ROBT - One Size Does Not Fit AI
2023-12-19 18:40:00 ET
Summary
- ASIC chips are gaining significance across the AI landscape, offering tailored solutions that promise to redefine the future of AI applications.
- With each passing month, it seems, AI applications grow more integral to simplifying our work and enriching daily life.
- ASIC processors have surpassed CPUs and GPUs in Bitcoin mining, thanks to the custom chips’ lower electricity consumption and better computing prowess.
- As generative AI takes off, others are working on chips that can optimize their data centers.
- When it comes to AI, we believe one size doesn’t fit all - and that customization could unwrap plenty of attractive opportunities for chipmakers.
By Yan Taw Boon
ASIC chips are gaining significance across the AI landscape, offering tailored solutions that promise to redefine the future of AI applications.
Snazzy AI tech is high on holiday wish lists - from Ray-Ban Meta smart glasses that can record videos, to Humane AI Pins that act like wearable personal assistants and control an array of smart devices. With each passing month, it seems, AI applications grow more integral to simplifying our work and enriching daily life.
Thus far, graphics processing units (GPUs) - from the likes of Nvidia (NVDA) and AMD (AMD) - have been the primary workhorses for training general-purpose large language models (LLMs), such as ChatGPT. But for all their versatility, these standard GPUs can yield mediocre performance and efficiency, depending on how they’re used.
Now come custom Application-Specific Integrated Circuit ((ASIC)) chips, designed to deliver more efficient and tailored solutions for diverse AI applications.
Take the cryptocurrency chip market, for example: For years Bitcoin miners relied on Central Processing Units (CPUs) that inhaled gobs of energy. Now ASIC processors have surpassed CPUs and GPUs in Bitcoin mining, thanks to the custom chips’ lower electricity consumption and better computing prowess. Morgan Stanley (MS) estimates that, between 2023 and 2027, the market for AI-focused ASIC chips is expected to grow 85% annually, to $30 billion. 1
Tech giants are placing their bets on ASICs. Google ([[GOOG]], [[GOOGL]]) has been partnering with Broadcom (AVGO) to produce its 5 th -generation Tensor Processing Units (TPUs), which can handle AI workloads roughly 10 times faster than traditional CPUs and GPUs. 2
As generative AI takes off, others are working on chips that can optimize their data centers. Amazon (AMZN) has teamed up with Marvell (MRVL) to develop its Graviton AI chips, designed to improve processing speed and reduce power consumption. It is also collaborating with Alchip Technologies (ALCPF) on its Trainium and Inferentia lines to efficiently train AI models and enable them to make accurate predictions.
Meanwhile, Microsoft (MSFT) has partnered with Global UniChip and Marvell to develop its Maia 100 and Cobalt 100 chips, respectively. The Maia 100 chip will seek to consume less power when training AI models, while the Cobalt 100 seeks to ramp up processing speed. In the automotive market, Tesla (TSLA) is working with Alchip on its supercomputing AI chip, Dojo, designed to enable self-driving vehicles.
When it comes to AI, we believe one size doesn’t fit all - and that customization could unwrap plenty of attractive opportunities for chipmakers.
Sources: 1) Morgan Stanley ; 2) “ Google Says TPU v4 Supercomputer Performs Faster Than Nvidia’s A100 Chip, NS Business ”, April 5, 2023.
This material is provided for informational purposes only and nothing herein constitutes investment, legal, accounting or tax advice. This material is general in nature and is not directed to any category of investors and should not be regarded as individualized, a recommendation, investment advice or a suggestion to engage in or refrain from any investment-related course of action. Investment decisions and the appropriateness of this material should be made based on an investor's individual objectives and circumstances and in consultation with his or her advisors. Information is obtained from sources deemed reliable, but there is no representation or warranty as to its accuracy, completeness or reliability. All information is current as of the date of this material and is subject to change without notice. The firm, its employees and advisory accounts may hold positions of any companies discussed. Any views or opinions expressed may not reflect those of the firm as a whole. Neuberger Berman products and services may not be available in all jurisdictions or to all client types. This material may include estimates, outlooks, projections and other “forward-looking statements.” Due to a variety of factors, actual events or market behavior may differ significantly from any views expressed.
This material is not intended as a formal research report and should not be relied upon as a basis for making an investment decision. The firm, its employees and advisory accounts may hold positions of any companies discussed. Specific securities identified and described do not represent all of the securities purchased, sold or recommended for advisory clients. It should not be assumed that any investments in securities, companies, sectors or markets identified and described were or will be profitable.
Investing entails risks, including possible loss of principal. Investments in hedge funds and private equity are speculative and involve a higher degree of risk than more traditional investments. Investments in hedge funds and private equity are intended for sophisticated investors only. Indexes are unmanaged and are not available for direct investment. Past performance is no guarantee of future results.
This material is being issued on a limited basis through various global subsidiaries and affiliates of Neuberger Berman Group LLC. Please visit www.nb.com/disclosure-global-communications for the specific entities and jurisdictional limitations and restrictions.
The “Neuberger Berman” name and logo are registered service marks of Neuberger Berman Group LLC.
© 2009-2023 Neuberger Berman Group LLC. All rights reserved.
Editor's Note: The summary bullets for this article were chosen by Seeking Alpha editors.
For further details see:
One Size Does Not Fit AI