Sam Altman asks for $7 trillion to build chip - Weekly News Roundup - Issue #454
Plus: OpenAI Sora and AI agents; ChatGPT gets memory; Gemini 1.5; "meaty" rice; more humanoid robots; glowing plants go on pre-order; and more!
Welcome to Weekly News Roundup Issue #454. In this issue, we will take a closer look at Sam Altman’s idea to boost global semiconductor capacity, for which he is seeking a $7 trillion investment.
In other news, OpenAI has teased Sora, a text-to-video generator capable of producing high-quality video clips from a prompt, and has added memory to ChatGPT. Google has released Gemini 1.5 Pro, an update to their Gemini family of AI models. Additionally, more humanoid robots have showcased their capabilities. And over in biotech, researchers have created a hybrid food by combining rice with beef, and a glowing petunia is now available for pre-order in the US.
On February 8th, the Wall Street Journal reported that Sam Altman, the CEO of OpenAI, is in talks with various investors to raise $7 trillion (yes, trillion) for a new venture to reshape the world of chip-making and AI by opening a network of new semiconductor foundries to address the chip shortages. That number, $7 trillion, is larger than Microsoft’s market valuation (about $3 trillion). It is also larger than Japan’s or Germany’s GDP, the fourth and third economies by GDP in the world, respectively. If Altman is successful in this massive fundraising campaign, it would be the largest fundraising in history.
Let’s try to understand Altman’s plan by first understanding what the semiconductor industry looks like, what the implications can be and what criticism it attracted.
Today’s tech industry and its most advanced devices are built on the shoulders of just one company - TSMC. Every iPhone, iPad and Mac with Apple Silicon as well as Nvidia’s top GPUs running machine learning algorithms, AMD’s powerful and efficient CPUs and every Playstation 5, Xbox Series X and Series S is powered by chips made by TSMC. TSMC rose to become a chip-manufacturing giant worth $575 billion, making it the 13th company in the world by market cap.
However, the situation TSMC has found itself in causes problems for the tech industry. The company is a critical linchpin in the entire tech industry. They offer the most advanced chip manufacturing techniques and that makes the company the default option for those who seek top-class performance from their chips. The massive demand this generated means TSMC can't offer enough capacity to support customers, which will continue to next year, according to TSMC CEO C.C. Wei.
Depending on only one foundry providing the most advanced chips for the most in-demand and powerful products creates a fragile ecosystem with one massive bottleneck and weak point. If there is a production issue, the impact would cascade and would be felt by the entire tech industry. Another problem with TSMC is that the company is located in Taiwan (the letter T in TSMC stands for Taiwan). If something were to happen in Taiwan, either a natural disaster or a military conflict, it would take even years for the tech industry to recover.
TSMC is not the only chip foundry in the world. Other players include Samsung, Intel, Micron and Global Foundries, but these companies cannot compete with TSMC in terms of the quality and performance of the chips they produce.
The global semiconductor industry would benefit from decentralising and not having TSMC as the weakest link at the foundation of the entire tech industry. What Sam Altman proposes to do is address the supply side of chip manufacturing by significantly expanding global semiconductor capabilities. The required investment is high because spinning up a new semiconductor factory is not cheap. TSMC is investing $12-40 billion to build a semiconductor plant in Arizona. In 2021, Intel announced a $20 billion investment into two new fabs in Arizona.
Altman’s plan is ambitious. At the time of writing this article, 73 new semiconductor fabs are being built worldwide, 16 of which will be capable of producing chips with 10nm or smaller process technology. Assuming Altman’s plan would focus on the most advanced nodes and one such factory would require an investment of $50 billion (which is more than the new TSMC factory in Arizona) then with $7 trillion, we could build about 140 such factories if all that investment would go into building new semiconductor fabs. In reality, I think that number would be smaller but it would still be a big expansion in manufacturing capacity.
To secure funding for this massive project, Altman has been in conversations with a range of potential investors worldwide, including SoftBank CEO Masayoshi Son and the United Arab Emirates. The inclusion of the UAE may have an impact on geopolitics. The UAE had joint AI research projects with China. However, G42, a UAE-founded AI company, has recently sold all its China stakes to appease the US. The US government might be concerned with another player entering the semiconductor scene. The US sees semiconductors and AI as important technologies assuring its global dominance and wants to keep those technologies as close to the chest as possible, as we have seen with the bans on sales of GPUs to China. OpenAI, the darling of the current AI boom, promised to be a customer of these new foundries, which could raise some questions about US national security being tied to a foreign entity.
Interestingly, part of the $7 trillion plan involves funding power providers, too. Those massive GPU farms powering ChatGPT and other AI services require enormous amounts of power. Generating one image with an AI text-to-image generator can use as much power as charging a phone. Addressing the power consumption or the amount of power available could positively impact the AI industry. I’d like to point out that Sam Altman has also invested in Helion, a fusion power startup, which recently signed a deal with Microsoft for the tech giant to purchase electricity from Helion’s first fusion power plant, scheduled for deployment in 2028. Some of that $7 trillion might go to companies like Helion and other energy startups and companies.
Altman’s ambitious fundraising goal to expand AI capabilities by orders of magnitude has brought attention and criticism. As
points out in his Seven reasons why the world should say No to Sam Altman and What Could Possibly Go Wrong with Sam Altman’s New Ambitions? posts, there are serious questions about the impact of Altman’s plan on energy and climate as well as the sheer amount of resources we are talking about, including mining rare earth minerals needed to make all those chips. Critics argue that such a move could prematurely lock us into a specific type of AI technology. Instead, they suggest focusing on safety and exploring innovative new approaches rather than hastily scaling up current architectures that still have unresolved issues.In any case, Altman's $7 trillion fund shows how big AI has become. The question is whether that price tag represents the value of the potential impact AI can have on our lives, or if it reflects the hubris and the magnitude of the AI bubble we are currently experiencing.
If you enjoy this post, please click the ❤️ button or share it.
Do you like my work? Consider becoming a paying subscriber to support it
For those who prefer to make a one-off donation, you can 'buy me a coffee' via Ko-fi. Every coffee bought is a generous support towards the work put into this newsletter.
Your support, in any form, is deeply appreciated and goes a long way in keeping this newsletter alive and thriving.
🦾 More than a human
A new device let a man sense temperature with his prosthetic hand
Researchers in Italy and Switzerland have developed "MiniTouch," a device that enables individuals with prosthetic hands to sense temperature, bringing us a step closer to fully sensory prosthetic limbs. The current prototype features only one sensor, but the team plans to expand it to include more sensors for temperature and touch.
🧠 Artificial Intelligence
OpenAI is developing “AI agents.” Here’s what that means.
According to a report published by The Information, OpenAI is working on "AI agents" that can execute complex tasks. Trained by observing human interactions with computers, these agents could formulate a step-by-step plan for a complex task and then execute it. If successful, OpenAI could bring about another dramatic change in how we use computers, similar to the impact ChatGPT had in 2023. However, OpenAI faces challenges in ensuring the reliability of these agents and protecting users' private information.
OpenAI Sora: Creating video from text
OpenAI introduces Sora, a text-to-video generator, capable of generating high-quality videos up to a minute long from a text description. Sora is still an internal project but some visual artists, designers, and filmmakers were granted access to gain feedback on how to advance the model to be most helpful for creative professionals. OpenAI did not disclose when we can expect Sora to be publicly available.
OpenAI: Memory and new controls for ChatGPT
OpenAI is rolling out a memory feature for ChatGPT, enabling it to retain information across conversations to improve interaction without users needing to repeat details. This test phase targets a small group of free and Plus users, with plans for a broader rollout. OpenAI also plans to introduce memory capabilities to its GPTs.
Google introduces Gemini 1.5
Just a week after releasing Gemini Ultra to the public, Google is launching Gemini 1.5 Pro, which offers quality comparable to Gemini 1.0 Ultra while utilizing less compute power. Gemini 1.5 Pro features a standard 128,000-token context window, although some users are experimenting with a context window of up to 1 million tokens. Moreover, the new model boasts improved understanding across all modalities, including video. Demis Hassabis also revealed that Gemini 1.5 Pro is a Mixture-of-Experts type of model. Currently, Gemini 1.5 Pro is available in a limited preview for developers and enterprise customers, with the date for a broader rollout yet to be announced.
Judge rejects most ChatGPT copyright claims from book authors
A California judge largely sided with OpenAI in a copyright infringement case, dismissing most claims against its ChatGPT model but allowing a direct copyright infringement claim to proceed. The court found insufficient evidence for most allegations but permitted further action on the use of copyrighted works without permission under California law.
Nvidia replaces Alphabet as Wall St's third most valuable company
With a valuation of $1.825 trillion, Nvidia has passed Alphabet to become the third most valuable company on Wall Street. Nvidia is riding the AI wave high as its GPUs are highly sought-after by many companies to train and run their AI models.
OpenAI researcher Andrej Karpathy departs firm
Andrej Karpathy, an artificial intelligence researcher and one of the founding members of OpenAI, has left OpenAI (again). In his post on X, Karpathy said there was no drama involved and that he will now focus on his own projects and “see what happens”. Maybe one of those projects is his idea of an operating system powered by large language models.
GOODY-2 - The world's most responsible AI model
Meet GOODY-2, the world’s safest and most responsible AI model. It is so safe it will politely refuse to answer any question that is outside its ethical principles’ bounds (which is every question). GOODY-2 may not ace every benchmark but it scores 99.8% on the PRUDE-QA test.
If you're enjoying the insights and perspectives shared in the Humanity Redefined newsletter, why not spread the word?
🤖 Robotics
Everything You Wanted to Know About 1X’s Latest Video
1X, a Norwegian company developing humanoid robots, posted a video showing their wheeled humanoid robots doing a bunch of random stuff. IEEE Spectrum sat down with Eric Jang, Vice President of Artificial Intelligence at 1X, to discuss these robots and to ensure everyone those are real robots, not humans dressed like robots.
2023 robot orders down 30% from 2022 in North America, according to A3
After experiencing two years of record sales, North American robot orders saw a 30% decline in 2023, according to the Association for Advancing Automation (A3). This downturn, especially visible in automotive manufacturing, is expected to persist until the latter half of 2024. A3 attributes this slowdown to several factors, including a sluggish US economy, increased interest rates, and an over-purchase of robots in 2022 due to supply chain concerns. However, A3 is optimistic about future growth, particularly in non-automotive industries, as businesses begin to recognize the advantages of using robots.
New robot guide dog shows not only human jobs are threatened by AI
A team from the University of Glasgow presents Roboguide, a robot guide dog. Utilizing the Unitree Go1 robot dog as a base, the team developed the prototype to assist blind and partially sighted individuals in indoor spaces. Thanks to its suite of sensors, Roboguide is aware of its surroundings, and with its integrated large language model, the robot can understand human commands and respond verbally.
MagicLab's humanoid can toast marshmallows, fold clothes and dance
Another humanoid robot enters the scene. In this video, MagicBot from MagicLab shows what it is capable of by roasting a marshmallow, folding baby clothes, watering a plant, dancing (every humanoid robot these days needs to show off their dancing skills) and doing magic tricks.
🧬 Biotechnology
Scientists grow 'meaty' rice hybrid food for protein kick
Scientists from South Korea have created a new type of hybrid food – "meaty" rice that they say could offer an affordable and eco-friendly source of protein. The process involved coating the rice in fish gelatin to help the beef cells adhere and then leaving the grains in a petri dish to culture for up to 11 days. According to the team at Yonsei University in South Korea, this rice contains 8% more protein and 7% more fat and is a bit firmer and more brittle than regular rice. Moreover, compared to regular beef, it has a smaller carbon footprint since the production method eliminates the need to raise and farm large numbers of animals.
Bioluminescent houseplant hits US market for first time
The US company Light Bio is beginning to accept pre-orders for their genetically engineered plants that glow. This bioengineered petunia (Petunia hybrida) looks like a normal plant during the day but glows a faint green in the dark. The company plans to ship a batch of 50,000 firefly petunias in April. I’d love to have one but I have to wait until these kinds of plants are available in the UK… or I have to make one myself.
Thanks for reading. If you enjoyed this post, please click the ❤️ button or share it.
Humanity Redefined sheds light on the bleeding edge of technology and how advancements in AI, robotics, and biotech can usher in abundance, expand humanity's horizons, and redefine what it means to be human.
A big thank you to my paid subscribers, to my Patrons: whmr, Florian, dux, Eric, Preppikoma and Andrew, and to everyone who supports my work on Ko-Fi. Thank you for the support!