Discover more from H+ Weekly
NVIDIA Rides the Generative AI Wave High - H+ Weekly - Issue #429
This week - UK and Germany invest in AI sovereignty; Meta launches own AI code-writing tool; Apptronik shows a new humanoid robot; growing muscles in a lab; Xiaomi presents a new robo-dog; and more!
Big tech companies, such as Microsoft, Alphabet or Meta, and the new wave of AI startups, are heavily investing in generative AI. All those companies need the computing power required to train their large machine learning models. A demand that Nvidia is uniquely positioned to fulfil and is massively benefiting from.
On Wednesday, August 23rd, Nvidia released their quarterly earnings report which showed $13.5 billion in revenue, up 101% over the prior year and over $2 billion more than the $11.2 billion Wall Street analysts had predicted. This news brought Nvidia stock on Wednesday beyond $500 per share, from $471 to $508, the highest price in the company’s history. The generative AI boom tripled Nvidia’s value this year, adding an extra $700 billion to the company’s valuation and making Nvidia the first chip firm to become a $1 trillion company.
“Data center compute revenue nearly tripled year on year, driven primarily by accelerating demand for cloud from cloud service providers and large consumer internet companies for our HGX platform, the engine of generative and large language models,” said Colette Kress, Nvidia’s executive vice president and CFO, in the post-earning reporting call.
The demand for Nvidia chips is massive. Nvidia may have over 300,000 orders for H100 - their ultra-high-end GPU. As Andrej Karpathy tweeted, “Who’s getting how many H100s and when is top gossip of the valley rn”. This high demand, however, will take some time to fulfil. Analysts estimate that the demand exceeds supply by at least 50%, which causes another GPU shortage (we have covered that in more detail in H+ Weekly Issue #427). A shortage from which Nvidia is making even more money. According to Tom’s Hardware, Nvidia is making a massive profit on H100 GPUs. The cost to build one such GPU is $3,320 and they sell for $25,000 to $30,000, according to the report.
The question is, how long the hype around generative AI will last and how long Nvidia will be able to ride this wave? We have seen that the adoption of ChatGPT or Bard is not as high as some might expect. Meanwhile, ChatGPT is losing traffic and Gartner placed generative AI at the peak of inflated expectations. There are also controversies around the copyrighted materials used in training large language models and it is still uncertain how the AI industry will be regulated.
Jensen Huang, founder and CEO of Nvidia, remains optimistic. “You’re seeing that data centers around the world are taking that capital spend and focusing it on the two most important trends of computing today: accelerated computing and generative AI,” Huang said. “And so I think this is not a near-term thing. This is a long-term industry transition, and we’re seeing these two platform shifts happening at the same time.”
From H+ Weekly
Becoming a paid subscriber now is the best way to support the newsletter.
If you enjoy and find value in what I write about, feel free to hit the like button and share your thoughts in the comments. Share the newsletter with someone who will enjoy it, too. That will help the newsletter grow and reach more people.
🦾 More than a human
Using a combination of brain-computer interfaces and machine learning, two teams of researchers were able to enable locked-in people to talk again and convey a limited range of facial expressions with a virtual avatar. All the patients need to do is think about what they want to say. Then a machine learning algorithm translates their brain waves into written words at a rate of 78 words per minute, using a 1,024-word vocabulary, with an error rate of 23%.
Older mouse brains rejuvenated by protein found in young blood
Researchers have found that a protein named platelet factor 4 (PF4), which is involved in wound healing, can improve learning and memory in ageing mice. Old mice that received doses of PF4 showed an improvement in the hippocampus and increased levels of molecules that promote synaptic plasticity. Aged mice injected with PF4 also did better than aged control mice in cognitive tests. These results may help diagnose Alzheimer’s disease earlier (lower levels of PF4 can be used as a biomarker) and could lead to new anti-ageing therapies targeting cognitive decline.
Brain stimulation helps people with Parkinson’s walk
Japanese researchers have discovered a way to improve a walking condition known as “Parkinsonian gait” by painlessly zapping patients’ brains every time they take a step. The therapy resulted in a significant increase in walking speed, step length, and gait symmetry for people in the treatment group, but not those who received the sham stimulation. Treated participants also reported less severe freezing of gait than at the start of the study.
🧠 Artificial Intelligence
AI-generated art cannot be copyrighted, rules a US Federal Judge
A federal judge upheld a finding from the U.S. Copyright Office that a piece of art created by AI is not open to copyright protection. In the decision, the judge wrote that copyright has never been granted to work that was “absent any guiding human hand,” adding that “human authorship is a bedrock requirement of copyright.”
UK to spend £100m in global race to produce AI chips
The UK government has announced a plan to spend £100m to secure AI hardware from companies such as Nvidia, AMD and Intel. Part of the plan involves obtaining 5000 GPUs from Nvidia. However, this might be too little, too late according to some experts. The UK accounts for just 0.5% of global semiconductor sales. Rishi Sunak’s government revealed plans in May to invest £1bn over 10 years in semiconductor research, design and production, a step dwarfed by the US’s $52bn (£41bn) Chips Act, and EU subsidies of €43bn (£37bn).
Germany doubles funds for AI ‘made in Europe’
The German government announced its intention to double the funds for investments in AI, increasing the available funds to nearly €1 billion. With the funds, Germany is looking to set up 150 new university labs dedicated to researching artificial intelligence, expand data centres, and increase access to datasets for training advanced AI models. The goal is to then convert the research and skills to “visible and measurable economic success and a concrete, noticeable benefit for society.” Additionally, the government says it hopes to show the unique selling point of AI “Made in Germany” (or “Made in Europe”).
Meta launches own AI code-writing tool: Code Llama
Looks like Meta is planning to win the generative AI game by appealing to the open-source community. A month ago, Meta released and open-sourced Llama 2, their large language model. Now, Meta released Code Llama, an open-source LLM trained to write code, positioned as a competitor to OpenAI’s Codex. Code Llama will use the same community license as Llama 2 and is free for research and commercial use.
Thinking about God makes people more likely to trust AI
According to new research, people who were actively thinking about God across eight experiments were less averse to AI and more willing to consider AI-based recommendations in a variety of contexts, including in the choice of movies, financial products, dental treatments, and romantic partners. The author of the research suggests that AI companies should take into account these results and consider experimenting with AI-based recommendation services with more religious populations first.
Meet Apptronik’s Apollo - a new robot joining the upcoming wave of humanoid robots. The robot is roughly human-size, standing 1.7 meters tall and weighing 73 kilograms, with a maximum payload of 25 kg. It can run for about 4 hours on a swappable battery. The company has two of these robots right now, and it is building four more. The company hopes its robot will find a place in logistics and manufacturing, but Apptronik promises “endless potential applications long term.” This article also contains an interview with Apptronik’s CEO and CTO, where they share more information about the custom actuators that can make Apollo more affordable. And by “affordable”, they mean $50,000 and less.
Xiaomi released their second-generation robotic dog, CyberDog 2. In the press release, Xiaomi said the robot is the size of a Doberman, weights 8.9 kg and is equipped with 19 sensors, including an RGB camera, an AI-powered interactive camera, 4 ToF sensors, a LiDAR sensor, a depth camera, an ultrasonic sensor, a fisheye lens sensor, a force sensor, and two Ultra Wideband (UWB) sensors. Xiaomi also promises to make CyberDog 2 open-source and allow third-party developers to add new functionalities. The price for this robo-dog is $1,789.
Fake Meat Is Bleeding, but It’s Not Dead Yet
After a couple of years of growth and hype, the meat alternative companies are hitting a wall. Beyond Meat told investors on the earnings call that net revenues had decreased 31% year-on-year. In total, sales of plant-based meat fell by 1% in the US in 2022, following a year of zero growth in 2021. However, the situation is different in Europe, where there is a demand for meat alternatives as Europeans are more open to making changes in their diet and are generally reducing how much meat they consume.
Building muscle in the lab
Researchers from ETH Zurich have found a way to turn connective tissue cells into muscle cells, using a cocktail of small molecules and proteins to “reprogram” these cells. This method is safer than other methods that use genetic engineering, which might introduce unwanted mutations. Researchers hope this method can help treat people suffering from muscle diseases or, alternatively, be used in producing lab-grown meat.
Biosensor bacteria would glow to flag bowel inflammation
Bioengineers have engineered a bacteria that acts as a biosensor detecting bowel inflammation. The idea is that a patient swallows the bacteria, which then traverse the entire gastrointestinal tract and are gathered at the opposite end. If a disease is detected, the bacteria will produce a fluorescent protein. So far, this approach has been tested in rats and has demonstrated its effectiveness. The researchers stated that these bacterial biosensors could be further developed to identify and track other gastrointestinal conditions.
H+ Weekly 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.
H+ Weekly is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.
A big thank you to my paid subscribers, to my Patrons: whmr, Florian, dux, Eric, Preppikoma and Andrew, and to everyone who bought me a coffee on Ko-Fi. Thank you for the support!
Thank you for reading and see you next Friday!