On Tuesday, NVIDIA and Google Cloud announced an expansion of their partnership, introducing AI infrastructure and software for large-scale generative AI models and accelerated data science tasks.
In a fireside chat at Google Cloud Next, Google Cloud CEO Thomas Kurian and NVIDIA CEO Jensen Huang discussed Google’s utilization of NVIDIA H100 and A100 GPUs for internal research and inference in divisions such as DeepMind.
Huang highlighted the deep collaboration that led to NVIDIA GPU acceleration for the PaxML framework. Developed by Google, the Jax-based ML framework is designed for training large-scale models, offering experimentation and parallelization capabilities.
According to NVIDIA’s blog post published on Tuesday, Google employed PaxML with NVIDIA GPUs for internal model development, including DeepMind and research projects. Both companies also announced PaxML’s immediate availability on the NVIDIA NGC container registry, allowing AI developers to use NVIDIA® H100 and A100 Tensor Core GPUs.
“Google Cloud has a long history of innovating in AI to foster and speed innovation for our customers,” Kurian said. “Many of Google’s products are built and served on NVIDIA GPUs, and many of our customers are seeking out NVIDIA accelerated computing to power efficient development of LLMs to advance generative AI.”
Moreover, Google will be incorporating serverless Spark with NVIDIA GPUs via its Dataproc service. The collaboration aims to enable data scientists to speed up Apache Spark workloads.
NVIDIA’s Stock Close at a Record High
Following the partnership expansion announcement, NVIDIA’s stock rose by over 19% on Tuesday, closing at a record high above $487 with a market cap of $1.2 trillion.
Adding to the momentum of NVIDIA’s stock price surge, Iris Energy, a Nasdaq-listed Bitcoin mining platform, revealed its purchase of 248 NVIDIA H100 GPUs for approximately $10 million on Tuesday.
The company said that the GPU purchase is for generative AI purposes as the technology represents an additional opportunity.
While Bitcoin mining remains Iris Energy’s core business, the company said that generative AI could enable it to power the following:
Additionally, the company said that the GPUs could allow it to service adjacent computing markets such as generative AI.
“Leveraging our next-generation data centers into generative AI is an exciting opportunity, particularly given current industry shortages in rack space and compute,” Daniel Roberts, Co-Founder and Co-CEO of Iris Energy, said in a statement. “We believe demand for sustainable computing is unlikely to go away, and feel we are uniquely positioned to capture ongoing growth in the broader industry; whether that be ASICs for Bitcoin mining, or GPUs for generative AI and beyond.”
Could Tesla Emerge as a Rival to NVIDIA?
On Tuesday, Tesla launched a massive $300 million NVIDIA H100 GPU AI computing cluster comprising 10,000 units, to facilitate the training of Full Self Driving (FSD) – the automaker’s label for its beta testing program focused on achieving fully autonomous driving.
According to Sawyer Merritt, the latest H100 GPUs can enable Tesla to enhance FSD training speed and quality.
“Due to real-world video training, we may have the largest training datasets in the world, hot tier cache capacity beyond 200PB – orders of magnitudes more than LLMs,” declared Tim Zaman, AI Infra & AI Platform Engineering Manager at Tesla.
However, it’s widely known that the demand for NVIDIA GPUs surpasses the available supply. Merritt revealed that Tesla is investing over $1 billion to further develop its own supercomputer, named Dojo, to increase the company’s compute capabilities.
Dojo uses a hyper-optimized, in-house custom chip design. If built to scale, Tesla’s chips could potentially rival NVIDIA.
Oppenheimer analyst Rick Schafer added that Tesla plans to spend over $2 billion on AI training this year, and another $2 billion in 2024.
As a result of Tuesday’s news, Tesla shares rose more than 7% on Tuesday, their largest single-day percentage increase since March and their highest in three weeks.
Tesla CEO Elon Musk has always hinted his AI ambitions, even before forming xAI in July, his new artificial intelligence venture. Musk was a co-founder at OpenAI, but stepped down from the company’s board in 2018.
NVIDIA’s Q2 Momentum Continues
Last week, NVIDIA reported robust Q2 2023 financial results, revealing an impressive 101% YoY surge in revenue, accompanied by an impressive 854% rise in earnings per diluted share.
The surge in company revenue owes itself to the heightened demand for its H100 AI infrastructures, particularly from prominent cloud service providers and other developers in the generative AI sector, as underscored by Tuesday’s announcements.
Analysts from Oppenheimer, Goldman Sachs, BofA, and others have raised their price targets for NVIDIA’s shares to $650 while HSBC and BNP Paribas are particularly bullish as they increased their price targets to $800 and $745 respectively.
Source: mPost
