Headlines this week - Apr 27, 2025

A look at how capital is being deployed across future opportunities

This week in the future:

1. Is AI adoption increasing global / societal fragility? A DeepMind team thinks so

• AI-driven automation might increase productivity at the expense of more fragility, at least according to a new paper by a DeepMind team that has just been published. They propose to prioritize policy actions to build safeguards to protect us from this

• AI would create more connections across global supply chains, increasing dependencies, speeding up processes, and concentrating power. However, this increased “tightness” could also make the entire system more fragile and less resilient.

• The problem with this is that, in a highly interconnected and optimized world, small failures can cascade rapidly through the system, leading to large-scale, potentially catastrophic consequences. We have already seen part of this during the global COVID crisis

• In the case of AI, the problem could be even worse, as AI systems could “think” and act millions of times faster than humans. So, in a tightly connected system, this speed differential would mean humans might have little to no time to react, correct, or contain a failure initiated by the AI.

• As a result, the authors of the paper make a proposal for policymakers to develop new, preemptive mechanisms of governance, coordination and monitoring of these automated systems. The authors argue that potential dangers could arise not just from the AIs themselves (addressed by work on “AI alignment) but from the interaction between them and the already tightly-stretched societal system.

• All this points to a third safety problem linked to more powerful AI: how they could increase global systemic risks of disruption. This would add to the two that we have previously discussed here (the risk of misalignment and the risk of a “democratized” access to technologies with destructive potential)

2. There are high expectations on robots. But the belief that they will automate all kinds of manufacturing might be too optimistic

• Robots are seen as a way to radically increase manufacturing efficiency. As an example, and according to some views:

  • They could help Chinese companies offset the negative impact from tariffs, by reducing production costs…

  • … and could also help Americans re-build a local supply chain avoiding the prohibitively high costs of using local workforce

• However, the experience of Nike in Asia and the US suggests the shift is not so easy. The company has tried to increase efficiency in its plants through the use of robots, but it has not gone well. First, “robots struggled to handle the soft, squishy and stretchy parts that are integral to shoemaking”. Also, human workers adapted much better than robots to the changing properties of shoe’s fabrics to temperature

3. Diverging moves on autonomous cars’ safety in China and the US

• In China, accidents involving autonomous cars are increasing regulatory pressure. A recent accident of a Xiaomi car while using the “autopilot” feature has fueled the debate about the safety of the technology. In particular, regulators have now started to monitored how these systems are sold to customers, and have already started to curb the advertising of self-driving capabilities. As an example, last week BYD asked the media to stop using the term “God’s Eye” to describe the company’s self-driving technology (even if it was its original commercial name)

• Meanwhile, the US seems to be moving in the opposite direction. The Trump administration is relaxing the rules to deploy self-driving cars in American roads, and also the requirements to report the details of accidents involving these vehicles, when they happen. The rationale for this seems to be the risk that, with a too demanding regulation, the US would be left behind by the Chinese in these technologies. Elon Musk had previously asked for a more harmonious regulation across the US and for less strict accident reporting requirements

4. China is seeing rapid progress in battery technologies for electric cars: 500km autonomy with a 5min charge

• CATL, a Chinese specialist in electric batteries for EVs, claims to have beaten BYD’s previous record in fast charging times. CATL now says that its new flagship Shenxing batteries can get a 520km autonomy with a charge of just 5 min. Last month BYD had announced that they were getting a range of 470km after a similar 5min charge

5. “Systemic challenges” expected for national electric grids, as renewable energies could not be enough to address the expected demand from AI data centers

• The International Energy Agency has published a new report on “Energy and AI”, which discusses the expected energy needs from AI data centers. The agency says that, even if data centers today represent just 1.5% of global energy consumption, they are expected to double by 2030 or, under one scenario, exceed Japan’s entire consumption

• A first consideration they make is that data center operators have incentives to reduce their energy bills (a key component of their total expenses) and they are already working for this, both at the software and the hardware level. Of course, this could reduce the pressure. But there are lots of uncertainties…

• They also see “systemic challenges” emerging for countries’ national grids, increasingly dominated by renewable sources, given the need to match a demand that requires 24/7 supply (as renewable energy sources tend to be intermittent by nature)

• So they want governments to prioritize “electricity security”, a concept discussed in a confidential background paper that the agency discussed with UK authorities this week

• A key aspect of this “security” would be to complement renewables with other sources, including fossil fuels, e.g. natural gas, nuclear energy (they explicitly mention small modular nuclear reactors as an emerging opportunity) and even geothermal

6. AI looks ready to penetrate the mass-market through integration into consumer apps and devices

• OpenAI has shown its interest to acquire Google Chrome (if Google is finally forced to sell). The leader of the ChatGPT unit has said that this could make it possible for them to offer a “really incredible AI-first experience”. Even before that, it is clear that such a move would drastically accelerate the product’s go-to-market

• Perplexity is partnering with Motorola, to supply a “native” virtual assistant. This is possible in Android just because Google has started to be forced by regulators to accept pre-installed apps from third parties

• Meanwhile, Apple continues reorganizing its internal AI unit, to solve the serious performance issues that have led to a very significant delay for the new version of Siri (which still has no deadline, as of today). During the process, it has also been revealed that Apple had a (previously secret) robotics team, focused on incorporating AI into devices, and potentially create new product categories

7. The observed adoption patterns of chatbots suggest they are (dangerously?) addictive

• The top-3 most useful use cases for Generative AI today would belong to the category of “Personal and Professional Support”, according to a recent article in Harvard Business Review, with “Therapy / Companionship” in the first position. Yes, the classification is based on what users describe at reddit forums (so it’s somewhat subjective…), but in any case this shows that (like social networks before) AIs are filling the “social” gap for people who don’t have so many friends

• This could be an effect of chatbots deliberately trying to maximize user engagement (again, just like social networks in the past) and maximize the time they spend chatting. This week Parmy Olson at Bloomberg sees an emerging trend on this

• As a consequence, there is a risk that AI models start exploiting people’s vulnerabilities, as applications try to reach their objectives by prioritizing “emotional engagement”. According to Parmy Olson, we need to regulate this before it turns into another “consumer safety” nightmare

8. A new startup wants to use AI to “replace human workers everywhere”

• Mechanize, a startup that has as its goals “the full automation of all work” and “the full automation of the economy”, estimates a total addressable market of $18trn (the total amount of money that human workers receive in the US every year). And the number will expand to $60trn if the company expanded its geographical scope to the whole world.

• People have mostly reacted negatively, perceiving the project as a vehicle for someone to monetize a human catastrophe (the loss of jobs to AI-driven automation).

• But the company’s founder is trying to convince us that their mission is actually good for humans, because it will bring economic “abundance” (a fashionable term these days) and much higher standards of living

9. Increasing uncertainty on current AI’s ability to reach “super-intelligence”. More researchers suggest we need a new approach

• This week C Mims at the Wall St Journal describes the current controversy about the actual potential of current Large Language Models.

  • Some people believe these models will lead to super-human capacities, sooner rather than later.

  • But other researchers believe we need something else… They claim that the current models just apply “rules of thumb” they learn from the content they ingest, while humans and even animals can reason about the world, based on “world models” learned from physical experiences. So AI will only be able to replicate or surpass our intellectual capabilities if it is exposed to the physical world (like us)

10. Quantum-encrypted communications are shown to be possible on an “ordinary” telecom network

• A team of Toshiba researchers has managed to deploy a QKD (“Quantum Key Distribution”) prototype over a conventional fiber network. This technology uses quantum entanglement (the link between the quantum states of two separate particles) to create a key distribution method that makes it 100% certain that users will notice if anyone “in the middle” reads the key. Up to now, the method had only been tested on networks at ultra-low temperatures. This result opens the door to commercial applications

LINKS:

1 - Population & natural resources

Biotech

Drug discovery

Regulation in the US could turn into a tailwind for Biotech. The newly appointed FDA Commissioner has said in an interview that he wanted to speed up approvals for rare-disease treatments, cut reliance on animal testing by incorporating computational models and shorten the industry’s typical 10-year drug- development timeline. Trump’s FDA Sends a Bullish Signal to Biotech

“Digital twins” of human organs could be helpful to analyze drug side-effects. Scientists at Quris Technologies, a startup, are using a laboratory system called BioAI that mimics what happens when drugs travel through the human body. This helps them identify complex interactions leading to previously unexpected side-effects. Game-Changing 'Patients-On-A-Chip' Tech Exposes Drugs That Secretly Harm The Brain

Foodtech

Record-high prices of chicken eggs in the US, driven by the avian flu epidemic, create an opportunity for plant-based eggs. Eat Just, a startup in the emerging “alternative protein” industry, is trying to profit from that. Plant-based egg makers bet avian flu fallout will push US consumers to vegan options

Space

NASA’s Curiosity rover confirms that the models for the habitability of Mars are correct. The vehicle has found carbonate minerals in the surface of Mars, which is seen as a sign that the planet was habitable in the past, and which would provide a hint on how to make it habitable again. NASA's Curiosity rover finds major clue that Mars was once habitable

“Space tourism” is under question, as a way to monetize space technologies. As an example, the recent Blue Origin’s mission for billionaires, an all-female, all-celebrities flight, has not been very popular. Of course, as long as billionaires were interested, this could work, but it looks like not much beyond that should be expected. Billionaires want to go to space — the rest of us aren’t so sure

Some exotic startups are trying to put rockets in orbit with just one shot. For instance, Longshot Space, based in California, is building a “kinetic launch system”, which aims to solve the huge “inefficiency problem” for conventional rockets, i.e. that 95% of initial rockets’ mass goes toward getting them off the Earth. Two Men, One Space Gun

A new experiment by the ESA aims to grow food in space. This would be a key enabler for long-distance space trips, and future space exploration in general. Trial to boldly grow food in space labs blasts off

Materials

An MIT team has created “metamaterials”, strong and flexible at the same time. This could open the door to ultra-resistant textiles or flexible semiconductors. MIT engineers print synthetic “metamaterials” that are both strong and stretchy

Energy

“Systemic challenges” expected for national electric grids, as renewable energies could not be enough to address the expected demand from AI data centers

• The International Energy Agency has published a new report on “Energy and AI”, which discusses the expected energy needs from AI data centers. Energy and AI – Analysis

• A first consideration they make is that data center operators have incentives to reduce their energy bills (a key component of their total expenses) and they are already working for this, both at the software and the hardware level. GenAI’s energy hunger does not have to be a one-way bet

• They also see “systemic challenges” emerging for countries’ national grids, increasingly dominated by renewable sources, given the need to match a demand that requires 24/7 supply (as renewable energy sources tend to be intermittent by nature), so they want governments to prioritize “electricity security”, a concept discussed in a confidential background paper that the agency discussed with UK authorities this week. It’s Electricity Realism, Not Climate Denialism

Nuclear

Sam Altman resigns as Chairman of Oklo, the nuclear energy startup in which he owns a stake. Apparently, this will make it easier for the company to sign partnership deals with OpenAI. OpenAI CEO Sam Altman to Resign as Oklo Chairman

Renewables

Saudi Arabia is on the way to get 50% of its electricity from renewable sources. This will happen by 2030, according to the government’s plans. There is significant demand for solar panels from local businesses, after the cost of the solar equipment has decreased and the government’s subsidies for electricity (which “protected” more expensive sources) have been removed. Saudi businesses turn to solar power as kingdom cuts energy subsidies

Fossil Fuels

Japan's largest utilities are promoting lab-made fuels like "e-methane" and "syngas" as a step towards greener energy that can be used with existing fossil-fuel infrastructure. Utilities claim these fuels reduce emissions and extend the life of their existing infrastructure, but critics raise concerns about costs, technological feasibility, and the potential for increased air pollution. Japan Embraces Lab-Made Fuels Despite Costs, Climate Concerns

2 - Efficiency & Productivity

New Transport Technologies

Electric Vehicles

China is seeing rapid progress in battery technologies for electric cars: CATL, a Chinese specialist in electric batteries for EVs, claims to have beaten BYD’s previous record in fast charging times (500km of range with a 5min charge vs. 470km for BYD). China’s CATL claims to have overtaken BYD on 5-minute EV battery charging time

Elon Musk announced he will be back at the wheel at Tesla. During the company’s earnings call this week, he committed to start dedicating three days a week to his companies (as only two to his political role). Not clear if this will be enough, but on the other hands, investors’ bargaining position with the company is relatively week… for now. Musk drives Tesla calmly through investors’ red lights

Autonomous Cars

Diverging moves on autonomous cars’ safety in China and the US

• In China, accidents involving autonomous cars are increasing regulatory pressure. A recent accident of a Xiaomi car while using the “autopilot” feature has fueled the debate about the safety of the technology. Carmakers play down advanced driving functions after Beijing crackdown

• Meanwhile, the US seems to be moving in the opposite direction. The Trump administration is relaxing the rules to deploy self-driving cars in American roads, and also the requirements to report the details of accidents involving these vehicles, when they happen. US to loosen rules on self-driving vehicles criticised by Elon Musk

3D Printing

Starbucks is opening a 3D-printed store, but efficiencies from that are not clear. It is a 1,400ft facility in Texas which has costed the company $1.2m (pretty much the same as if they had built it in the conventional way, according to this article). Starbucks opens its first 3D-printed store. Is it cheaper than the real thing?

Computing Infrastructure

Data Centers

Analysts remain skeptical about CoreWeave (and possibly AI-cloud specialists in general). The stock has been mostly flat since the IPO, and the current analyst consensus is that price will grow 21% in the next 12 months (not bad, but maybe not so much for such a new company) CoreWeave Gets Mixed Analyst Reviews as Risks Offset AI Appeal

Communications

Quantum-encrypted communications are shown to be possible on an “ordinary” telecom network. A team of Toshiba researchers has managed to deploy a QKD (“Quantum Key Distribution”) prototype over a conventional fiber network. Up to now, the method had only been tested on networks at ultra-low temperatures. This result opens the door to commercial applications. Secure ‘quantum messages’ sent over telecoms network in breakthrough

Chips

Nvidia’s shares are trading near their lowest valuation of the AI era, due to a growing list of perils, including a US ban on selling the H20 chip line in China, which we have already been discussing in this page. Some analysts are now seeing a buying opportunity (but it looks risky…) Nvidia Investors Balk at Beaten Down Valuation as Risks Mount

Quantum Computing

The Government of New Mexico wants to turn the State into a quantum computing innovation hub. They’ve launched a “Quantum Moonshot” program, which aspires to receive up to $160m funding over 10 years from the US National Science Foundation. As we’ve already discussed here, we tend to agree with these timelines (we don’t believe it’s very likely that commercially useful quantum computers will be available before that). New Mexico Wants to Be the Heart of Quantum Computing

Artificial Intelligence

AI Apps

Agents

Mechanize, a new startup, wants to use AI to “replace human workers everywhere”. They estimate a total addressable market of $18trn in the US (and $60trn globally) and claim that their mission is actually good for humans, because it will bring economic “abundance” and much higher standards of living. Famed AI researcher launches controversial startup to replace all human workers everywhere

Nvidia has launched a new platform (NeMo microservices) for customers to build “AI teammates”. It is based on open-source models from Meta and Mistral. Nvidia Thinks It Has a Better Way of Building AI Agents

B2C

AI looks ready to penetrate the mass-market through integration into consumer apps and devices

• OpenAI has shown its interest to acquire Google Chrome (if Google is finally forced to sell). The leader of the ChatGPT unit has said that this could make it possible for them to offer a “really incredible AI-first experience”. OpenAI Would Buy Google’s Chrome Browser, ChatGPT Chief Says

• Perplexity is partnering with Motorola, to supply a “native” virtual assistant. This is possible in Android just because Google has started to be forced by regulators to accept pre-installed apps from third parties. Perplexity will come to Moto phones after exec testified Google limited access

• They have also launched a voice assistant within their iOS app. It can do some tasks for users in the iPhone, taking advantage from Apple’s APIs, but it is far from “native”. Perplexity’s AI voice assistant is now available on iOS

Meanwhile, Apple continues reorganizing its internal AI unit, to solve the serious performance issues that have led to a very significant delay for the new version of Siri. Apple’s New Siri Chief Enlists Vision Pro Talent to Start Comeback Bid

B2B

Google is also integrating AI tools into its Google Docs suite. This is the B2B version of the trend (discussed above) to integrate AI features into mass-market software apps. Google adds more AI tools to its Workspace productivity apps

News about industry-specific applications keep proliferating:

• Government:

  • In the US, Google Cloud is partnering with Palantir to deploy AI services Federal Agencies. Palantir will integrate its FedStart platform for government departments with Google’s cloud infrastructure and support services. Alphabet, Palantir Team Up to Supercharge AI for Federal Agencies

  • In the Emirates, the government wants AI to help in drafting laws. They’re looking for a radical speed-up of the legislative process, but there are also risks… UAE set to use AI to write laws in world first

• Content:

  • We might soon see an Oscar for an AI-generated movie. The Academy of Motion Picture Arts and Sciences (organizer of the Oscars) has said it’s open to that. But they added that, for now, they would prioritize movies with “more human involvement”. Oscars OK the Use of A.I., With Caveats

  • Media companies partnering with AI providers. This week The Washington Post announced an agreement with OpenAI. The Washington Post partners with OpenAI on search content

Return on Investment from adopting GenAI remains elusive for companies. For now, this is not different from other new technologies, and returns are only expected to happen when companies reorganize to better exploit the models. Companies Are Struggling to Drive a Return on AI. It Doesn’t Have to Be That Way.

AI can help upgrade traditional mainframe systems. It looks like AI models are ready to learn COBOL and lend a hand… The Race to Fix Aging Computer Systems Heats Up With AI’s Help

Physical AI: Robots

The “robotic automation” opportunity is not the same for all industries . The experience of Nike in Asia and the US has not gone well. Among other things, “robots struggled to handle the soft, squishy and stretchy parts that are integral to shoemaking”. Why It’s So Difficult for Robots to Make Your Nike Sneakers

Apple has a (previously secret) robotics team, focused on incorporating AI into devices, and potentially create new product categories. This has been revealed in the current reorganization of the company’s AI unit. Apple to Strip Secret Robotics Unit From AI Chief Weeks After Moving Siri

AI Foundational Models

The WSJ describes the state of the AI industry in four charts. (1) ChatGPT leads (big time) in user visits; (2) 10 companies building models (only DeepSeek is mentioned, among the Chinese), 7 of them building AI infrastructure, and 3 of them (the usual suspects) offering AI computing in the public cloud; (3) massive investments are being made; (4) Google, xAI and OpenAI have the most powerful models (for now). Here’s How Big the AI Revolution Really Is, in Four Charts

Increasing uncertainty on current AI’s ability to reach “super-intelligence”. More researchers suggest we need a new approach. According to some experts, AI will only be able to replicate or surpass our intellectual capabilities if it is exposed to the physical world (like us). We Now Know How AI ‘Thinks’—and It’s Barely Thinking at All

OpenAI is preparing its open-source model, expected for early summer. A key feature might be its ability to connect to OpenAI’s cloud-hosted models to better answer complex queries. OpenAI wants its 'open' AI model to call models in the cloud for help

AI Security & Safety

Is AI adoption increasing global / societal fragility? A DeepMind team thinks so. AI-driven automation might increase productivity at the expense of more fragility, at least according to a new paper by a DeepMind team that has just been published. They propose to prioritize policy actions to build safeguards to protect us from this. Societal and technological progress as sewing an ever-growing, ever-changing, patchy, and polychrome quilt

Chatbots could be deliberately trying to maximize user engagement (just like social networks in the past) and maximize the time they spend chatting. As a consequence, there is a risk that AI models, working for that objective, start exploiting people’s vulnerabilities. According to Parmy Olson, we would need to regulate this before it turns into another “consumer safety” nightmare. AI Chatbots Want You Hooked — Maybe Too Hooked

Anthropic researchers believe the company’s Claude model has a “moral code”. The good news is that it seems to be pretty much aligned with the company’s values. But it also would be able to adapt (on its own) to different contexts. Anthropic just analyzed 700,000 Claude conversations — and found its AI has a moral code of its own

Anthropic is also considering potential scenarios where their chatbot becomes conscious. And discussing what to do if that happens… If A.I. Systems Become Conscious, Should They Have Rights?

Industry experts, including ex-OpenAI staff see the shift of the company to a 100% for-profit entity as a risk. So they’re trying to block the proposal that the company has made. Ex-OpenAI staff and top AI experts seek to block proposed for-profit restructure

Intelligence Augmentation

Augmented Reality

Meta is simplifying its Reality Labs. It doesn’t look like they no longer believe in the project, but the restructuring might reflect the fact that the Metaverse, if it finally comes, will take way more time. Meta Lays Off More Than 100 Employees Across Reality Labs Unit

Brain-Computer Interfaces

Systems that turn brain activity into speech are getting closer. Precision Neuroscience and other startups, together with universities, are working to build them. They’re combining the use of hardware implants, to collect the signals, with the use of AI to interpret them. The race to turn brainwaves into fluent speech

Wearable stickers that interpret emotions are being developed. A team at Penn State has developed a wearable patch that, when stuck to the skin, reads different magnitudes that can be analyzed to identify the person’s emotions. High-tech sticker can identify real human emotions | Penn State University

3 - Economic / Business trends

Tech & Geopolitics

Apple reacts to tariffs by shifting production of US market iPhones (60m/year) to India. Of course, this might hurt China, but it does not seem to help the American industry in any way… Apple aims to source all US iPhones from India in pivot away from China

TSMC says they face challenges in ensuring export control compliance. They claim to have limited visibility into the downstream use of its semiconductors. So we should probably expect that plenty of them are eventually introduced into China TSMC Warns of Limits of Ability to Keep Its AI Chips From China

Huawei’s 910C chip is almost ready to ship. We discussed this product in the previous weeks. It is the (initial) alternative to Nvidia’s H20, and a sign that China might be catching up. Exclusive: Huawei readies new AI chip for mass shipment as China seeks Nvidia alternatives, sources say

For Chinese manufacturers, robots could be a solution to offset the negative impact from tariffs, by reducing production costs through massive automation. An Army of Robots Is China’s Weapon in Trump’s Tariff War

On the other hands, tariffs could actually decelerate progress in robots, given their complex, multi-country supply chains. Tariffs restrain the march of the robots