A few days ago, I published a post “Why software underperforming amid the AI craze?“, Today’s article lists some practical examples and figures for your reference, so that readers can have an idea of generative AI cost.
Generative AI requires huge hardware expenditures
First of all, this wave of artificial intelligence is triggered by generative artificial intelligence such as ChatGPT developed by OpenAI. Generative artificial intelligence requires huge capital investment in data centers and servers, and is even larger than cloud computing. Related companies will spend US$200 billion to expand data centers in 2024, 50% higher than in 2023.
Traditional AI costs less
Please note: Artificial intelligence has been developed for nearly half a century and has been evolving silently. However, it has not made any major breakthroughs, been widely accepted and adopted, or caused a trend. It was not until the release of generative artificial intelligence that people’s views on the practicality of generative artificial intelligence changed. In the past, the cost of hardware construction required for artificial intelligence applications was not high, and it was actually insignificant in terms of enterprise expenses.
Note: There have always been artificial intelligence manufacturers, large and small, in the market. They have been around for decades, but they have not received the high attention they have today. In my post “Major Artificial Intelligence Listed Companies in the U.S. Stock Market”, I listed some of the more famous listed artificial intelligence companies in the U.S. stock market for readers.
LLM capabilities is decided by hardware
The core of generative artificial intelligence is actually the LLM (large language model) of artificial intelligence, and the ability of LLM is directly proportional to its scale. That is, the higher the expenditure on data centers, servers, and infrastructure, the higher the probability of success in building LLM.
Because of this, relevant companies spend huge sums of money to purchase a large number of GPUs to develop LLM. But here comes the problem. For traditional software companies, the barrier to entry is too high to create new application scenarios with generative artificial intelligence.
Some famous cases
Anthropic’s estimation
The CEO of Anthropic, OpenAI’s biggest rival, when he interview with “In Good Company” podcast said in June 2024, : The current cost of AI model training is US$1 billion, and in the next three years, this number may rise to US$10 billion or even US$100 billion.
How much does each version of ChatGPT cost?
The training costs of various versions of OpenAI’s ChatGPT large model are as shown in the following table:
| Release time | Training cost (in USD) | |
| ChatGPT 5.0 | Summer, 2024 (estimated) | US$1.25 billion to US$2.25 billion |
| ChatGPT 4.0 | March 14, 2023 | US$100 million |
| ChatGPT 3.0 | June, 2020 | USD 4.5 million |
| ChatGPT 2.0 | Febuary, 2019 | USD 40,000 |
The training cost of well-known AI large models
The training costs of various versions of the well-known artificial intelligence large models on the market are as shown in the following table (the latest one is at the top):
| Release time | Training cost (in USD) | |
| Gemini Ultra | 2023 | 191,400,000 |
| Llama 2 70B | 2023 | 3,931,897 |
| ChatGPT 4.0 | 2023 | 78,352,034 |
| PaLM (540B) | 2022 | 12,389,056 |
| Megatron-Turing NLG 530B | 2021 | 6,405,653 |
| ChatGPT 3.0 175B (Davinci) | 2020 | 4,324,883 |
| RoBERTa Large | 2019 | 160,018 |
| BERT-Large | 2018 | 3,288 |
| Transformer | 2017 | 930 |
Related articles
- “Why software underperforming amid the AI craze?“
- “How Much Does Generative AI Cost?”
- “Major artificial intelligence companies in US stocks market“
- “Artificial intelligence investment trap“
- “The artificial intelligence bubble in the capital market is forming“
- “OpenAI, the Generative Artificial Intelligence rising star and ChatGPT“
- “The Importance of Small Language Models (SLM) in AI, how is Apple AI different from others?“
- “Artificial intelligence benefits industries“
- “Four chip companies account for one-third of S&P 500 gains so far this year＂
- “Top five lucrative artificial lucrative intelligence listed companies“
Disclaimer
- The content of this site is the author’s personal opinions and is for reference only. I am not responsible for the correctness, opinions, and immediacy of the content and information of the article. Readers must make their own judgments.
- I shall not be liable for any damages or other legal liabilities for the direct or indirect losses caused by the readers’ direct or indirect reliance on and reference to the information on this site, or all the responsibilities arising therefrom, as a result of any investment behavior.
