RTN: is Embodied AI turning the corner?
new Apple hardware, ARM pops, Nvidia spending spree
AI x Physics
AI is moving fast, with text and image-based use cases grabbing the media's attention. Less evident, is the incredible amount of progress that AI and high-performance computing are making in Physics, with multiple breakthroughs in the last few weeks alone.
This week, Google released a new CFD (computational fluid dynamics) framework for simulating turbulent flows using Tensor Processing Units (TPUs). This type of modeling is important for many applications like weather prediction, combustion engineering, aerodynamics etc. but is hard due to data dimensionality issues and the range of spatial and temporal scales involved.
Stanford also just released PhysObjects - a dataset containing human and ML annotations of physical properties like mass, fragility, and deformability for household objects, fine-tuned on a visual-language model (VLM). This provides a way for machines to quantify humans’ intuitive visual understanding of everyday objects.
Machines’ reasoning of the real-world is accelerating, two weeks ago an autonomous drone beat the world’s best drone racers. In August, researchers produced the largest CFD simulation model with 40bn cell resolution on an AMD cluster. This week, Wayve also used a VL(A)M model to improve learning and explainability of its autononomous driving model (though this is not strictly a physics breakthrough). These advances demonstrate AI's potential to tackle deeply rooted problems in machine perception, planning and reasoning of the physical world.
Robust physics simulations and intuitive physical understanding will transform robotics, generative design, simulation and digital twins. The downstream impacts span nearly every industry dealing with physical processes and products.
📣PSA: we’d love to expand the readership of this newsletter, please help us by forwarding this to a friend of colleague who might enjoy it
Top Stories
Nvidia is picking up where VCs left off (The Information, paywall)
Prefab homes startup Mighty Buildings raises $52m (TechCrunch)
European AI defense company, Helsing, raises $200m round (FT, paywall)
Rockwell Automation acquires AMR leader Clearpath Robotics (Robot Report)
Sustainability Trends Report 2023 (Generation Investment Mgmt)
Elon: Entrepreneurs are missing out on opportunities in physical industries (Fortune, paywall)
Argo AI founders start new AV trucking company (Robot Report)
We now have sand-sized semiconductors, to track Parmesan!? (Spectrum IEEE)
China’s SMIC 7nm chip, why US sanctions have failed (SemiAnalysis)
TSMC will start trial production in Arizona Fab in 2024, however chips will need to be sent to Taiwan for advanced packaging (Tom’s Hardware)
The US just discovered a lithium deposit which might be worth up to $1tr
Chinese battery manufacturers are expanding to Hungary (Rest of the World)
New database launched to search UK ‘Areas of Research Interest (ARI)’
Palantir goes all in on Defense, with Government Web Services:
Palantir: Visual history of US military-industrial complex Intel release new Thunderbolt 5 (USB-C), shares more on their Open Source strategy:
🦾 Manufacturing and Robotics
ARM stock pops, IPO window is slightly more open
ARM went public (again) on the NASDAQ last Thursday, the stock listed at $51 and jumped to $64.56 by the end of trading Friday, jumping 26%
Importantly, this positive price action shows *some* latent demand for technology IPOs, which is particularly positive given ARM’s anaemic growth. The company is currently priced at a 120x P/E multiple
Off the back of this, Instacart, has raised its IPO price target to a $10bn valuation
We previously covered (in RTN23) Softbank’s need for ARM to go public, given their margin loan secured against 75% of the stock. You can expect Softbank to be extending their loan capacity
Apple released new hardware, many incremental changes and a few big leaps
There are a number of notable updates to traverse a number of topics we care about here, namely semiconductors, climate and manufacturing and supply chain control
The big update (other than new iPhone) is Apple's new A17 SoC which has been built on TSMC 3nm (N3) fab process v 5nm (5N) process for the A16
The chip has a 6-core CPU and GPU configuration, with stated processing speed improvements of 10% and 20% respectively, while wafer prices went up 25% and SRAM scaling stopped
Apple also announced the new Apple Watch Series 9 will come with its S9 SiP chip, which can process ML workloads on-edge
They also announced (some combinations) of their Watch were carbon neutral and have shown their increasing ability to diversify manufacturing away from China. This iPhone is the first that will also be made in India
🌍 Policy and Geopolitics
The EU is upset that China is able to produce EVs cheaper than in Europe
In last week’s edition, we covered German automakers getting hosed by Chinese EV makers at the Munich Auto Show. Maybe Ursula reads this newsletter? 👋
Either way, the EU has seemingly decided that free markets don’t extend to China, whilst forgetting that China is also the EU’s third largest export partner, with some $200bn of annual exports
The EU has launched an ‘anti-subsidy’ probe into the increasing prevalence of Chinese EVs in Europe:
"Global markets are now flooded with cheaper electric cars. And their price is kept artificially low by huge state subsidies,"
