Headlines this week - Mar 2, 2025
A look at how capital is being deployed across future opportunities
This week in the future:
Doubts about a potential deceleration of investment in AI infrastructure (chips, data centers, energy) seem now dissipated (temporary at least), after Nvidia’s results call this week. The company reported a revenue growth of around 80% yoy, driven by sales of their AI chips. Even if some analysts were disappointed (mainly by the company’s guidance for 2025), this does not look like deceleration to anyone
But some things have started to change:
New models are being explored to fund deployments: Meta is building a new fund with external investors, with a target size of $35bn, which will be dedicated to investments in AI infrastructure. This reminds us of similar models that are still being used in the telecom industry to support fiber deployments
Nvidia is also evolving. Their recent success has been partially driven by a new line of processors (Blackwell) with more memory and less precise calculations, which is more appropriate for inference (I.e. the process through which trained models make decisions) rather than training. So their results would actually support the view that computing power (and infrastructure business models) are progressively shifting to inference
Competition to launch new AI models keeps heating up. In part, this is related to the current trend of increasing computation in the inference phase (also affecting infrastructure, as we just mentioned). New “hybrid” models are being produced, combining fast, “intuitive” answers to some questions, with deeper, more “reflective” answers to others
Anthropic has announced Claude 3.7, which they describe as “the first hybrid-reasoning model”
OpenAI has launched GPT-4.5, which they say is very effective in reducing hallucinations
Tencent has announced Hunyuan Turbo S, a new model that, according to their claims, has better performance than DeepSeek (the “rock star” among Chinese models right now)
But AI takeoff in the economy (the productivity wave that S Nadella and others expect) could take time
Tyler Cowen (an economist and a thought leader on the impact of new technologies) expects take-off to be “relatively slow”. Among other decelerating factors, he mentions: (1) inefficient sectors, with increasing weight in GDP, will be slow to adopt AI; (2) humans in the loop will become bottlenecks for productivity growth; (3) diffusion of new technologies has traditionally taken more time than first expected; (4) historically, GDP growth has almost never been much higher than 2% (for “somewhat mysterious reasons”). He also points out that financial markets are not discounting the kind of productivity expansions that people like Satya Nadella are talking about, even if there is plenty of information about the technology being deployed.
Meanwhile, consumer AI apps are proliferating. Lots of new products are being developed / launched:
Amazon has announced a new version of Alexa, incorporating GenAI and some “agent” capabilities. It will be free for Amazon Prime subscribers (starting next month)
Meta is launching a standalone app for MetaAI, to (more) directly compete with ChatGPT, Google Gemini, Perplexity, or even Europe’s LeChat (by Mistral). This is significant because it means offering a Meta service from outside the company’s “family of apps”
New Quantum Computing announcements:
Amazon presented a new chip, produced through the company’s collaboration with the Caltech team led by quantum computing “giant” John Preskill
But useful quantum computers could be relatively far away in the future. Analysts recognize that building practical machines will only be possible by the end of this decade, and most probably later
Can Europe become self-sufficient in strategic technologies?
More voices claim for Europe’s self-sufficiency in technology: Last week M Schaake, at the FT, argued that a tech ecosystem independent from the US is a key strategic mandate, to protect Europe’s sovereignty.
Europe is a global leader in nuclear fusion research. An announcement this week by a team in France, which managed to maintain a fusion reaction active for 22 minutes (beating a previous record from a Chinese team), confirmed that Europe is a leading power in this emerging technology
Local startups are working to make Europe autonomous also in space technologies: Isar Aerospace, a German startup founded seven years ago, is building a (privately funded) rocket, that they’re planning to test-flight from mainland Europe
China keeps working to catch up with the US in space technologies:
Building a (global) alternative to Starlink is a priority. SpaceSail, a Chinese startup, is working with the government of Brazil to build a new satellite broadband constellation. Chinese authorities perceive these services as a strategic priority, due to their ability to “transcend national boundaries, penetrating sovereignty and unconditionally covering the whole world”.
Startups are moving ahead with reusable rockets, too. Deep Blue Aerospace, another startup, is working on reusable rockets, which they see as a “disruptive technology that will have huge market potential in the future”. The company has a valuation of around $700m and would be planning an IPO for 2028.
A revolution might be under way in preventive medicine. Biotech labs are developing next-generation blood tests that will be able to detect (1) aging of different organs, pointing to higher risks of illness, including cancer; and (2) specific types of serum nanoparticles believed to be early signs of some cancers. The first type of test is on the way to be commercially viable, but initially at high prices
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 :-)
LINKS:
1 - Population & natural resources
Biotech
A revolution might be under way in preventive medicine. Biotech labs are developing next-generation blood tests that:
Evaluate the aging of different organs, pointing to higher risks of illness, including cancer. ‘Quick and easy’ blood tests for ageing organs detect disease risks
Detect specific types of serum nanoparticles believed to be early signs of some cancers. AI Unveils Hidden Nanoparticles – A Breakthrough in Early Disease Detection
Space
China keeps working to catch up with the US in space technologies:
Building a (global) alternative to Starlink is a priority. SpaceSail, a Chinese startup, is working with the government of Brazil to build a new satellite broadband constellation. Chinese rivals to Musk's Starlink accelerate race to dominate satellite internet
Startups are moving ahead with reusable rockets, too. Deep Blue Aerospace, another startup, is working on reusable rockets, which they see as a “disruptive technology that will have huge market potential in the future”. China’s Leading Private Rocket Company Has Eye on Listing in 2028
Europe is also moving to become self-sufficient. Isar Aerospace, a German startup, is building a (privately funded) rocket, that they’re planning to test-flight from mainland Europe. German startup to attempt the first orbital launch from Western Europe
2 - Efficiency & Productivity
Energy
Nuclear
Europe confirms its global leadership in nuclear fusion research. An announcement this week by a team in France confirmed that Europe is a leading power in this emerging technology. French Scientists Beat China’s Fusion Record With 22-Minute Plasma Reaction
In Japan, regulatory barriers are delaying the restart of the world’s largest nuclear facility. Post-Fukushima rules require the construction of anti-terrorism equipment that the company says won’t be ready until 2029. Tepco Falls on Risk of Delay to Restart World’s Top Atomic Plant
Fossil Fuels
Data center energy consumption already has significant impact on carbon emissions. Research from Californian universities estimates that the cost of treating illnesses connected to this pollution was $5.4bn since 2019, only in the US. Pollution from Big Tech’s data centre boom costs US public health $5.4bn
New Transport Technologies
Electric Vehicles
EVs are driving a deep transformation of the car industry. A new book tells the story of how combustion engine engineers (at companies like Ford) are being made redundant by the shift to electric cars. The Engineers that EVs Made Expendable
New players (often from China) are on the way to dominate the industry. In this new world, companies like China’s Xiaomi, previously only a smartphone vendor, are now threatening Ford and GM. The Chinese EV Maker Threatening Ford and GM
There are also EV startups in the West, but they’re suffering. Lucid, one of the startups that were launched after Tesla’s success, is struggling. The CEO is leaving and a new plan to double production and introduce an SUV model has been launched. Lucid chief resigns as EV maker pledges to double production
Artificial Intelligence
AI: Apps
B2C
Lots of new products being developed / launched:
Amazon is launching a new “GenAI” version of Alexa. “Alexa+” will incorporate AI-agent capabilities, and will be free for Amazon Prime subscribers. Amazon Unveils Long-Awaited AI Version of Alexa Voice Assistant
Alexa+ will be commercially launched next month. Initially for Echo speakers, then (over subsequent months) with other versions, including smartphone apps. Amazon to Launch Generative-AI Alexa+ Next Month
Meta is launching a Meta AI chatbot app. This will reach users outside the company’s “family of apps”, in competition with existing apps from Meta’s direct rivals in AI (OpenAI, Gemini, Claude, Perplexity, and others). Meta’s AI chatbot will soon have a standalone app
Perplexity wants to integrate AI in the web browser. The company is working on a web browser (“Comet”) that will integrate AI. Analysts are skeptical, due to high competition in the browser market. Perplexity wants to reinvent the web browser with AI—but there’s fierce competition
TikTok will offer AI tools to facilitate content creation. They’re planning to substitute their current “creator marketplace” with a new platform (“TikTok One”) with AI-enabled features. TikTok sunsets its creator marketplace for TikTok One, a broader solution with AI tools
Quora is doing the same with app creation. Quora is going one step beyond, and is launching a platform (“Poe”) with AI-powered tools for users to create their own apps. Quora's Poe now lets users create and share custom AI-powered apps
AI is changing the way students work. A recent survey in the UK shows that more than 90% undergrad students are now using AI to do their work, compared with ~70% a year ago. Surge in UK university students using AI to complete work
Estonia wants to embrace the trend. The government is partnering with OpenAI and Anthropic on an initiative (“AI Leap”) to teach artificial intelligence skills to high school students. Estonia launches AI in high schools with US tech groups
B2B
Expectations on AI impact are high, but there are still lots of uncertainties:
Eric Schmidt (ex-Google) is bullish on the opportunities. He expects “a new renaissance in human knowledge and capability”. According to his WSJ essay, AI could accelerate drug discovery, make it possible for a single person to run a whole company, personalize education, or create new materials for space exploration. Opinion | AI Could Usher In a New Renaissance
But Tyler Cowen expects take-off to be “relatively slow”. In this short essay he mentions several reasons why productivity growth could be lower (or take more time) than what many people expect. Why I think AI take-off is relatively slow
Salesforce shows how vendors are trying to address adoption challenges:
Acceleration of AI adoption through Agents is at the core of the company’s equity story. The company describes AI Agents as a tool designed to handle multi-step processes on behalf of users without human supervision. Salesforce Stock Revival Hinges on Success of AI Agents
They are partnering with Google in enterprise AI solutions. The agreement includes the combination of Google’s Gemini model with Salesforce’s agents. Google Cloud and Salesforce are partnering to bring enterprise AI to more businesses.
New use cases are continuously being explored:
Coding: Google is now offering a version of its AI-powered coding tool (“Google Gemini Assist”) for free, targeting students, hobbyists, freelancers, and startups. Google Gemini’s AI coding tool is now free for individual users
Weather forecasting: A new AI system in Europe is expected to provide predictions for up to 15 days ahead, and to improve the tracking of extreme weather events. Weather forecasting takes big step forward with Europe’s new AI system
AI: Robots
Small robots working as a collective can be used to create new “smart” materials. Researchers at UC Santa Barbara and TU Dresden have developed a material-like collective of robots with behaviors inspired by biology. Robots in the collective are programmed to assemble themselves together into various forms with different material properties. How to get a robot collective to act like a smart material
AI: Foundational Models
Competition to launch new AI models keeps heating up:
Anthropic has announced Claude 3.7, which they describe as “the first hybrid-reasoning model” Anthropic Launches the World’s First ‘Hybrid Reasoning’ AI Model
OpenAI has launched GPT-4.5, which they say is very effective in reducing hallucinations OpenAI reveals GPT-4.5 amid flurry of new AI model releases
Tencent has announced Hunyuan Turbo S, a new model that, according to their claims, has better performance than DeepSeek (the “rock star” among Chinese models right now) Tencent Releases Turbo AI Model It Says Is Faster Than DeepSeek
DeepSeek is speeding up the release of R2, its new model. It was originally planned for May 2025, but it will probably come earlier. DeepSeek rushes to launch new AI model as China goes all in
xAI’s new model, Grok 3 looks successful, for now. Daily active users for Grok’s U.S. app (now powered by Grok 3) grew more than +260% last week, while global daily active users climbed 5x week-over-week. Grok 3 appears to be driving Grok usage to new heights
Valuations remain strong. Anthropic just closed a funding round at a $61.5bn valuation. So there is no significant “post-DeepSeek effect”… Exclusive | AI Startup Anthropic Finalizing $3.5 Billion Funding Round
AI: Security & Safety
AI Safety is a hot topic, according to investors… At least, that’s how it looks if your founder is Ilya Sutskever (an OpenAI founder, and -for many people- a genius). His new company, Safe Superintelligence, is raising more than $1bn at a $30bn, with not even a product prototype to show. Sutskever is happy that the company will be “fully insulated from the outside pressures of having to deal with a large and complicated product and having to be stuck in a competitive rat race.” Pre-product AI ‘company’ now valued at $30bn
AI: Infrastructure
Doubts about a potential deceleration of investment in AI infrastructure (chips, data centers, energy) seem now dissipated (temporary at least):
A rumor about Microsoft cancelling leases for AI data centers made investors nervous at the start of the week. The source was a report by TD Cowen, an investment bank. They said that the company had cancelled leases for a capacity of approx. 200MW (so it looked substantial). Microsoft Cancels Leases for AI Data Centers, Analyst Says
Microsoft has denied this rumor. In an email to CNBC, the company said that their plans to invest in AI infrastructure remain on track. Microsoft denies claims it is cancelling data centre leases
After that, Alibaba announced a $52bn investment in the next 3 years. This is not such a large figure, compared with previous announcements by Amazon, Microsoft or Meta, but it is significant coming from China (where DeepSeek was supposed to have found a way to make infrastructure less relevant…) Alibaba to Spend More Than $52 Billion in AI, Cloud Over Next Three Years
Also, Nvidia’s results just showed that investment is not decelerating. The company reported a revenue growth of around 80% yoy, driven by sales of their AI chips. Even if some analysts were disappointed (mainly by the company’s guidance for 2025), this does not look like deceleration to anyone. Nvidia’s Bumper Sales Show AI Bonanza’s Strength Nvidia revenues jump almost 80% on booming AI chip sales
On the other hand, Big Tech companies are exploring new models to fund deployments:
Meta is building a new fund with external investors, with a target size of $35bn, which will be dedicated to investments in AI infrastructure. This reminds us of similar models that are still being used in the telecom industry to support fiber deployments. Meta in Talks to Raise $35 Billion for Data-Center Financing Led by Apollo
Amazon is using savings from operational automation to fund the infrastructure expansion. About a quarter of the company’s massive investment this year will be dedicated to automate its retail operations. The savings from this will contribute to improve the investment business case. Amazon bets savings from automation can help fuel AI spending boom
CoreWeave’s IPO is also a sign of confidence. The company is a data center operator, with AI as key revenue growth driver. So their decision to go public looks like a positive sign for infrastructure investment momentum. Data centre operator CoreWeave lays groundwork for IPO
AI: Chips
Nvidia is adapting to the latest trends in AI computing. The company’s recent success has been partially driven by a new line of processors (Blackwell) with more memory and less precise calculations, which is more appropriate for inference. How Nvidia Adapted Its Chips to Stay Ahead of an AI Industry Shift
Huawei could become the Chinese “AI chip champion”. The company is radically improving the efficiency of their chip production line, which is leading to profitability in this business (a key step for Chinese self-sufficiency in this industry). Huawei improves AI chip production in boost for China’s tech goals
Microsoft sees US government’s chip export restrictions as part of the problem. They are complaining to President Trump, arguing that the ban could be a “strategic mis-step” Microsoft urges Donald Trump to rethink AI chip export controls
Quantum Computing
Useful quantum computers could still be relatively far away in the future. R Waters (FT) estimates that building practical quantum computing machines will only be possible by the end of this decade, and most probably later. Quantum computing is struggling to reach its silicon moment
In the meantime, announcements continue:
This week was Amazon’s turn. After Microsoft announcements last week, Amazon presented a new chip, produced through the company’s collaboration with the Caltech team led by quantum computing “giant” John Preskill. Amazon’s first quantum computing chip makes its debut
In a social media post, Preskill explained why this is relevant. He claimed that “there is still far to go, but we hope that Ocelot’s unique architecture will shorten the path to quantum utility that benefits the world.” Amazon announces Ocelot quantum chip | John Preskill
And Quantum Computing startups are still attractive for investors. Quantum Machines, an Israeli startup, has just raised $170m, the fifth-largest ever round for a quantum computing company. Startup Quantum Machines Raises $170 Million in Series C Round
Connectivity
Google wants to start the next generation of wireless connectivity (using visible light): Taara, a “moonshot project” at Google’s X lab, is working on a solution to provide high speed wireless broadband via laser beams. Now the company is ready to launch a new “Taara chip” that could lead to actual commercial deployments. Google’s Taara Hopes to Usher in a New Era of Internet Powered by Light
Intelligence Augmentation
Brain-Computer Interfaces
A solution to pilot drones with thoughts is being developed. A research team in China is building a two-way brain computer interface to communicate drones with human brains. In one demonstration, volunteers using it were able to steer a drone through a 3D obstacle course using their minds. Two-way brain-computer interface pilots drones with thoughts
3 - Economic / Business trends
Tech & Geopolitics
The idea that Europe needs to be self-sufficient in technology is gaining traction. Last week M Schaake, at the FT, argued that a tech ecosystem independent from the US is a key strategic mandate, to protect Europe’s sovereignty. Europe’s dependence on US tech is a critical weakness
In China, DeepSeek is still perceived as geo-political game-changer. According to this Chinese analyst, the company’s achievements with its R1 model reveal how the Chinese leadership in hardware production is turning into an advantage to build cheaper AI models. China’s superpower of scaling will spur DeepSeek’s competitive threat
The Chinese government is actively pushing for DeepSeek adoption across the country, looking to reinforce the company by creating economies of scale. DeepSeek spreads across China with Beijing’s backing
Abu Dhabi’s royals want to control global AI systems like they previously controlled oil production. Sheikh Tahnoon bin Zayed Al Nahyan, the United Arab Emirates’ national security adviser and brother of the president, is leading a new AI fund (MGX) that plans to manage more than $70bn for AI investments globally (including participation in the American Stargate project). The ‘Spy Sheikh’ Taking the AI World by Storm
AI Economics
Investments in AI infrastructure could be not so beneficial as some people say. Data center building may require many human workers, but once the facilities are built, the number of jobs needed to operate them might decrease quite substantially. At least this is what happens with the massive data center that OpenAI is building in Texas, which could be run by just 100 full-time employees (vs. 1,500 used in the construction stage). The AI Data-Center Boom Is a Job-Creation Bust
AI copyright is turning into an issue:
UK artists want restrictions on how their content can be used for AI training. In the UK a group of pop artists have launched a “protest album” against a UK government’s proposal to alter copyright law in order to help tech companies train GenAI models. A musical supergroup fights AI — with a silent protest album
This is seen as part of a worldwide revolt by creative artists and content companies, which want to be compensated if their materials are used for model training. Several start-ups (ProRata.ai, TollBit, Human Native.ai) are experimenting with new economic models that would maintain incentives for creators. But it is not yet clear if AI companies will accept these new arrangements. Help is coming in the AI copyright wars
Thank you for great insights, as always. Looking forward to reach you in the coming days !