Headlines this week - Feb 2, 2025
A look at how capital is being deployed across future opportunities
This week in the future:
This was week 1 after DeepSeek, and the technology industry is currently kind of obsessed with the topic. Some messages from the (many) discussions in press and social media:
The emerging consensus seems to be that Open Source LLMs have won, and that this will lead to a commoditization of these models. Some analysts have compared DeepSeek (as an Open Source “operating system” for AI) with Android (an Open Source operating system for smartphones). In this scenario, OpenAI, Anthropic or Google could now aspire to be the Apple of AI (with a less open, more integrated approach to end-user applications)
In any case, as a result of DeepSeek and other Open Source models, many people are expecting an acceleration of innovation in AI apps, which could be beneficial for end users, and for the economy. It could also catalyze an “explosion” of new startups (so some VCs are rather optimistic…) All this could benefit companies like Meta and Apple (which are more interested in monetizing apps than in monetizing the “engines” or “models” behind them)
There are different views about the implications on infrastructure investments. The stock market has severely penalized US semiconductor companies this week, as their large growth expectations linked to AI have drastically cooled down. However, many voices are claiming that computing capacity is still going to be key, even if its role eventually changed from training to inference, and even if it would not depend so much on Nvidia’s chips
In any case, this looks like a challenge to the previous geopolitical status quo, in which the US was assuming an advantage vs. China in these technologies, that now does not look so clear. It remains to be seen how the different actors react. The US may be tempted to become more open (reduce chip export controls? OpenAI open sourcing some of its tools?) or more closed (more / deeper controls? A Manhattan Project for AI?)
In other topics:
US energy companies suffered a stock market selloff this week, after the DeepSeek news, and this revealed how much their valuations rely now on the growth perspectives for AI infrastructure (and its associated energy demand)
Boom Supersonic, a startup building a “next-generation”, silent supersonic commercial plane, crossed the speed of sound with a prototype this week
Helion, the nuclear fusion startup supported by Sam Altman, among others, has just raised $425m at a $5.4bn valuation
Investors’ money keeps flowing to startups planning to use AI for medical research. An example is Manas AI, a startup led by ex-LinkedIn R Hoffman, that has raised $25m to develop new drugs against cancer, using AI
Jeff Bezos’ Blue Origin consolidating as an alternative to SpaceX in rockets
I hope you’ll enjoy it. Please share this with your friends if you think it is interesting. And give me feedback if you’re missing something :-)
1 - Population & natural resources
Biotech
Momentum grows for the application of AI to medical research:
A project to use AI against cancer raises $25m. Manas AI (the company) is led by ex-LinkedIn R. Hoffman. Medical research and drug discovery startups received $3.34bn funding in 2024 (5x vs. 2023). Exclusive | Reid Hoffman Raises $24.6 Million for AI Cancer-Research Startup
Chinese researchers have developed an AI model that makes the edition of proteins easier / more agile. Protein editing is key to develop drugs based on how the designed proteins interact with other molecules. AI-driven multi-modal framework improves protein editing for science and medicine
The US FDA has just approved a bioengineered blood vessel to treat vascular trauma, which could help patients that don’t have any spare veins to reroute blood flow. The “artificial” vessels are grown in the lab from pigs’ arteries cells. This Blood Vessel Was Grown in a Lab With Real Human Cells
Foodtech
Could bacteria be the food of the future? Scientists and startups are trying to identify microorganisms that could be a protein source, as a more efficient alternative to traditional foods (using less land, water, and fertilizer). Seeking New Foods, Scientists Look To Bacteria, Algae, and More
Space
Jeff Bezos’ Blue Origin has now been confirmed as an alternative to SpaceX, after the successful launch of their New Glenn rocket. With successful New Glenn flight, Blue Origin may finally be turning the corner
2 - Efficiency & Productivity
Energy
After DeepSeek, the market starts to questions the previously bullish expectations for energy companies:
A “reality check” for the energy industry: If progress in AI can happen without a massive increase in computing, we won’t need so many giant data centers, and our energy needs won’t grow as previously expected. AI’s energy obsession just got a reality check
A selloff of energy stocks was triggered after the news about DeepSeek. The International Energy Agency views this as a proof that the relation between energy demand and AI progress is “not well understood”. DeepSeek-driven sell-off puts power demands of AI in doubt, says IEA
Valuations in the energy industry have become very dependent on AI needs. Some companies lost as much as 15% of their market value on a single day. DeepSeek Reveals How Badly Energy Industry Needs AI for Growth
This could benefit the green transition (under question in the last few months because green energies were apparently not enough to address the demand tsunami that was coming). DeepSeek sticks a fork in utility investors’ socket
Nuclear
Helion, a leading nuclear fusion startup, has a valuation of $5.4bn. They just closed a $425m founding round. S Altman (OpenAI) and P Thiel are among its backers. US nuclear fusion start-up backed by Sam Altman and Peter Thiel secures $425mn
Renewables
Surprisingly, Middle East oil production leaders are moving to lead the world in renewables. Masdar, an Emirati oil company is building a $6bn solar plant. Meanwhile, Saudi Aramco is investing in the production of lithium (a key element for batteries). Middle East becomes fastest-growing renewables market outside China
New Transport Technologies
Electric Vehicles
The crisis in US demand triggers a crisis in Tesla:
Demand is cooling down in the US, and Tesla is suffering: The company’s 4Q24 results were mixed, but the car division reported a -8% fall in sales vs. 4Q23 (first decline in more than 10 years), due to promotions used to stimulate demand. Tesla Caps Roller-Coaster Year With Mixed Fourth-Quarter Earnings
In Europe, anti-China regulation could also penalize Tesla. So the company is now suing the EU over the tariffs imposed on imports of electric vehicles from China. Tesla sues EU over tariffs on electric vehicles from China
Robotaxis are becoming the company’s big hope. In the call, Musk announced the launch of a robotaxi service in Texas within six months, and linked this to the company’s recovery. Tesla results disappoint but Musk touts coming robots and ‘cybercabs’
Autonomous Cars
Will people accept self-driving technology? Some concerns published this week about this:
People don’t seem to trust fully driverless technology. The share of US drivers who say they are “afraid” of automated cars has risen from 54% in 2021 to 66% in 2024, while trust has dropped from 14% to 9% in 2024. The Big Question: would you trust a driverless car?
Even the promise of safer roads can be questioned. Current numbers of fatalities are already very low for human drivers, and the bizarre mistakes often made by autonomous cars suggest it won’t be easy for them to perform better. Can self-driving cars save us from ourselves?
Supersonic Airplanes
Boom Supersonic crosses a milestone. The company is building a “next generation Concorde”, with silent engines that would open more markets and make the airplane commercially profitable. Their XB-1 (small) prototype broke the sound barrier in a test flight this week. Boom Supersonic jet breaks sound barrier to open way for new ‘Concorde’
Artificial Intelligence
This was week 1 of the “DeepSeek era”
There was a stock market “blood bath” on Monday, with Nvidia at the center. People talked about DeepSeek “turning Tech and Wall St upside down”. The Day DeepSeek Turned Tech and Wall Street Upside Down
Leading analysts and industry experts have been busy making comments. Some of them actually agree about some key points about what happens next:
S Rosenbush (WSJ) compares DeepSeek with Android. He converges with the “AI commoditization” theme, and believes DeepSeek could turn into “the Android of AI” (so maybe American Big Tech giants will now fight to play Apple’s role?). DeepSeek’s Big Question: Where Does AI’s True Value Reside?
M Siegler argues that AI business models based on charging users may be hurt. From this perspective, Facebook is in better shape than Google, Microsoft or OpenAI. AI, Uh, Finds a Way
B Thompson sees Open Source as the big winner, and the commoditization of AI models (i.e. the challenges for businesses based charging) as a clear implication. DeepSeek FAQ
Parmy Olson (Bloomberg) argues (probably correctly) that capital intensity (I.e. money) will now not be enough to as a competitive advantage / moat for OpenAI, Google and the other Big Tech giants. DeepSeek Shows Silicon Valley’s Huge Blindspot on AI
Dan Gallagher (WSJ / Heard on the St) thinks the panic is overblown. He’s specifically talking about the chip companies. DeepSeek Won’t Sink U.S. AI Titans
A Karpathy (a leading technical expert) argues that computing (I.e. facilities, and chips) will remain crucial. He also believes that this shows the power of Reinforcement Learning (vs. “imitation” / “brute force” learning). Andrej Karpathy (@karpathy) on X
AI applications could be the big opportunity now. The emerging takeaway from almost everyone seems to be that Foundational Models are being commoditized, and that this will now catalyze innovation in apps, with hardware still under question (but possibly also an opportunity, given the importance of “compute” as a driver of progress). Software Stocks In Vogue Again on DeepSeek’s Efficiency Promise
AI: Apps
Agents
At initial stages, agents are expected to be highly specialized in very repetitive, low-risk tasks. According to P Olson at Bloomberg, companies waiting for “Super Intelligence” may lose a shorter-term opportunity to create value. Skip the Hype, Here’s How AI ‘Agents’ Can Really Help
B2C
Meta’s AI chatbot becomes more powerful, with a capacity to remember the user profiles. Meta AI will use its ‘memory’ to provide better recommendations
B2B
OpenAI just released “ChatGPT Gov”, a version of its chatbot designed for US government agencies. This happens in the context of the close relationship they are now building with Trump. OpenAI launches ChatGPT plan for US government agencies
Mercor, a startup using AI to match profiles with jobs, initially focused on the AI industry, is now valued at $2bn, after a recent funding round in which they’ve raised $75m. The valuation has increased by 8x in just 1 year. Exclusive | AI Recruiting Startup Mercor Rockets to $2 Billion Valuation
AI: Robots
Robots continue to attract investors:
SoftBank wants to invest in a startup developing a “general-purpose brain” for robots. Skild AI (the startup) is building an LLM for robots, and is trying to raise funds at a $4bn valuation. Softbank (very active these days) is negotiating a deal. SoftBank set to invest in robotics start-up Skild AI at $4bn valuation
Boston Dynamics may think it’s time for its IPO. Hyundai Motor, who owns the American robotics pioneer, is rumored to be preparing the offer, but they have not yet confirmed any details. Boston Dynamics IPO may be in works
AI: Foundational Models
Arguments about commoditization of Foundational Models are proliferating:
DeepSeek R1 obtains comparable results to OpenAI models, at a much lower cost and with the possibility of even smaller versions that could easily run on mobile phones and tablets. Yes, R1 is weaker than OpenAI models “on specific types of problem solving”, but it’s just a start… How DeepSeek’s AI Stacks Up Against OpenAI’s Model
DeepSeek may be just the start. Many more companies across the world are now developing similar Open Source models (that could now also take advantage of DeepSeek’s innovations).DeepSeek’s ‘Open AI’ Should Terrify Sam Altman
Alibaba’s announcements this week are a clear proof of this:
On Monday, Alibaba launched a new family of AI “agentic” models that can control PCs and mobile phones (similar features to OpenAI’s recently announced “Operator”). Alibaba's Qwen team releases AI models that can control PCs and phones
Competition in China is heating up. The way Alibaba presented its new products, with direct comparisons vs. DeepSeek’s models, indicates the high level of competition currently happening within China, too. Alibaba Unveils Upgraded AI Model, Claims It Surpasses Rival DeepSeek-V3
DeepSeek also presented a new model, for multi-modal applications, directly targeting OpenAI’s DALL-E and Stable Diffusion. DeepSeek unleashes ‘Janus Pro 7B’ vision model amidst AI stock bloodbath, igniting fresh fears of Chinese tech dominance
Sam Altman is responding to the pressure:
OpenAI will now fast-track “much better models” that would be on the way. OpenAI’s Altman vows ‘better models’ as China’s DeepSeek disrupts global race
On Friday they released o3-mini, a smaller “thinking model” that (like DeepSeek’s products) fact-checks itself before giving out results. OpenAI launches o3-mini, its latest ‘reasoning’ model
Will he make OpenAI open again? Altman has even said (also on Friday) that there is an internal discussion at OpenAI about the “possibility” of releasing some of the technology within its AI models and publish more research showing how its systems work. Sam Altman’s Answer to DeepSeek Is Giving Away OpenAI’s Tech
AI: Security & Safety
Have we already crossed the “danger” threshold with AI? This (apparently true) story about a young guy who built a potentially dangerous device simulating nuclear fusion, using Anthropic’s Cloud AI for instructions, would be a good example of how AI is “democratizing” dangerous knowledge. A Young Man Used AI to Build A Nuclear Fusor and Now I Must Weep
Controversy among technology experts about AI risks. A few days ago at Davos, D Hassabis (DeepMind) claimed that Artificial General Intelligence “could threaten civilization if it runs out of control or is hijacked by bad actors”; D Amodei (Anthropic) described “1984 scenarios, or worse” and Y Bengio (University of Montreal) said that “science doesn’t know how we can control machines that are smarter than us”. Meanwhile, Y LeCun (Meta) perceives hypocrisy in comments like these, being made by people who work for companies running to improve models that remain “closed”, and looking for competitive advantages. AI leaders clash over safety and $100bn Stargate project
AI: Infrastructure
DeepSeek’s new models have put into question the need of large investments in computing infrastructure. DeepSeek’s ‘aha moment’ creates new way to build powerful AI with less money
As a consequence, there was a shockwave through markets on Monday, as investors are now much more uncertain about the likely future investment in AI hardware. Wall Street set to open lower as DeepSeek sows doubts about AI spending
In this uncertain context, SoftBank does not seem to be scared:
They’re negotiating an enormous (up to $25bn) investment in OpenAI (the biggest-ever for Sam Altman’s company). OpenAI is expected to use all this money to invest in infrastructure. SoftBank in Talks to Invest as Much as $25 Billion in OpenAI
The deal could apparently be closed at a $300bn valuation of OpenAI (almost 2x in just 1 year…) Exclusive | OpenAI in Talks for Huge Investment Round Valuing It Up to $300 Billion
For SoftBank, this is on top of an additional $19bn that they have committed for the Stargate project, also linked to AI infrastructure (so their total exposure reaches $44bn, for now…) OpenAI’s Sam Altman and SoftBank’s Masayoshi Son Are AI’s New Power Couple
In their results calls this week, Big Tech companies also confirmed their commitment to invest:
Meta ($60-$65bn in 2025) and Microsoft ($80bn in 2025) explicitly confirmed their investment plans. Microsoft, Meta Talk Up Their Big AI Ambitions and Spending Plans
At Meta, Y LeCun argues these large investments are still required. Like other experts (see A Karpathy’s comments above), he claims that computing capacity is still key, even if it might now be focused on inference (and “reasoning”), rather than (“raw”) training. https://www.threads.net/@yannlecun/post/DFVXCiMuHWL?xmt=AQGzAbhAq44OO9kPKthPgEp6soT26QV1tuXQxSsBIO8mHA
The “official” reaction at Meta is that their view has actually been vindicated (particularly about Open Sourcing). DeepSeek Shows Meta’s A.I. Strategy Is Working
Meanwhile, at Microsoft, S Nadella welcomed the AI demand acceleration that DeepSeek would be expected to bring. He invoked “Jevons paradox” for that (i.e. commoditization often accelerates product demand, and can actually improve profits). Satya Nadella (@satyanadella) on X
An interesting first move by Microsoft has been to become more “neutral” with their AI infrastructure, which now is supporting DeepSeek too. Microsoft brings a DeepSeek model to its cloud
Finally, independent investors (like Blackstone) are also running to justify their data center investments. Blackstone has already invested $80bn in computing facilities. Blackstone defends $80bn data centre investment as DeepSeek shakes market
AI: Chips
The “DeepSeek tsunami” is making investors reestimate the demand of high-end chips (see above). DeepSeek Undercuts Belief That Chip-Hungry U.S. Players Will Win AI Race
The largest impact is being felt by Nvidia (the global leader in high-end chips). DeepSeek’s Rise Exposes Nvidia’s Weakness
Nvidia had a massive sell-off on Monday, that has continued during the rest of the week, after a brief / weak rebound on Tuesday. Nvidia shares resume fall despite gains in European chip stocks
Not even commitments by Microsoft and Meta to maintain their investment plans have been enough to contain Nvidia’s fall. Either because investors are skeptical on Microsoft and Meta’s commitments or because Nvidia’s valuation would rely on many other customers. Microsoft and Meta AI Splurge Still Not Enough for Nvidia
Even if “compute” remained as a key resource for AI progress, Nvidia might not be able to capitalize on that. For example, in China, Huawei has been reported to have been working with AI companies, including DeepSeek, to adapt models trained on Nvidia GPUs. Why Nvidia investors are spooked by Chinese AI upstart DeepSeek
Meanwhile, other giants in the semiconductor supply chain could actually benefit from the DeepSeek developments, e.g.: if this finally catalyzed an acceleration of AI demand (and all kinds of chips behind it)
This is the case for TSMC (which the WSJ has shown as a potential arbitrage opportunity, this week). The Taiwan company is the leading manufacturer of chips, so even if Nvidia lost market share, TSMC could benefit from whoever was replacing them (except maybe some Chinese players). DeepSeek’s Deeper Impact Is Positive for Some Hard-Hit Stocks
And this is also the case for ASML, which builds machines for chip manufacturing. They presented results this week and claimed that orders were exceeding previous forecasts. ASML Orders Exceed Forecasts in Defiance of DeepSeek AI Selloff
Quantum Computing
Xanadu, a Canadian startup, has presented an “easily scalable” model of quantum computer, built on server racks. Aurora, the company’s prototype is still a 12-qubit machine (vs. e.g. IBM’s 1,121 qubits), but they see this more as a problem of limited investment (up to now) than as a technical issue. This quantum computer built on server racks paves the way to bigger machines
Intelligence Augmentation
Brain-Computer Interfaces
Echo Neurotechnologies, a Neuralink competitor, has raised $50m to develop brain implants. The company is lead by a famous neurosurgeon and a16z is one of the investors. Andreessen Horowitz Backs Neuralink Rival in New Round
Whole Brain Emulation
Scientists at Delft University have developed an artificial environment for neurons to grow like in a real brain, using 3D printing. The objective is to turn this into a tool to analyze how conditions like Alzheimer’s affect neural growth. TU Delft develops 3D-printed brain-like environment that promotes neuron growth
3 - Economic / Business trends
Tech & Geopolitics
The events this week have shown that the US might not be so far ahead of China in AI technologies. China could be catching up in spite of chip export controls. As we said last week, this is seen as a “Sputnik moment” by many. The global AI race: Is China catching up to the US? / DeepSeek is the Fork in History the US Needed
Does the US need a Manhattan Project for AI? The assumption that the country getting AGI first would become the (only) global super-power is making many experts in the US think that way. This could lead to more secretive technical work. So, even if many people think Open Source has won, the next step is not so evident. The Manhattan Project Was Secret. Should America’s AI Work Be Too?
OpenAI recent deal with Los Alamos seems to point in this direction. The company will run its o1 model in Los Alamos’ Venado supercomputer, (officially) to identify new approaches to treating diseases, protect the US power grid, enhance America's cyber and general security, and contribute to high-energy physics research. OpenAI inks deal with Los Alamos lab to cram o1 into Venado
Reasons to be worried: DeepSeek would now be working to migrate its model to Huawei chips, with some sources claiming that this leads to a very limited performance loss (-5%) and a very significant cost reduction (-70%). Olala🇻🇳 🇨🇳 🇷🇺 (@olalatech1) on X
It is unclear if DeepSeek “distilled” its models from OpenAI’s. There are several investigations under way. But does it matter? OpenAI says it has evidence China’s DeepSeek used its model to train competitor