Headlines this week - Dec 29, 2024
A look at how capital is being deployed across future opportunities
This week in the future:
US sanctions are not having the expected impact on China’s progress with advanced AI models. DeepSeek, a Chinese startup, just open sourced a model which performs almost at the same level as the most advanced OpenAI / Anthropic ones
US policymakers are now considering to expand restrictions to less advanced chips (that seem to be useful for innovative Chinese developers to create these powerful AI models)
The fossil fuel industry is under pressure from the boom of electric car sales in China (expected to surpass combustion engine sales next year). As demand for transport fuel falls, the industry’s revenue mix is shifting to petrochemical products, but here China may also have a big effect, due to the country’s excess refining capacity
Consensus grows about a shift in the approach to increase models performance, from “scaling up” of “raw capacity” to “reasoning”. OpenAI’s newest model (o3) vindicates this shift, with a very significant performance leap vs. previous models (justified by the new approach)
AI safety researchers at Anthropic confirm that advanced models show resistance against “corrections”. The company’s model, Claude, tried to hide its misalignments vs. the expected values when researchers tried to train away these misalignments
Land availability emerges as a potential bottleneck for AI development. Data center companies look to develop in appropriate areas, but this sometimes could lead to crowd out other real estate needs such as housing and retail. The US city of Atlanta might be an early example of this
Quantum Computing experts advise companies and governments to change their encryption algorithms to “quantum-safe” ones (i.e. not breakable by a Quantum Computer using the Shor’s Algorithm). Bitcoin is also included (as an attack could cause a tsunami of economic losses)
Change expected at NASA during the Trump administration. More commercial partnerships are expected, including with SpaceX and Blue Origin, and the priority is likely to be to launch manned missions to the Moon and Mars
AI military applications continue to be a priority. Palantir and Anduril, two tech / AI companies targeting Pentagon deals, are now looking to partner with SpaceX and OpenAI to increase their scope, amid increasing demand for AI weapons and defense systems
Meta planning an “augmented reality” version of their Ray-Ban smart glasses. The company is planning to add an internal display that will show notifications (including interactions with the MetaAI chatbot)
I hope you’ll enjoy it. Please share with your friends if you think this is interesting. And give me feedback if you’re missing something :-)
1. Population & natural resources
Biotech
Capital markets are not working efficiently for Biotech. Larger, more mature startups in “hot” themes get most of the funds, and this works against innovation in the industry. Falling interest rates could help attract more funds and contribute to solve the problem (but not completely). Why bigger is not always better in biotech
Space
A milestone is reached in solar research. A solar probe launched in 2018 is about to enter the solar atmosphere. This mission is significant because it requires advanced materials to withstand extreme heat and radiation, and will provide critical insights into the solar system's fundamental properties. A Spacecraft Is About to Fly Into the Sun’s Atmosphere for the First Time
The Trump administration plans to make NASA more efficient (with a little help from SpaceX). The team is considering consolidating NASA centers and accelerating the Artemis lunar program. All this includes more room for commercial partnerships, e.g. for manned missions to the Moon and Mars., which could benefit SpaceX and other firms. How might NASA change under Trump? Here’s what is being discussed
Engineers are leaving Boeing. The average tenure of Boeing engineers has decreased, as they seek opportunities at companies like Blue Origin and SpaceX. Let’s see if the new CEO can contain the leak. Boeing battles brain drain as engineers chase the allure of space
Tech & Geopolitics
Chinese AI startup DeepSeek just published a new model that performs at the level of the most advanced in the West. DeepSeek’s developers have used a mixture-of-experts (MoE) architecture to efficiently handle tasks, which helps them overcome the challenges from lack of access to advanced chips. The model is open-source. DeepSeek-V3, ultra-large open-source AI, outperforms Llama and Qwen on launch
Sanctions are not preventing China from making rapid advances in AI, Chinese AI startups are making significant progress despite U.S. restrictions on access to advanced chips. This is being achieved through several innovations in the way models are developed. Don’t Look Now, but China’s AI Is Catching Up Fast
US Commerce Secretary Gina Raimondo supports this view. She has actually said that export controls are not enough to hold back China's technological progress. Raimondo Says Holding Back China in Chips Race Is a ‘Fool’s Errand’
And investors too… The recent stock rally for SMIC (the leading Chinese semiconductor company) seems linked to bright perspectives in AI. This would only make sense as long as SMIC evolved to build more advanced chips or (alternatively) if Chinese AI developers found ways to make progress with less advanced processors. China’s SMIC Extends Rally on AI Enthusiasm, Accelerated Chip Localization
Restricting access to high-end chips only may not be enough for the US to maintain resilience in the chip supply chain. China’s semiconductor companies are specializing in “legacy” chips, capturing market share and hurting American competitors, which are becoming increasingly weaker. America’s Chip War With China Will Spread
There are signs that Washington is starting to shift attention to more mature chips. The US government just launched an investigation of China’s alleged anti-competitive practices in the semiconductor industry, targeting both foundational and mature chips. US launches probe into Chinese semiconductor industry
2. Efficiency / Productivity
Energy
Nuclear
Russia is exporting its nuclear energy know-how to other countries. They are currently building 10 plants in countries including Bangladesh, China, Egypt, India, Iran and Turkey. Russia aims to be global leader in nuclear power plant construction
Renewables
Uncertainty is growing for the industry. Biden’s rush to issue loans to renewable energy companies (before he leaves) highlights concerns about a change of policy in the US. Some analysts are dismissing the issue, given the competition with China on renewable technologies, but everyone agrees that uncertainty is growing. Joe Biden rushes to issue cleantech loans in bid to secure legacy
A new material for solar cells emerges (perovskite vs. silicon). A Japanese firm just revealed its plans to mass-produce perovskite solar films as part of a Japanese government-subsidized $2 billion project. Perovskite-based solar panels have advantages vs. current silicon ones, and they will also reduce dependence on current suppliers. Sekisui Chemical to Mass-Produce Solar Films in $2 Billion Project
Fossil Fuels
The fossil-fuel industry is preparing for the (long?) future, after all. Companies like Italy’s Eni are committed to continue exploration and have plans use the most advanced computing tools to find new reservoirs. Eni fires up €100mn supercomputer in race to find oil and gas reservoirs
But a shift to electric cars (e.g. in China) could have a massive (negative) impact. China's increasing adoption of electric vehicles (see section below) is already causing a decline in demand for gasoline and diesel. This shift is leading to a reduction in the proportion of oil demand accounted for by these transportation fuels, from 51% in 2018 to 44% in 2024. China’s EV Boom Is Starting to Pinch Oil Producers
The future of the industry might rely on plastic. As demand for transport fuels decreases, oil and gas companies might save the day with sales of petrochemical products (e.g. plastics). The share of this segment is already increasing. However, US companies could also suffer price pressures from China, where there is massive refining capacity. Driving an EV? Big Oil Hopes You Don’t Cut Down on Plastic Too
New Transport Technologies
Electric Vehicles
An electric car boom in China. As already mentioned above, EV sales in China will surpass sales of internal combustion engine cars in 2025. Sales are predicted to grow by 20% to over 12m cars in 2025, while sales of traditionally powered cars will fall by more than 10% to under 11m. China’s EV sales set to overtake traditional cars years ahead of west
Supersonic Airplanes
Quiet supersonic jets are getting closer. NASA’s project to develop one such aircraft (the X-59) just reached a milestone, as the first maximum afterburner engine test was completed. These innovative jets look to address a critical barrier for supersonic airplanes adoption (i.e. noise that limits their feasible routes). NASA’s X-59 Quiet Supersonic Jet Reaches a Major Milestone
Artificial Intelligence
Consensus grows about a shift in the approach to increase models performance. In his “predictions for 2025” analysis, R. Waters from the FT agrees that approaches based on more “reasoning”are promising. He also mentions agent-like capabilities, integration with current technologies (i.e. AI as a “co-pilot” more than a substitute), and a sustainable dominance by Nvidia of the chip space (linked to the company’s “middleware moat”) as key trends for the coming year. Four AI predictions for 2025
Debates continue about the best path to adoption. Gillian Tett (also from the FT) also talks about agents, and argues that ensuring that humans control AI agents, not the other way around, should be a key priority in the coming deployments. She also says that deployments should probably adapt to local cultures (with different perception of the relation of humans with non-human objects and technologies). We need to think again about what the ‘A’ in AI signifies
AI: Apps
Meta expects AI characters to fill social networks. In what could be seen as a regulator’s nightmare, Meta is developing a product that will make it possible for (human) users to easily create AI-based characters for social networks. There are already some (rather extreme) examples of this, like SocialAI, a “personalized” social network where an individual interacts exclusively with AI agents (but the experience is really poor…) Meta envisages social media filled with AI-generated users
Investors expect people to adopt chatbots as “personal teachers”. At least that’s what we can conclude from the bearish outlook currently affecting online education companies, previously attractive for investors. GenerativeAI is being used to develop much more powerful alternatives. Investment in online education groups plummets following rise of AI
Military applications continue to be a priority. In the current geo-strategic context, the US is trying to accelerate. Palantir and Anduril, two tech / AI companies targeting Pentagon deals, are now looking to partner with SpaceX and OpenAI to better capture the opportunity and/or increase their (defense-related) “addressable market”. Palantir and Anduril join forces with tech groups to bid for Pentagon contracts
AI: Robots
Will robots (ever) substitute scientists? For now, this American startup is trying to automate some lab-technician tasks. The approach, as with many other robot initiatives under way, is using LLMs to power the robots. Tetsuwan Scientific is making robotic AI scientists that can run experiments on their own
AI: Foundational Models
Investors believe Google might be catching up with OpenAI. Google has just announced several developments in AI, including Gemini 2.0 (its most powerful LLM for now), a new AI chip (Trillium), and a new AI agent prototype (Project Mariner). All these seem to have contributed to the company’s stock recently reaching a record high. Google’s push to re-establish lead in AI boosts investor confidence
OpenAI’s newest model (o3) shows a big performance leap vs. previous models. It scores 75.7% on the ARC-AGI benchmark under standard compute conditions, with a high-compute version reaching 87.5% (previous models scored 32% or less in the same tests). According to the company, the model shows a qualitative shift in AI capabilities, by adapting to tasks it has never encountered before. The results also support the approach based on “reasoning” (o3 uses chain-of-thought (CoT) reasoning combined with a search mechanism and a reward model that evaluates and refines solutions). OpenAI’s o3 shows remarkable progress on ARC-AGI, sparking debate on AI reasoning
AI: Security & Safety
Recent research with Anthropic’s Claude confirms that advanced AI models resist to be “corrected”. More specifically, in internal work within Anthropic, the model tried to hide its misalignments vs. the expected values when researchers tried to train away these misalignments. The takeaway is that AI models may not be easily corrected or realigned once they develop a complex set of goals. This raises concerns about the effectiveness of current alignment strategies and highlights the importance of developing new methods that can handle the AI's potential resistance to goal changes. Why Worry About Incorrigible Claude?
OpenAI claims it has used a new safety paradigm in its new o3 model. “Deliberative alignment”, as the company calls it, is based on making the model “think” in the programmed safety policy during the inference phase (a critical one for the performance of these new models). The claim is that this new method decreases the rate at which the models answer “unsafe” questions while improving their ability to answer benign ones. OpenAI trained o1 and o3 to 'think' about its safety policy
AI: Infrastructure
Land availability as a bottleneck for AI development? The explosive growth of data centers in the Atlanta metropolitan area (+76% power capacity in just 1 year), driven by AI and linked to incentives from local governments, is starting to lead to a competition for land, with data centers starting to crowd out other real estate needs such as housing and retail. The Data-Center Boom Eats Up a Lot of Land. Atlanta Says It’s Gone Too Far.
In China, Apple’s supplier Foxconn is diversifying into data centers. They are partnering with Zettabyte, a specialist in software solutions to make data center operations more energy efficient. Foxconn Invests in AI Data Center Firm Zettabyte to Boost Sustainable Computing
AI: Chips
SoftBank’s Masa Son will invest $100bn in the US, and chip production is a key priority for him. He aims to have the first batch of shipment-ready AI chips by 2026, with a prototype as early as the summer of next year. This will be a key part of his strategy to become a serious contender in the AI hardware market. He also wants to leverage SoftBank’s investment in Arm (the global leader in chip design frameworks). Masayoshi Son’s Quest to Become the Next Nvidia
Quantum Computing
Bitcoin needs to prepare for a Quantum Computing future. The first Quantum Computers may still be 5-6 years away (at the very least), and their applications will probably be much more limited than what some people are saying (and selling :-) ). But breaking common cryptography codes (via Shor’s Algorithm) is likely to be one of them. So many (actual) experts are recommending companies and governments to start working in quantum-safe cryptographic keys. Bitcoin (and blockchain in general) is not an exception. A Looming Threat to Bitcoin: The Risk of a Quantum Hack
Intelligence Augmentation
Augmented Reality
Meta planning an “augmented reality” version of their Ray-Ban smart glasses. The company is planning to add an internal display that will show notifications (including interactions with the MetaAI chatbot). The new product is currently planned for the second half of 2025. Meta to add display to Ray-Bans as race over smart glasses intensifies
3. Economic Trends
Could AI drive a significant increase in productivity? As we know from the last 3-4 decades, where digital technology was supposed to have a massive effect, this is not clear at all. However, recent data from the US look promising, and some economists are expecting investment in AI to lead to a +3% productivity growth in the coming years. The U.S. Needs a Productivity Miracle. It Might Just Get One.
Donald Trump is expected to bring major changes for the global order. However, no one really knows what to expect. This FT article discusses several scenarios, from the least optimistic (for the US) to the most positive one. Europe is not looking too good in any of them… How America First will transform the world in 2025