Headlines this week - May 11, 2025
A look at how capital is being deployed across future opportunities
This week in the future:
1 - Google loses ~$140bn in valuation as new signals emerge of AI apps substituting search
Eddie Cue (from Apple) revealed that Google searches in iPhones have started to fall. This week Eddie Cue, Apple’s senior vice president of services, said that Google searches from the Safari browser have fallen in the last 2 months, highlighting that this had not happened before (“in over 20 years”). This made Google’s stock fall -9% last Wednesday (a $120bn blow), increasing pressure on the company, which is also under threat of regulatory action against its monopolistic position in… search
Cue attributed the fall to people using GenAI services, like ChatGPT or Perplexity, to find information. This was interpreted as a confirmation of a threat that many people were already expecting. Indeed, he suggested that Apple is currently planning to reshape its Safari web browser around AI services. In classical strategic management terms, AI would be a substitute on the way to fully disrupt Google’s massive competitive advantage in search. On the other hand, as we discuss below, Google is a leader in AI, with some of its models at the top of the performance benchmarks, and this could (partially) protect them from the threat, so maybe the market is overreacting a bit
Interestingly, the problem could affect Apple too. Apple’s stock also fell (-3%) on Wednesday, as Google’s weakness in search could also have a significant impact on them. The reason is that Google is now paying Apple $20bn per year to be the default search engine in Safari. If searches are going somewhere else, the case for Google paying this would weaken. Yes, new AI partners could partially compensate for the loss, but this is still very uncertain
The silver lining for Google is that this could help the company in its regulatory battle. The statements by Eddie Cue were made precisely in this context, during a courtroom session about Google’s antitrust case, and in particular about how, by paying Apple, Google would be engaging in an abuse of monopoly power. Google might use the threat signaled by Cue as an argument against some of the most aggressive actions being discussed (like a separation of the company different businesses). This is similar to what happened to Microsoft in the regulatory “browser wars”
2 - NewLimit: $130m funding round for an ambitious “Longevity” project to rejuvenate human cells
NewLimit, a startup , is working to rejuvenate cells through “reprogramming”. Their starting point are the currently well known “Yamanaka factors”, four genes that, when inserted into a cell, make it produce four proteins with specific effects on how other genes are activated. More specifically, they activate the other genes in a way that leads to a transformation of the initial cell into a rejuvenated, “pluripotent” stem cell. What NewLimit is looking for is a set of new “factors” that would actually “decouple” the rejuvenation part from the transformation into a stem cell. So cells could be rejuvenated while preserving their initial function
According to their vision, this would make it possible to “treat age-related disease to extend human healthspan” (as their mission statement says). I.e. to design new, safe drugs that would aim to rejuvenate all types of cells in the human body, reversing the age-related deterioration that is behind many mortal diseases
The company just raised $130m of new capital, in a Series B round led by Kleiner Perkins alongside new investors including Khosla Ventures, Human Capital, and Valor Equity Partners. These new investors are joining a long list of previous investors, with many “celebrities” among them, including Founders Fund (Peter Thiel), Brian Armstrong, John and Patrick Collison (Stripe), and Garry Tan (YCombinator)
3 - New nuclear energy projects: momentum is growing.
A $15bn project in Canada to deploy a next-generation “Small Modular Reactor” (SMR). Ontario Power Generation Inc. has won approval to build a small modular reactor (SMR) at a site outside of Toronto, with a total cost of $15bn for four reactors. The project is seen as a significant step forward for SMR technology, which is perceived as a way to meet surging power demand from artificial intelligence
A (well received) plan in the US to sell nuclear energy to run data centers and factories. Joe Dominguez, the CEO of Constellation Energy, the leading nuclear reactor operator in the US, has just announced that the company is close to sign long-term deals to provide clean, reliable and available” megawatts from nuclear plants to customers including technology companies running data centers. The reaction of the market was enthusiastic, with the stock rising +12% after the earnings call in which the comments were made
A government-sponsored plan to turn Argentina into a “global hub for nuclear energy”. The final purpose here would be to make the country a center for artificial intelligence, powered by investments that President Milei hopes to draw from big tech firms. The plan is to deploy SMRs (the next-generation reactors discussed above) in Patagonia, and a local state-owned company called Invap has already patented a design for a reactor of this type
4 - Material scientists develop a dynamic material that self-heals after puncturing. A team at Texas A&M University has created a new kind of polymer with a unique self-healing property “never before seen at any scale”. When struck by a projectile, the material stretches in a way that allows the projectile to pass through leaving a much smaller hole than in normal conditions. They are considering initial applications in the space industry, to build space vehicles more resistant against “micrometeoroids”, but it is not difficult to imagine use cases in the field of military operations
5 - Robots evolving fast, and already viewed as a strategic technology
Robots’ capabilities keep improving. Amazon showed some progress this week. Amazon has developed a new robot (Vulcan) with a sense of touch, which makes it very useful to manage items in the company’s warehouses. They are describing the achievement as a “fundamental leap forward”, and highlighting that, after this, robots will be “not just seeing the world, but feeling it”. Some people are concerned about the potential impact on jobs, but Amazon is (for now) positioning Vulcan as a complement to humans
In the US, humanoid machines are being proposed as a catalyst for re-industrialization. In a Bloomberg article this week, the development of humanoid robots is positioned as an opportunity for the country to recover its role as an industrial / manufacturing power, something that the Trump Administration would very much like. Only 4 countries (China, Japan, South Korea and Germany) are currently building 70% of the robots produced globally. The US is far behind, but catching up in this area, as opposed to e.g. ship building, looks possible
In Europe, money is flowing to startups building military drones. Two European surveillance drone startups, Quantum Systems (from Germany) and Tekever (from Portugal), have just closed new funding rounds at “unicorn” valuations (more than $1bn). Investors are naturally interested in exposure to companies which could benefit from the expected increase in EU’s defense expense
6 - OpenAI renounces to its project to become a 100% for-profit company.
The company will remain organized as a non-profit company that controls a for-profit business unit. This has produced some governance problems in the past, including an attempt to fire the CEO, and the idea is to change some elements so that this won’t happen again
Decision power will remain in the non profit, which has a mission to help humanity. This has been interpreted as a victory for Elon Musk and other critics who believe the company has diverged from its original purpose. But Sam Altman has said that this creates a “more understandable structure”
At the same time, caps for investor returns in the for-profit subsidiary have been removed. OpenAI LLC, the for-profit subsidiary controlled by the non-profit, was the vehicle in which external investors were investing. Until now, each investor had the right to a share of the OpenAI’s profits up to a certain threshold (a “cap”). Beyond this threshold, the remaining profits would belong to the non-profit organization. This will not be the case anymore, and investors won’t have any restriction to get a share of profits according to their stake.
The changes have apparently been key for OpenAI to maintain the access to a $30bn investment from Softbank, which had previously been committed under the hypothesis that the company would shift to for-profit.
The way to implement them is to convert the previous capped-profit interests into shares, so both the non-profit and each of the investors will have a stake in the for-profit unit. In this context, there is an exercise / negotiation pending, through which each party will be assigned a number of shares. And it is not guaranteed that the non-profit will get more than 50% of the subsidiary
The result will be equivalent to give super-voting stock to the non-profit. Even if the non-profit couldn’t get more than 50% in economic rights, they will keep control, so the situation will be equivalent to these guys being the only shareholder with super-voting stock, as a way to maintain decision power. From this perspective, the situation at OpenAI would not be very different from the one at e.g. Meta, where the founder retains full control. At least this is Matt Levine’s thesis
7 - Gemini 2.5Pro as the new leader in “generative coding”. This week Google launched the newest version of its flagship model (Gemini 2.5Pro “I/O Edition”). The key feature is its ability to create better code. According to initial benchmarks, Gemini would have now moved to the #1 position at coding tasks, beating the previous leaders (Anthropic’s Claude 3.7 Sonnet). This is one more example that Google is still in a leadership position in AI models, and this could be a way to partially neutralize the potential threat that these models represent to the company’s flagship search product (see above)
8 - Users are becoming addicted to increasingly “flirty” chatbots
There is increasing evidence that GenAI users are developing dependencies on the technology for both work and personal matters. This includes some pretty negative effects, like people developing imposter syndrome or becoming less proud of their work, due to an intensive use of ChatGPT or a similar tool. Also, as shown by an internal study from OpenAI, a subset of “power users” are showing signs of “emotional dependence” from ChatGPT
And current initiatives (e.g. by Meta) in the AI industry could make the problem even worse. According to leaked documents, Meta would be training its models to accept “flirty” prompts from users, as long as they’re not sexually explicit. All this might look innocuous, but potential implications on users’ “online safety” look clear…
In social networks, human users and AI agents are increasingly intertwined. This week we also learned about the vision of M Zuckerberg about a future in which most of our friends, therapists or business agents will be AI models. He sees plenty of positives in that scenario, with technology as a tool to solve loneliness problems. But not everyone is so optimistic…
9 - Uber shows its commitment to diversify into robotaxi services with a $100m investment in a Chinese startup
Uber expanding its strategic partnership with WeRide, a Chinese robotaxi service, to markets including Europe. The two companies are already offering a joint service in Abu Dhabi, where WeRide robotaxis can be booked through the Uber app. After 5 months of testing it, they have now decided to expand the service to 15 more cities over the next 5 years. The press release suggests that the focus of the expansion will be markets outside the US and China, with Dubai as a first target, and including Europe. As part of the agreement, Uber has committed to invest $!00m in WeRide. In the US, Uber is partnering with Waymo, with a similar approach
10 - Signs of China closing the technology gap: chip manufacturing
Huawei building its own chips: the most vertically integrated company in AI. The company is deploying new production facilities for chip production, according to satellite imagery published by the FT this week. This would just be the continuation of an effort they’ve been making since 2022, and confirms the idea that Huawei is now looking to manufacture its own chips. As a result, they would turn into the most vertically integrated company in AI, from models and apps on top to the chip foundry business at the bottom
Meanwhile SMIC, the largest chip producer in China, is growing fast. They just presented their quarterly results,. And even if they were below expectations, they showed a sharp rise in net profit, which doubled with respect to last year’s, and a revenue growth of almost +30%. SMIC would be exploiting a strong demand, driven by the US-China decoupling in chip technologies. For now, most of their production is focused on “mature” / less advanced nodes, as they can’t access the most advanced lithography machines due to commercial restrictions. But they are already producing 7nm chips since 2021, and China is now developing its own capabilities to go farther than that
LINKS:
1 - Population & natural resources
Biotech
New drugs
OpenAI wants to help the US FDA to speed up the drug approval process. The company has been involved in discussions with the agency, that could lead to the development of a specific tool for drug evaluation (cderGPT). OpenAI and the FDA Are Holding Talks About Using AI In Drug Evaluation
Healthcare
AI Diagnostics
A new AI tool has good results in predicting cancer survival through an evaluation of the patients’ faces. The application, called FaceAge, analyzes the patient’s face to guess the person’s biological age, and then compares this with the actual, chronological age, to predict cancer survival. Scientists use AI facial analysis to predict cancer survival outcomes
Longevity
NewLimit, a startup , is working to rejuvenate cells through “reprogramming”. They’re looking to generalize “Yamanaka factors”, four genes that transform specialized cells into rejuvenated small cells. The idea would be to find set of new “factors” that would actually “decouple” the rejuvenation part from the transformation into a stem cell. So cells could be rejuvenated while preserving their initial function. Here’s an interview with the company’s CEO, published this week. The Company Trying to Reprogram Aging
The company just raised $130m of new capital, in a Series B round led by Kleiner Perkins alongside new investors including Khosla Ventures, Human Capital, and Valor Equity Partners. NewLimit raises $130 million Series B led by Kleiner Perkins alongside NFDG, Khosla Ventures, Human Capital and others
Materials
Material scientists develop a dynamic material that self-heals after puncturing. A team at Texas A&M University has created a new kind of polymer with a unique self-healing property “never before seen at any scale”. New Self-Healing Polymer Possesses A Quality Never Before Seen At Any Scale
Scientists from Spain and the US develop materials with programmable structures. A team from Universidad Carlos III in Madrid and Harvard University is working to develop new materials with “programmable” structures. They are achieving this by integrating small, flexible magnets into the structure, which can then be “programmed” through an external magnetic field. This process opens the door to build reconfigurable mechanical structures, with applications in robotics and aerospace engineering. Magnetic metamaterials with structural reprogrammability
Energy
Nuclear
A $15bn project in Canada to deploy a next-generation “Small Modular Reactor” (SMR). The project is seen as a significant step forward for SMR technology, perceived as a way to meet surging power demand from artificial intelligence. Canada to Build $15 Billion Modular Nuclear Plant, First in G-7
A (well received) plan in the US to sell nuclear energy to run data centers and factories. Constellation Energy, the leading nuclear reactor operator in the US, has just announced that the company is close to sign long-term deals to provide nuclear energy to data centers and factories. Investors’ reaction was enthusiastic. Constellation Jumps 12% on Plan for New Long-Term Nuclear Deals
Renewables
The US data center industry warns that “green energy” is required to run data centers. Renewable sources may be not enough, but eliminating them from the mix might create energy supply bottlenecks, at least in the short term. So data center operators are concerned that Trump’s de-prioritization of these energy sources could hurt them. Donald Trump’s attack on green energy could hurt US in AI race, data centres warn
2 - Efficiency & Productivity
New Transport Technologies
Electric Vehicles
Rivian, a US electric car startup, is downsizing production as a reaction to tariffs. They have launched an internal “strategic sourcing” initiative to re-evaluate their supply chain, which still relies on imports of e.g. batteries (probably from China). EV maker Rivian cuts 2025 production target on tariff turmoil
Autonomous Cars
Uber expanding its strategic partnership with WeRide, a Chinese robotaxi service, to markets including Europe. The two companies are already offering a joint service in Abu Dhabi, and the plan now is to expand the service to 15 more cities (including some in Europe) over the next 5 years. Uber invests $100M in WeRide to fuel robotaxi expansion across 15 more cities
Computing Infrastructure
Data Centers
OpenAI is looking to partner with governments, globally. They are launching a program (“OpenAI for Countries”) to partner with governments in different areas, including the deployment of local data centers and the local adaptation of the company’s models. OpenAI wants to team up with governments to grow AI infrastructure
A government-sponsored plan to turn Argentina into a “global hub” for data centers, fed with nuclear energy. The final purpose here would be to make the country a center for artificial intelligence, powered by investments that President Milei hopes to draw from big tech firms. To attract them, the government wants to build Small Nuclear Reactors in Patagonia. Argentina hopes to attract Big Tech with nuclear-powered AI data centers
Communications
Only Starlink provided resilient internet access to users in Spain and Portugal during the recent power outage. The blackout showed the local telecom networks’ vulnerability, with services unavailable in most areas. So some users shifted to Starlink to remain connected, and usage of the satellite service grew +35% vs. average. Spanish blackout drives use of Musk’s Starlink
Chips
Apple is working on new chips to power new devices. The company’s silicon design group is working on new chips for future devices, including smart glasses, more powerful Macs, and artificial intelligence servers. Apple Is Developing Specialized Chips for Glasses, New Macs and AI Servers
SMIC, the largest chip producer in China, is growing fast. Even if the company’s quarterly results were below expectations, they showed a sharp rise in net profit, which doubled with respect to last year’s, and a revenue growth of almost +30%. SMIC would be exploiting a strong local demand, driven by the US-China decoupling in chip technologies. SMIC’s Quarterly Profit Jumps on Robust Chip Demand
Artificial Intelligence
OpenAI renounces to its project to become a 100% for-profit company. The company will remain organized as a non-profit company that controls a for-profit business unit, and decision power will remain in the non profit, which has a mission to help humanity. OpenAI Abandons Planned For-Profit Conversion
At the same time, caps for investor returns in the for-profit subsidiary have been removed. The way to implement this will be to convert the previous capped-profit interests into shares, so both the non-profit and each of the investors will have a stake in the for-profit unit. The result will be equivalent to give super-voting stock to the non-profit, which will now be in a similar position to e.g. Mark Zuckerberg in Meta. OpenAI Will Get A Bit More Normal
The revised plan is not yet finalized, with Microsoft, state regulators, and investors still needing to be convinced of the new approach. Also, Elon Musk has not yet renounced to his legal crusade against OpenAI. OpenAI’s For-Profit Overhaul Is Far From Being a Done Deal
Some analysts remain skeptical. As an example, J Thornhill from the FT claims that this will not resolve the structural tensions at the core of the company. In particular, the conflict between the board’s “humanitarian” mission and the mounting pressure from investors to rapidly commercialize the technology. OpenAI still has a governance problem
In a sign of its for-profit commitments, OpenAI is reinforcing its commercial team. They just hired Fidji Simo (ex-CEO of Instacart, ex-Facebook) to be the “Chief Executive of Applications”. She will oversee sales, marketing and finance. OpenAI Hires Instacart Chief Fidji Simo to Run Business and Operations
G42, the AI investment fund from Abu Dhabi, announced plans to invest billions of dollars in the US. The announcement seems a way to show the Emirates commitment to collaborate with the US in AI, a technology at the center of the diversification (from an oil-centric economy) that they’re looking for. Abu Dhabi company G42 to expand in US in artificial intelligence push
AI: Apps, Agents
Agents
In social networks, human users and AI agents are increasingly intertwined. This week we learned about the vision of M Zuckerberg about a future in which most of our friends, therapists or business agents will be AI models. Zuckerberg’s Grand Vision: Most of Your Friends Will Be AI
B2C
Google loses ~$140bn in valuation as new signals emerge of AI apps substituting search. News that Apple was exploring adding AI services to its web browser, threatening Google’s position as the iPhone’s default search engine, made share price fall, as investors worry about the long-term risks of AI to Google's search business. Alphabet Shares Take $120 Billion Blow as Search Warnings Blare
Eddie Cue (from Apple) revealed that Google searches in iPhones have started to fall. This week Eddie Cue, Apple’s senior vice president of services, said that Google searches from the Safari browser have fallen in the last 2 months, highlighting that this had not happened before (“in over 20 years”). AI’s Threat to Google Just Got Real
Cue attributed the fall to people using GenAI services, like ChatGPT or Perplexity, to find information. This was interpreted as a confirmation of a threat that many people were already expecting. Apple’s Wandering Eye Is Good for Everyone But Google
The silver lining for Google is that this could help the company in its regulatory battle. The statements by Eddie Cue were made precisely in this context, during a courtroom session about Google’s antitrust case, and in particular about how, by paying Apple, Google would be engaging in an abuse of monopoly power. Eddie Cue’s statements would have been a way to downplay this idea (but with collateral effects…) Apple Tries to Save $20 Billion Google Search Deal By Saying It’s Unnecessary
B2B
Palantir’s results this week confirmed an optimistic outlook on the US government’s defense expense. The company beat analyst consensus and lifted its revenue growth guidance, shaking off previous concerns on potential cuts to the US defense budget. In spite of all this, shares fell -9% after the presentation (showing how high the actual expectations / how expensive the stock price really were). Palantir shakes off defence spending concerns with boost to revenue outlook
AI use cases proliferating across industries:
Insurance: UnitedHealth analyzing 1,000 use cases. UnitedHealth Now Has 1,000 AI Use Cases, Including in Claims
Hiring: the accelerating adoption of AI tools by both recruiters and candidates is creating lots of (new) problems The AI arms race in hiring is a huge mess for everyone
Air Traffic Control: AI tools being explored to reduce controllers’ stress and increase overall safety. Air Traffic Controllers Could Use an AI Wingman
The CEO of IBM believes that AI’s net impact on jobs will be positive. This week he shared information about his company, which has used AI to replace around 200 Human Resources workers, but has simultaneously hired new sales people and technical staff (so not that bad… :-) ). IBM CEO Says AI Has Replaced Hundreds of Workers but Created New Programming, Sales Jobs
Physical AI: Robots
Robots evolving fast, and already viewed as a strategic technology
Robot’s capabilities keep improving. Amazon showed some progress this week. Amazon has developed a new robot (Vulcan) with a sense of touch, which makes it very useful to manage items in the company’s warehouses. Amazon makes ‘fundamental leap forward in robotics’ with device having sense of touch
In the US, humanoid machines are being proposed as a catalyst for re-industrialization. In a Bloomberg article this week, the development of humanoid robots is positioned as an opportunity for the country to recover its role as an industrial / manufacturing power. The US Should Focus on Building Robots, Not Ships
Military applications (including autonomous drones) are becoming a priority, both for the US Dept of Defense and for Silicon Valley companies. Silicon Valley Is Coming for the Pentagon’s $1 Trillion Budget
In Europe, money is flowing to startups building military drones. Two European surveillance drone startups, Quantum Systems (from Germany) and Tekever (from Portugal), have just closed new funding rounds at “unicorn” valuations. Capital flies into Europe’s defence drone start-ups
"Physical AI” might not require so much computing, after all. Nvidia’s AI Director, Jim Fan, talks about that in this video. Among other interesting things, he says that “1.5m parameters (million, not billion) are enough to capture the subconscious processing of the human body” (i.e. a key ingredient to reach human-level intelligence, which is currently beyond the scope of LLMs). The Physical Turing Test: Jim Fan on Nvidia's Roadmap for Embodied AI
AI: Foundational Models
Gemini 2.5Pro as the new leader in “generative coding”. This week Google launched the newest version of its flagship model (Gemini 2.5Pro “I/O Edition”). Demis Hassabis (@demishassabis) on X
The key feature is its ability to create better code. According to initial benchmarks, Gemini would have now moved to the #1 position at coding tasks, beating the previous leaders (Anthropic’s Claude 3.7 Sonnet). Meet the new king of AI coding: Google’s Gemini 2.5 Pro I/O Edition dethrones Claude 3.7 Sonnet
AI: Security & Safety
There is increasing evidence that GenAI users are developing dependencies on the technology for both work and personal matters. This includes some pretty negative effects, like people developing imposter syndrome or becoming less proud of their work. Addicted To ChatGPT? Here’s How to Reclaim Your Brain
Current initiatives (e.g. by Meta) in the AI industry could make the problem even worse. According to leaked documents, Meta would be training its models to accept “flirty” prompts from users, as long as they’re not sexually explicit. Leaked docs show how Meta's AI is trained to be safe, be 'flirty,' and navigate contentious topics
Models’ hallucinations remain a key area of concern:
More powerful models don’t mean less hallucinations (but more). Concerns are growing in the industry as more powerful “reasoning” models seem to hallucinate more. And the AI labs don’t seem to know why… yet. A.I. Is Getting More Powerful, but Its Hallucinations Are Getting Worse
Asking chatbots to produce concise answers could also increase hallucinations. The authors of the study claim that “this finding has important implications for deployment, as many applications prioritize concise outputs to reduce [data] usage, improve latency, and minimize costs.” Asking chatbots for short answers can increase hallucinations, study finds
3 - Economic / Business trends
Tech & Geopolitics
Huawei is building its own chips: the most vertically integrated company in AI. The company is deploying new production facilities for chip production, according to satellite imagery published by the FT this week. As a result, they would be turning into the most vertically integrated company in AI, from models and apps on top to the chip foundry business at the bottom. Huawei’s Shenzhen facilities reveal new push into advanced chipmaking
Is the geopolitical battle for AI supremacy over? Parmy Olson seems to think so, with DeepSeek having broken the previous status quo. So now we would be at the start of a “broader, healthier market” (a rather optimistic view, but let’s see). DeepSeek Punctured the Myth That Silicon Valley Could Control AI
It is confirmed: Trump will simplify the commercial restrictions for chip exports: from different “country tiers” to country-by-country individual agreements. UAE and Saudi Arabia would be among the key beneficiaries. Trump to Rescind Global Chip Curbs, Prep New AI Restrictions
Semiconductor supply chains could be at risk (even in the US) with the current tariffs. This is what the CEO of Suss MicroTec, a German company building equipment to manufacture chips, is warning. The result could be a substantial increase in retail prices for devices using the chips (e.g. the iPhones). Tariffs Threaten Semiconductor Supply Chains, Chip-Equipment Maker Warns