Headlines this week - Dec 15, 2024
A look at how capital is being deployed across future opportunities
This week in the future:
The debate on the future of AI models goes on: most (but not all) people believe that what has been the dominant approach until now (“raw scaling” of ever-larger models, using an increasing amount of computing power) is now reaching a dead-end, and will be replaced by (a) smarter, more focused systems that achieve better results with fewer resources, and/or (b) systems with “longer thinking” before giving an answer
SpaceX just reached a $350bn valuation, according to a deal that several investors have done with the company’s employees, to buy their shares
Many question marks on the recent decision to fire Intel’s CEO, apparently linked to the board’s lack of patience, and also to their concerns about the CapEx required to develop the (strategic) Foundry business (which could now be on the way to be sold to an external party)
GM is abandoning its robotaxi initiatives, and some people have started to question the potential profitability of this business, suggesting that it would be more valuable to directly sell autonomous cars to consumers
Google launched Gemini 2.0, their new “flagship” AI model, with strong capabilities to support AI agents. The company also presented several prototypes of agents, confirming the industry momentum around them
Nuclear energy is emerging as a strong “renewable” option, while “traditional” alternatives (like offshore wind) are starting to struggle. Fusion remains the “holy grail” for the field, and startups (like Fuse) are being created with the aspiration to be “the SpaceX of nuclear fusion”
Armies and police forces could be among early adopters of robots. AI-controlled weapons are becoming a hot space, with Anduril (a $14bn US-based startup) at the front of it (for now)
“Mirror biology” (based on molecules that mirror the ones that build current living systems) is emerging as an existential threat for humans (an life on Earth)
New “altermagnetic” materials could make data storage cheaper, faster and more sustainable
Zuckerberg’s vision about the Metaverse looks fully vindicated, with Google (with Samsung) and several startups now working to develop smart glasses. The commercial success of the Meta Ray-Bans has clearly helped with this
Population & Natural Resources
Biotech
The price of biotech progress: risk
Is aversion to risk decelerating progress? This article discusses the trade-offs between the positive impact of biotech research (e.g. gain-of-function analysis) and the potentially catastrophic risks involved. Public pressure against these risks might be discouraging some researchers to work in this field. The Long, Contentious Battle to Regulate Gain-of-Function Work
A new bio-threat was announced this week. A group of scientists published a paper in Science Magazine, warning about the (huge) risks from developing “mirror biology” (molecules that are the physical mirror image of life’s building blocks). Not everyone agrees, but it looks scary. Synthetic ‘mirror’ microbes pose unprecedented threat to life, scientists warn
AI as a catalyst
Money keeps flowing to initiatives to use AI to make progress in medicine. Dimension Capital, a VC targeting life sciences startups using digital technologies, just raised $500m. VC Firm Dimension Raises $500 Million to Bet on AI and Medicine
This field is a key priority for Microsoft. At least tha t’s what the company’s head of AI says… Microsoft’s Mustafa Suleyman hires ex-DeepMind staff for AI health unit
Riding the Ozempic wave
Good news for investors in the Ozempic and other weight-loss drugs. Ozempic, Wegovy and Mounjaro seem to have more practical applications, beyond weight-loss, including heart disease and Alzheimer’s. So their business perspectives are becoming even better. What other conditions could weight-loss drugs treat?
Foodtech
We need alternative ways to produce cheap food. Current ultra-processed foods look increasingly unsustainable. This week a paper linked them to colorectal cancer. So we need different ways to solve the problem of feeding the whole world’s population. Scientists find potential link between ultra-processed foods and cancer
Space
It’s official: SpaceX is worth $350bn. A group of investors (including the company itself) have agreed to buy stock currently owned by employees at $185/share, equivalent to a $350bn capitalization. SpaceX’s valuation soars to $350bn in employee share deal
Concerns on sustainability issues for satellites is growing. As an example, the environmental impact of the process now being used to dispose of “space junk” (satellites after their useful life) is still unknown. The world’s next big environmental problem could come from space
Materials
Altermagnetic materials, recently discovered, could revolutionize data storage. Altermagnetic materials can sustain magnetic activity without being magnetic themselves. So they could be used to build cheaper, faster and more sustainable memory components. New magnetic flow has potential to revolutionise electronic devices
Tech & Geopolitics
Nvidia exposed to higher risks in China. On top of increasing restrictions in the US to export products to China, now China is also launching an antitrust investigation against the company. So the future of their business in the country is becoming very uncertain. China launches antitrust probe into Nvidia
US commercial restrictions for chip technologies may not be so useful. An example this week came with news that Russian armed forces have been able to buy chips from Intel and others through a very simple process, involving intermediaries. Russia’s Military Found a Surprisingly Simple Way to Buy US Chips
The EU is softening its attitude vs. Trump with respect to technology. We clearly have much more to lose than the Americans, so it is probably a good idea to swallow some of that pride… EU tech chief strikes conciliatory tone with Elon Musk
Efficiency / Productivity
Energy
Nuclear
Fuse: a startup in the race to develop commercially feasible nuclear fusion technology. Fuse's ambition to be the "SpaceX of fusion," and they are aiming to collaborate with various partners to refine the same technology that has already shown promising results in the lab. Inside a Fusion Startup's Insane, Top-Secret Opening Ceremony
Renewables
BP and Shell reduce their exposure to renewable energies. Amid signs that US regulators will now be more open to the use of fossil fuels, BP and Shell are starting to abandon their renewable power generation projects, previously expected to reach the same revenues as oil and gas by the 2030s. BP and Shell rein in electricity ambitions to escape ‘valley of death’
This move may be not enough, at least for BP. BP, which has put its offshore wind assets into a jointly-owned business with a Japanese power generation company, still has a huge debt load, and investors are worried about this. BP has only scratched the surface in solving its identity crisis
Offshore wind power generation is decelerating. At least that is what the biggest vendor of offshore wind equipment in the West (GE Vernova) anticipates. GE Vernova Expects More Trouble for Struggling Offshore Wind Industry
Fossil Fuels
Fossil fuel companies also want to capture the data center opportunity. Exxon just announced they’re building a gas power plant specifically designed to feed data centers. It will include carbon capture technologies. Exxon Plans to Sell Electricity to Data Centers
New Transport Technologies
Electric Vehicles
After the fiasco of a 100% European battery startup (Northvolt), European car vendors start to partner with Chinese battery giants. A new $4.3bn battery factory will be built in Spain by Europe’s Stellantis, in partnership with China’s CATL. China’s CATL to build $4.3bn battery factory with Stellantis in European expansion
Autonomous Cars
General Motors closes its robotaxi program. This comes after an investment of $10bn in the last 10 years. The decision is justified by long times and high costs to scale the business, in an increasingly competitive field. General Motors Scraps Cruise Robotaxi Program
Part of the problem is their lack of credibility with investors (for this). GM’s investors tend not to believe the company can successfully compete with Waymo or Tesla, and this is making it difficult for them to raise the required capital. GM’s robotaxi dream has been prematurely clamped
The announcement has opened a debate on the robotaxi business model. Robotaxis are perceived as a loss-making business, and it could be easier to reach profitability by selling autonomous cars to customers (a project that GM has not abandoned). The end of Cruise is the beginning of a risky new phase for autonomous vehicles
Startups keep emerging, to challenge Waymo and Tesla. E.g.: Wayve, which is using a “fully AI-centric” approach to autonomous driving, in contrast with Waymo’s hybrid approach, which combines AI learning with hand-coded instructions. Wayve’s AI Self-Driving System Is Here to Drive Like a Human and Take On Waymo and Tesla
3D Printing
The guy who killed UnitedHealthcare’s CEO used a 3D-printed gun. A really sinister way for 3D printing to start its renaissance (?) The ‘Ghost Gun’ Linked to Luigi Mangione Shows Just How Far 3D-Printed Weapons Have Come
Artificial Intelligence
AI: Agents
AI Agents will be the next big thing in the AI field during 2025. Announcements and discussions about them proliferate every week:
For this FT columnist, they could help accelerate AI adoption. This seems reasonable, as agents could require less adaptation of how people work The AI agents are coming
Google has this as a key priority. Making agents work better is a significant part of what the company’s new LLM (Gemini 2.0) brings to the table. Google races to bring AI-powered ‘agents’ to consumers
In Google’s vision, agents are expected to autonomously manage people’s computing interfaces:
The file system (i.e. Google Drive): Gemini AI can now summarize what’s in your Google Drive folders
The web (i.e. Google Chrome): Google unveils Project Mariner: AI agents to use the web for you
Google also showed an updated version of a more advanced prototype (Astra): This agent will be able to “see the world as a person”. It is described as an “all-seeing”, “all-hearing” and “all-remembering” tool, so expect some interesting privacy debates around this. How Google's Project Astra gives an AI agent eyes
Amazon joins the club. The company’s new R&D lab in San Francisco (Amazon AGI SF Lab) will focus on building “foundational” capabilities for AI agents. Amazon forms an AI agent-focused lab led by Adept's co-founder
AI: Apps
Speak, a startup using GenAI to teach foreign languages is now a unicorn. OpenAI is an investor. OpenAI-backed Speak raises $78M at $1B valuation to help users learn languages by talking out loud
Apple keeps deploying “Apple Intelligence”. New features coming with iOS 18.2 (in the appropriate countries :-) ) iOS 18.2 is here with Apple Intelligence image generation features in tow
AI can help people recover their speech (in a way). A model can be trained with clips of a person’s voice so that it can talk in (exactly) the same way. This can help people recovering their speech after losing it (but only if they recorded themselves before that…) AI can bring back a person’s own voice
Commercial AI-generated films are coming. TCL, the world's largest TV manufacturer, just demoed some AI-generated films in Hollywood. For now, they look technically limited and production requires human support. But this is expected to change. Of course, there are plenty of (business and ethical) concerns about a future when these products will be good enough. I Went to the Premiere of the First Commercially Streaming AI-Generated Movies
AI: Robots
Armies and police forces are expected to be robot early adopters:
Anduril is working on products that fully automate military actions. The company, founded by Palmer Luckey (previously founder of Oculus, later acquired by Facebook), already has a $14bn valuation (that is expected to grow fast). We saw a demo of the new AI system powering Anduril’s vision for war
Europe is entering the game, too. Rheinmetall, a German arms producer, is partnering with Auterion, a Swiss-American technology startup, to develop software for military drones. They see the war in Ukraine as a possible commercial target. Rheinmetall, Auterion to Develop Standardized Software for Military Drones
Police in China is already patrolling with robots. As the picture in this article (that looks as a sci-fi scene) shows. Police in China take hi-tech rolling robot out on patrol
AI: Foundational Models
Google presented Gemini 2.0, its latest LLM, with “multimodality” as a key feature. Gemini 2.0, Google's newest flagship AI, can generate text, images, and speech
Powering AI agents is now a key mission for Gemini. As we already discussed above, when talking about Google’s AI agent initiatives. Google Rolls Out Faster Gemini AI Model to Power Agents
Meanwhile, people keep talking about “long thinking” as the new frontier for models, with raw scaling increasingly seen as a decelerating approach. Get Ready for ‘Long Thinking,’ AI’s Next Leap Forward - WSJ
Linked to these views, people are also paying attention to smaller models. Microsoft just presented Phi-4, a model that outperforms larger ("LLM”) models in mathematical reasoning, despite having significantly fewer parameters (14bn vs. hundreds of billions or potentially trillions). This is one more sign that the future may not lie in ever-larger models, but in designing smarter, more focused systems that achieve better results with fewer resources. Microsoft’s smaller AI model beats the big guys: Meet Phi-4, the efficiency king
Microsoft’s head of AI has (naturally) a neutral approach. In this long interview with The Verge, M Suleyman anticipates that larger models will continue to unlock "seismic gains" similar to those observed in previous generations, albeit at a slower pace due to increasing costs and complexity. And he also also acknowledges the value of smaller, more specialized models, particularly for consumer-facing products. Microsoft AI CEO Mustafa Suleyman on what the industry is getting wrong about AGI
It is becoming increasingly difficult to find new data to train new models (after already having used more or less the whole of the “open internet” to train the existing ones):
Reddit is becoming an attractive partner / target for AI companies, because Reddit posts are an attractive source of new data. How Years of Reddit Posts Have Made the Company an AI Darling
Public-domain books are being incorporated, after a partnership of Google with Harvard University. Harvard and Google to release 1 million public-domain books as AI training dataset
In China, ByteDance consolidates its AI strength, and is on the way to lead Generative AI too. TikTok-owner ByteDance takes lead in race to capitalise on AI in China
AI: Security / Safety
At AI companies, specific teams are being built to check how harmful can the models be. As an example, Anthropic’s “Frontier Red Team” is in charge of anticipating (and neutralizing) potentially harmful uses of AI models by “bad actors”. The AI Researchers Pushing Computers to Launch Nightmare Scenarios
AI could drastically amplify the “toxic content” problem. New challenges for online safety are emerging… Chatbots urged teen to self-harm, suggested murdering parents, lawsuit says
Amid all this negativity, some voices have a positive vision of AI as a tool for human “superagency”. According to Reid Hoffman (the LinkedIn founder), AI systems have the potential to actually enhance human agency, providing new “superpowers” that humans could take advantage of. AI is the frenemy of freedom
Policymakers need to find the right balance between controlling the bad and enabling the good. In the US, Trump seems to be inclining to a more “laissez faire” approach (in contrast, for example, with the UK). UK’s ambitions to police AI face Trump’s ‘starkly’ different approach
AI: Infrastructure
Silently, Dell has transitioned into an “AI Infrastructure” company. The company is now focused on AI servers (computing and storage), rather than personal computers. Michael Dell Spent 40 Years Preparing for an AI Boom No One Expected
Scarcity of physical sites with good access to the power grid could become a limitation for AI development, at least in the US. This also highlights how important energy (and access to it) is becoming for the emerging AI industry. US aim to lead on AI threatened by land shortage
AI: Chips
Nvidia under attack: regulators in China, and new competitors at home. Apart from the already mentioned regulatory investigation in China, Nvidia is also under the threat of vertical integration initiatives by some of its customers (e.g.: Google, more advanced, or Amazon, just announced) and also by emerging startups. Nvidia’s profit smorgasbord attracts an ant invasion
Broadcom is emerging as one of Nvidia’s rivals. Their valuation is soaring (just reached $1trn), and they have exciting new initiatives (including a partnership with Apple, to develop AI chips) Broadcom soars to $1tn as chipmaker projects ‘massive’ AI growth
Marvell, a smaller company, is also emerging as an AI-chip player. At more than $100bn, they have now become more valuable than Intel (!) They’re working with Amazon to design new AI chips (like the Trainium processor just announced) Meet the Small AI Chip Maker Now More Valuable Than Intel
Meanwhile, discussions continue about the fall (for now) of Intel:
Industry expert D. Patel calls the BoD “incompetent”, and claims that Gelsinger shouldn’t have been fired. According to his views, the decision might have been driven by “panic” at the high CapEx required to build / develop the “Foundry” business. But this business seems critical for the future. Intel on the Brink of Death
Ben Thompson agrees, and argues that Intel should create an independent “Foundry” unit, which could potentially receive substantial government subsidies, given the current geo-strategic context. About the BoD, he also says that it has been a “long-running disaster” Intel’s Death and Potential Revival
Rumors that the chip manufacturing unit could be sold seem to support the idea that Gelsinger was fired for opposing to this. Intel executives say a manufacturing spinoff is possible
Quantum Computing
Google made a big Quantum Computing announcement this week, and after the initial euphoria, reactions have cooled down a bit:
It was initially described as a “breakthrough”. Even if the good news were that it could make commercial quantum computers possible “within this decade”. Scientific breakthrough gives new hope to building quantum computers
The punch line was that the prototype is capable of tasks that a traditional supercomputer could not master in 10^24 years. However, the key point is that the value of this depends on what these “tasks” are… Google Makes New Quantum Computing Breakthrough
But professional physicists were more skeptical. On X, Sabine Hossenfelder said (about the “task” in question) that: “the particular calculation in question is to produce a random distribution. The result of this calculation has no practical use.” And also: “while the announcement is super impressive from a scientific pov and all, the consequences for everyday life are zero.” Sabine Hossenfelder (@skdh) on X
After the initial euphoria, comments cooled down a bit:
The WSJ said that “quantum computing’s payoff is years away at best”. Google’s Quantum Boost Doesn’t Really Compute,
And the Financial Times admitted that “it is still difficult to anticipate exactly when quantum computing effects will be felt or how sweeping they will be” The mysterious promise of the quantum future
Intelligence Augmentation
Metaverse
Zuckerberg’s vision about smart glasses and the Metaverse is starting to be vindicated:
Google is working with Samsung to develop smart glasses and VR headsets. The commercial success of Meta Ray-Bans can be seen as a driver of this. Google plans new smart glasses and VR headsets in Samsung partnership
More startups are also emerging in this space. An example is Looktech, a company that is building “AI-glasses” (very similar to the smart Ray-Bans). Looktech unveils AI glasses with personalized assistance and media capture
Meanwhile, Meta keeps working on the vision:
They are developing new AI models specifically targeting Metaverse applications. Meta releases AI model to enhance Metaverse experience
They just announced a new update for the Quest headset. Meta’s new Quest update has faster hand tracking and at-a-glance PC connections
At the same time, Apple is finding that “virtual computer screens” could be the killer application for the Vision Pro headset. The Vision Pro’s ultrawide Mac display is very close to being a killer app