[Introduction to New Wisdom]As soon as Altman's $7 trillion plan came out, the world was in an uproar. This amount accounts for 10% of global GDP, equivalent to 2.5 Microsofts, 3.75 Googles, 4 Nvidias, 7 Metas, or 11.5 Teslas.
Yesterday, when the news came out that Sam Altman had raised US$7 trillion to build a chip empire, public opinion was in an uproar.
Advertisement
7 trillion US dollars is 10% of global GDP, equivalent to the GDP of the United States, or 1/3 of China's GDP.
This number is really incomprehensible, unless OpenAI is convinced that its technology will fundamentally reshape the entire world. Otherwise, artificial intelligence is in a huge bubble.
Advertisement
7 trillion US dollars can buy 2.5 Microsofts, 3.75 Googles, 4 Nvidias, 7 Metas, and 11.5 Teslas.
Or simply buy Nvidia, Intel, Microsoft, and Google as a whole package.
With this money, Altman can buy Nvidia, AMD, TSMC, Broadcom, ASML, Samsung, Intel, Qualcomm, Arm and other companies, and the remaining money can be packaged into Meta and take home another $300 billion.
This money is basically the entire semiconductor ecosystem.
The total national product of the United States is 23.36 trillion; the United States spent 4 trillion on World War II; and with US$330 billion, it can solve the world's hunger problem by 2030.
No wonder everyone exclaimed: Altman's ambition is unprecedented and beyond imagination!
Altman said it was just “chaotic goodness.”
Altman wants to become a global overlord and overthrow human governments?
Friends also commented sharply on Sam Altman's ambition: the father of the fourth industrial revolution is here to cut the second wave of leeks.
Insider “Deep Space” said that if the report did not misinterpret Altman, then he is either crazy or engaging in an AI coup.
If the GDP of all countries in the world exceeds one trillion, it will rank 18th. If it can really raise 1% of the total global GDP, OpenAI will be rich as a country.
Such a large amount of money has been invested in the semiconductor industry, which cannot be absorbed by a single country and can only be spread globally.
This kind of behavior has gone beyond the scope of business, but is a national political behavior. And if this money is controlled by AI, AI will really overthrow the human government.
Source: Zhiyou “Deep Space”
Perhaps he wants to be Musk's “second” and complete his plan to “immigrate to Mars” in the field of integrated circuits?
A close friend “Jiuxianghelongya” said that it can be seen that Altman is also trying to save the country like Musk and turned to Middle East capital when being targeted by the board of directors.
However, only if his vision reaches 1 to 2 levels higher than the current ceiling of human chips, will he be able to see excess returns.
Friend “PENG Bo” said that the current strategic thinking of the United States is to attract money from all over the world. If all this money is used to invest in AI projects, the United States can lock the development of AI in these regions.
Fortunately, everyone now knows the strategic significance of local AI companies. DeepMind insists on being in the UK and Mistral insists on France.
It can be seen that Altman is playing a big game. AI, nuclear fusion, and world currency are all supporting facilities to reshape the industry chain of semiconductor AI algorithms.
The friend “Me from the Star” said that he has seen through Altman's ambition: what he wants to do is to connect various breakthroughs and rearrange the world.
If someone really controls AGI, it will change the way of life for a community with a shared future for mankind.
Computing power is the nuclear weapon of the future, and those with computing power will win the world. Today, the algorithms, predictions, and software of AI companies around the world have reached a fever pitch, and computing power is their trump card.
Computing power is the nuclear weapon of the future
In fact, Altman's efforts to solve the current GPU shortage and change the landscape of the semiconductor industry is nothing new. GPU has long become the core competitiveness of Silicon Valley's AI field.
Last year, Nvidia's high-performance GPU H100, which is expensive but still hard to find, has long been a hot topic in the LLM field.
Google, Amazon, Meta, OpenAI and Microsoft are all using NVIDIA's GPUs. It can be said that NVIDIA monopolizes the current AI computing power market and directly holds pricing power, so its revenue has skyrocketed.
The H100 has become so expensive that Musk even shouted: GPUs are now more in demand than drugs.
However, there are already reports that the Nvidia H100 will be sold out before 2024.
As for how serious OpenAI’s GPU shortage is? Altman has been forced to appeal: everyone should stop using ChatGPT!
We have a huge shortage of GPUs and the fewer people using our product the better.
We would be happy if people used less because we don't have enough GPUs.
Sam Altman said that OpenAI is already severely GPU limited and has had to postpone many short-term plans (fine-tuning, dedicated capacity, 32k context windows, multi-modality).
The shortage of GPUs has also caused many users to complain that the API is too slow. What's more serious is that training GPT-5 requires 50,000 H100. It can be seen that Altman felt that rather than spending money to buy it from Nvidia, he might as well build it himself.
As early as December last year, Altman was revealed to be engaged in “chip trading,” which was also suspected to be the trigger for the OpenAI palace battle.
At that time, foreign media revealed that OpenAI signed a letter of intent worth US$51 million with the AI chip startup Rain AI in 2019, and would purchase Rain AI's chips after they are launched.
Rain AI is developing a “brain-like” NPU chip that can significantly reduce the cost of AI computing power. It is expected to be tape-out in December and available in October 2024.
It is worth noting that Sam Altman, as a shareholder of Rain AI, also personally invested US$1 million.
According to a person who did not want to be identified publicly, Sam Altman was previously fired by OpenAI’s former board of directors, in part because of the entangled relationship between his other investments and OpenAI.
In addition to OpenAI, major technology companies have also begun to make plans. Recently, Meta CEO Xiao Zha once again stated that the first requirement for building “general artificial intelligence” is “world-class computing infrastructure.”
By the end of this year, Meta will have about 350,000 H100s, and if other GPUs are included, the total will have the equivalent computing power of 600,000 H100s.
Not only that, Meta is also developing its own dedicated AI chip. Among them, the newly upgraded second-generation self-developed AI chip Artemis will be officially put into production this year.
Coincidentally, Microsoft also released two customized chips-Azure Maia 100 and Azure Cobalt 100 in November last year.
Among them, the Maia 100 GPU, which uses TSMC's 5nm process and has 105 billion transistors, will be able to compete with NVIDIA (H100) and AMD (MI300X) in terms of computing power, is far ahead in network IO, and slightly behind in terms of memory bandwidth. Apparently backward.
In addition, it will also support Microsoft's first implementation of sub-8-bit data types, the MX data type. In the MXInt8 format, Maia's computing power can reach 1600 TFLOPS, and in the MXFP4 format it is 3200 TFLOPS.
At the same time, Google and Amazon have also been developing TPU and Trainium chips respectively for many years.
But they are obviously still pale in comparison to Sam Altman's grand layout.
NYU professor analyzes “six” reasons why we must oppose
In this regard, Marcus, a professor at New York University, wrote that when Sam Altman made an astonishing request of $700 billion, he seemed to have reached the moment when his career began to decline after the peak.
Going a step further, Marcus listed “six” reasons why people should come together and make it clear to the ambitious young CEO that the world should not and would not revolve around him alone:
– Energy climate
If the $700 billion worth of AI infrastructure were stretched to its limits, it would consume massive amounts of energy. And we shouldn’t fool ourselves into thinking that all this energy comes from “renewable” sources.
Someone on X compared GenAI's energy needs to the energy consumption of the entire country of Germany. In this regard, Marcus believes that this estimate may actually be conservative given that models tend to become larger and more expensive to train.
Obviously, consuming such a huge amount of energy is no small matter. GenAI, in particular, has so far presented more potential than actual results, with no compelling reason to justify sacrificing the planet for it.
– natural resources
According to a report by Fortune magazine in September 2023, AI tools have caused Microsoft's water usage to surge by 34%; Meta's Llama 2 model is said to use twice as much water as Llama 1; another 2023 study found that OpenAI's GPT-3 consumed 700,000 liters of water during training.
However, OpenAI has never disclosed the relevant data of GPT-4, so it is even more difficult to imagine what the data of GPT-5 will be like.
– Economic resources
$7 trillion is almost 10 times what the United States spends on education in a year and 21 times the cost of ending global hunger.
Funds are not endless and every expenditure comes with an opportunity cost.
Although people dream that generative AI can greatly increase the economic aggregate, the current situation is that OpenAI has not yet achieved profitability due to extremely high operating costs.
Besides, giving all the money to a person based on his promise is a bit too childish.
– Financial risk
A project costing US$7 trillion will inevitably be viewed as a behemoth with no room for error. Any form of financial aid could easily trigger a collapse of the global economy.
Even if AI does not pose a direct threat to us as some fear, the failure of the $7 trillion project could lead to a global financial recession worse than the subprime mortgage crisis of 2007-2008.
– human knowledge
OpenAI claims that without the copyright exemption, they simply would not be able to build their current project, which would put a huge strain on artists, musicians, writers, and other creators.
– negative effects
OpenAI bears virtually none of the associated costs, including the spread of disinformation and misinformation, cybercrime, and most recently, the fake book scam, which has harmed at least three authors I know of in the past two months .
The most recent one comes from well-known music writer Ted Gioia, who yesterday mentioned a fake work named “Frank” Gioia:
It's bad enough that LLM makes this kind of thing easy; what's even more worrying is that companies like OpenAI don't seem to mind passing all the social costs on to society at all, just like those who have recklessly emitted harmful chemicals in the past Like a material factory.
Moreover, we must also face the fact that the so-called “alignment problem” (that is, the consistency of AI's goals and human interests) is far from being resolved.
The natural resources required will be astronomical
Specifically in terms of resources, according to the analysis of Sasha Luccioni, climate leader and researcher at Hugging Face, the impact of Altman's ambitions on the environment is certain.
“If this plan were to be implemented, the amount of natural resources required would be unimaginable,” she told VentureBeat. “Even using renewable energy, which is not a given, the amount of water required would be unimaginable.” The amount of rare minerals and rare minerals is also extremely huge.”
OpenAI unlikely to provide more transparency
Luccioni said that there is currently a lack of transparency in environmental impact in the AI field, and as Altman launches a new round of financing, this situation is unlikely to improve in the short term.
A comparison of Google's PaLM 1 (2022) and its successor PaLM 2 (May 2023) shows that the amount of information provided in the paper is significantly reduced. Among them, the first paper has enough information to allow outsiders to estimate energy consumption.
“Today, these companies no longer even disclose the time required to train the model or the number of chips used, and the information disclosure is almost zero.”
Altman is very busy: Vision Pro is second best, and we need to use AI to find aliens
This weekend, Sam Altman made all kinds of shocking comments on X, and netizens even suspected that his account had been hacked.
References:
https://venturebeat.com/ai/sam-altman-wants-up-to-7-trillion-for-ai-chips-the-natural-resources-required-would-be-mind-boggling/
https://garymarcus.substack.com/p/seven-reasons-why-the-world-should?r=8tdk6
https://twitter.com/rowancheung/status/1756041213566714020
https://www.zhihu.com/question/643734366
https://www.zhihu.com/question/643734366/answer/3392217135