certain extent. By moving to Ethernet, OpenAI is able to reduce its reliance on a single supplier and improve the flexibility and stability of its supply chain.

Moreover, building an independent network solution is of great strategic significance to OpenAI. Ethernet, as an open standard, is able to provide OpenAI with greater technical autonomy. This means that OpenAI can flexibly customize and optimize the network architecture according to its own needs, without being constrained by proprietary technology. This autonomy not only helps OpenAI stay ahead of the curve in technology, but also provides more possibilities for future technological innovation and applications.

"Stargate" will not be the first "betrayal"

As the heart of the digital world, every change in the network architecture of the data center can have a profound impact on the industry. Microsoft and OpenAI's "Stargate" program, with its transformation of network solutions, heralds a major shift in the future direction of the data center industry.

The impact of the change of network solutions on the industry is first reflected in the redefinition of technical standards. With the advancement of the Stargate program, the adoption of Ethernet as a network architecture will likely drive the industry in a more open and standardized direction. This transformation can not only reduce the construction and operating costs of data centers, but also promote the diversification of technological innovation and applications, bringing more development opportunities to the entire industry.

At the same time, this change will also reshape the competitive landscape in the data center space. In the face of Microsoft's and OpenAI's "Stargate" plan, other data center operators and cloud service providers will have to re-examine their technology strategies and market positioning. Some companies may choose to follow suit and invest in the R&D and deployment of Ethernet technology; Others may seek to maintain a competitive advantage through technological innovation or service differentiation. In any case, the implementation of the Stargate program will accelerate competition and consolidation within the industry, and promote the development of the industry in a more efficient and sustainable direction.