Ai Cloud & Chip Opportunities Heat Up: Amazon is making their own AI Chips & AWS Ai Cloud Platform

Ai Cloud & Chip Opportunities Heat Up: Amazon is making their own AI Chips & AWS Ai Cloud Platform


Amazon Generative Ai Cloud Platforms

No alt text provided for this image
Sagemaker & All ways to make LLM Ai Cloud Platforms- See also: https://aws.amazon.com/generative-ai/technology/

Through its Bedrock and Sagemaker Jumpstart generative AI service, Amazon Web Services will offer access to a number of Large Language Models (LLM) aka Generative AI models including:

  • Titan- Amazon's LLM Model
  • Cohere
  • AI21
  • Meta/Facebook's Llama 2
  • Google-backed Anthropic
  • Hugging Face - LLM community
  • Stability AI- model for turning text into images

This is in competition with Microsoft Azure that will have ChatGPT and Meta's Llama 2. However, ChatGPT is not available to the general public yet through Azure, so Amazon is ahead of Microsoft. Microsoft is also building Open AIs ChatGPT 4 into all its Office Products through Co-Pilot.

Amazon Making its own Generative Ai Chips

Amazon launched its own ARM based server chips in 2018, Graviton, after it purchased Annapurna Labs. when it purchased a semiconductor manufacturer. Then in 2018 it started to develop its AI chips releasing Inferentia in 2019 and Trainium in 2021 for use in its Ai Cloud computers. Following Google's TPU chips in 2016.


AMD's AI Chip to Launch 4th Quarter of 2023 & Desperately Needed with Short Supply of NVIDA

AMD's new Generative AI chip the MI300X is due to release Q4 of this year and will challenge NVIDA's market share. AMD has collaborated with Microsoft on a chip focused narrowly on LLM Generative Ai models vs NVIDIA's more general purpose Ai/ML processing. AMD MI300X is the first GPU that is powerful enough to run LLM Ai model with 80 Billion parameters in memory. NVIDA does have 15 years of software to support its Ai chips which will take time to catchup with. This should help AMD massively increase its revenue:

No alt text provided for this image

CHIPS ACT Allocated $53B for Domestic Semiconductor Manufacturing

Under the CHIPS Act the Commerce Department has allocated $53 Billion in Federal Funds and tax credits to boost semiconductor manufacturing, through its Chips Program Office: see application info here. The is lead by Todd Fisher, from KKR, and a dozen other Wall Street team members to advance this investment.

No alt text provided for this image
https://www.wsj.com/articles/chips-semiconductors-manufacturing-china-taiwan-11673650917?mod=article_inline

Private Semiconductor Investment

No alt text provided for this image

US Restrictions on Sale of Chips to China

In 2022 the US implemented "rules [that] will require U.S. chip makers to obtain a license from the Commerce Department to export certain chips used in advanced artificial-intelligence calculations and supercomputing—crucial technologies for modern weapons systems, senior administration officials said." The US is considering to put more restrictions on chips, including not only NVIDA's high end Ai A1000 but also the A800. Part of the reason of the new rules is to prevent players like NVIDA from trying to get around the rules. But AMD sees these rules an an opportunity with the right kind of Ai chips for China vs the domestic chips. There has also been a rally from the US to prevent other countries like Japan and Netherlands from selling semiconductor manufacturing equipment to China.

China retailed with its own limits on export of raw materials on semiconductors and solar panels. But it seems to have much less of an effect.

AI Chip Start-Ups

Start-ups in the AI Chip space are raising a lot of capital but AMD is the one that is likely to take real market share from NVIDA.

No alt text provided for this image
https://www.ft.com/content/1fca1f48-44d8-43ac-bfdd-251b0025347e



d

要查看或添加评论,请登录

Kevin Assemi的更多文章

社区洞察

其他会员也浏览了