Winning first doesn't mean winning! How long can NVIDIA and OpenAI monopolize OpenAI?

Wallstreetcn
2024.01.21 11:44
portai
I'm PortAI, I can summarize articles.

On the iron throne, there is always a new king replacing the old one. The replacement for GPUs is emerging, and the first and foremost is ASIC (Application Specific Integrated Circuits). Major cloud providers are actively developing ASICs to reduce their reliance on NVIDIA.

In the AI boom of 2023, the undisputed leaders are two major winners: NVIDIA, the GPU manufacturer that dominates the market, and OpenAI, the creator of OpenAI.

However, the competition in the field of generative AI has just begun. There have been many examples in history of pioneers being surpassed. There was AltaVista, which developed a search engine before Alphabet-C, and Nokia, which once sold phones that could circle the Earth several times. But their names have faded away in the tide of business history.

Similarly, how long can NVIDIA and OpenAI maintain their leading advantages?

Alternatives to GPUs are emerging

In the gold rush of artificial intelligence, the initial winners were infrastructure providers such as chip manufacturers and cloud service providers. The clear leader in this market is NVIDIA, the giant of AI, which almost monopolizes the market.

According to data compiled by Orient Securities, as of Q4 2022, NVIDIA's market share in the GPU market is as high as 84%, far surpassing its competitors, thanks to its powerful CUDA ecosystem.

However, Deutsche Bank analysts pointed out that alternatives to GPUs are emerging, with ASICs (Application Specific Integrated Circuits) being the most prominent. ASICs are specialized circuits designed and manufactured for specific algorithms, tailored for workloads such as AI training. Major cloud vendors are actively developing ASICs to reduce their dependence on NVIDIA.

The most famous ASIC is Alphabet-C's TPU, which is said to be two to three times more efficient than similar GPUs. Amazon's AWS has also developed Trainium and Inferentia chips, with the latter reportedly offering a 50% improvement in performance per watt.

In May of this year, Meta released the first-generation Meta Training and Inference Accelerator (MTIA). In November of last year, Microsoft launched its own custom-designed chip, Maia AI Accelerator. Maia, which went online earlier this year, along with its equivalent CPUs, completes Microsoft's infrastructure from chips to software and cooling systems.

Although Deutsche Bank acknowledges that dethroning NVIDIA is not an easy task, especially since it not only has chips and data centers but also controls the CUDA ecosystem, embedding its products firmly within it. However, NVIDIA is already facing tougher competition in the data center switch market, and the rise of small-scale LLM is driving the rise of edge computing, which may lead to a decline in demand for GPU-driven data centers:

Competitors are entering the market. NVIDIA is the leader in the "switch" field, a technology that allows GPUs and CPUs in data centers to communicate with each other. However, its high-bandwidth, low-latency Infiniband switches face competition from Ethernet switches provided by Broadcom and Marvell, which seem to be more efficient in terms of scale. The rise of mini LLMs may lead to more AI processing at the "edge" of smartphones, cars, and other devices during the reasoning stage, eliminating the need to route to data centers filled with NVIDIA chips.

How secure is OpenAI's dominant position?

In addition to NVIDIA's monopoly in hardware, OpenAI is another leader in the AI field with absolute software strength.

As the pioneer of this generation's AI boom, OpenAI's chatbot has become synonymous with generative AI.

In terms of traffic, according to Similarweb data, OpenAI has a monthly visit volume of about 1.6 billion, which is more than 70 times that of Anthropic's Claude, and more than four times that of Alphabet-C Bard.

However, Deutsche Bank pointed out three major issues with OpenAI:

First, last year's power struggle exposed the irreconcilable tension in OpenAI's ethical identity core. In short, the greater the power, the greater the responsibility. OpenAI is now a provider of capital and technology, rather than the higher ideals of the "effective altruists" who dominated the board of directors of OpenAI 1.0.

Second, it seems to consolidate Microsoft's position as a power broker in Silicon Valley AI. It was reported that at this time last year, OpenAI received a $10 billion investment from Microsoft, and now the company is irreversibly indebted to Microsoft. Microsoft not only benefits from OpenAI's development, but also develops its own AI with this advantage, reducing its dependence on OpenAI. This includes bundling 365 Copilot into its products.

Third, the power struggle at OpenAI has shaken the attention of customers and regulatory agencies, making alternative products from competitors more attractive than before. It also puts pressure on regulatory agencies, urging them to think twice and not allow a small company with unconventional governance structure to make progress in a critical technology that could shake the fate of humanity. For Microsoft or Alphabet-C, it's one thing if they can't fail, but for a star startup, it's a completely different story. This may lead to stricter regulatory language in the future.