Industry Predictions 
Sanjay Gangal
Sanjay Gangal is the President of IBSystems, the parent company of AECCafe.com, MCADCafe, EDACafe.Com, GISCafe.Com, and ShareCG.Com. EDACafe Industry Predictions for 2023 – VerifAIJanuary 19th, 2023 by Sanjay Gangal
By Sandeep Srinivasan, CEO, VerifAI Inc
|
| Company Name | Capital Raised | Product | Primary Use-Case | Use-Model |
|---|---|---|---|---|
| SambaNova | > $1 Billion | SuperComputer | Faster Training of Large Language Models (LLM) | AI Platform as a Service HPC Hardware |
| Cerebras | > $750 Million | WSI SuperComputer | Faster Training for LLM | AI Platform as a Service HPC Hardware |
| GraphCore | > $692 Million | Multi form Factor Accelerator | Faster Training for GNN, LLM | AI Platform as a Service HPC Hardware |
| Groq | > $360 Million | Multi form Factor Accelerators | Fast Training for LLM , CV | AI Platform as Service Hardware Sales |
| Tenstorrent | > $ 350 Million | Multi form Factor Accelerators | Fast Training for LLM | AI Platform as Service Hardware Sales |
Significant Capital Raised and Single Use-Cases:
These companies have raised significant capital, and ironically, all have a single compelling use case, which is to train Large Language Models (LLMs) such as GPT-3 and Bert. Now, the question remains as to how many companies care to train their LLM rather than fine-tune Pre-trained models from hugging-face and other open sources.
Cloud hosting Business Model:
The business model for these large AI accelerators is to build their cloud offering for developers to use. Creating a cloud computing business in a hyper-competitive cloud-provider market is a very high bar. If the only value proposition is to provide faster training for AI models, only some developers would consider moving to such a cloud offering. It is hard to imagine that these AI supercomputer companies will make a dent in the cloud computing market and convince developers to move to their cloud to train LLMs!
Only a few organizations, such as Oak-Ridge National Labs, Livermore National Labs, and Argonne national labs, can afford to buy hardware from companies like Sambanova, Cerebras, and Groq. As a result, these companies may look like Cray Computing, where they sold one supercomputer to each of these large institutions and stopped there.
Challenges:
The biggest challenge for these AI-Chip companies is to convince one of the prominent cloud providers to use their hardware as a part of their compute-platform offering. But the bar is very high here, and the competition is fierce from internal efforts, such as the Google TPU.
2023 Predictions for AI-Chip Companies:
- Acquisition: There will be one or two winners in the AI-Chip space that AWS, GOOG, MSFT, or ORCL, may acquire, if they are lucky. The next level of opportunity is for HPE, DELL, CISCO, or other opportunistic OEMs to acquire them.
The least likely acquirers are INTC, AMD, and NVIDIA; However, these companies are constantly looking to improve their compute pipelines, but they will rarely pay for such highly valued startups that may have minimal revenues. - Potential Foreign Acquisition: The worst-case scenario for these companies is an acquisition by foreign companies, which would fall under US-fed scrutiny.
- Relevant for a Few, not Mainstream: If foreign acquisitions don’t happen, these companies will be the next generation of Cray Computer Inc, i.e., relevant for a few customers, not for the mainstream.
2023 Predictions for EDA
- Verification will be a growth category in EDA. Innovation in Verification will come from startups like VerifAI Inc. Verification Services and Products will grow at a modest 5-7% CAGR.
- EDA business model conflicts with open source ML/AI: The big 3 EDA companies will ponder their business models. AI/ML improves efficiency, which may lead to fewer EDA licenses used. EDA companies will be conflicted about improving efficiency versus selling more licenses. In addition, EDA has a traditional mindset of patenting and protecting its ‘IP.’ In the case of AI/ML, most frameworks are open-source, it’s harder to patent ML/AI algorithms.
- RISCV-specific EDA tools will open up new market opportunities
- EDA industry will be impacted by US trade policy with China. This may include decline in oveall revenues from APAC
About Author: Sandeep Srinivasan, Founder and CEO, VerifAI Inc. Sandeep is a serial entrepreneur, he has founded 4 startups. His previous 2 startups focussed on algorithms to reduce power consumption on IC’s. Sandeep founded a location based messaging startup that uses AI to keep track of your family. Two of the startups were acquired by larger companies. Sandeep’s expertise is Building Products, Teams and companies from Ground up. His technical expertise is broad , in the areas of ML, CAD (Timing/Clocks/PD) and Software Architecture. Sandeep started his career @AMD, held Senior R&D management positions @Cadence Design Systems and HLDS. Sandeep has a MS EE/CS from U Wisconsin Milwaukee and LEAD Program Certificate from Stanford GSB. Sandeep holds multiple patents in the area of location based messaging and machine learning for verification
Category: Predictions
