Headlines this week - Oct 5, 2025
A look at how capital is being deployed across future opportunities
This week in the future:
1 - Everyone agrees that “world models” will enable the next generation of AI, but not on how to build them
“World models” have become the new frontier in the race for super-intelligence. Major AI labs are now betting heavily on “world models”—simulations that help AI understand cause and effect—as the key to their next big breakthroughs. This new focus marks a significant shift in the industry’s approach to achieving AGI.
However, a key pioneer argues that the LLMs currently used are a “dead end” for this task. In a recent interview, Reinforcement Learning pioneer Richard Sutton recently argued that a new architecture, based on “raw” learning from experience, is needed to build true world models. This represents a direct challenge to the current dominance of large language models in the field.
This has sparked a new debate on how to effectively implement “world models”. An essay by DeepMind senior researcher Andrew Trask highlights the controversy sparked by Sutton’s argument. The debate centers on whether human-like knowledge and architecture are necessary for progress, or if pure computational scale is the only thing that matters in the long run. This is an extension of Sutton’s previous (and very well known) arguments in his now old “Bitter Lesson” paper.
2 - A key element in the “world paradigm” will be the tools to directly access “real world” data. Robots could compete with wearables for this
A new front could open between Musk and Zuckerberg over “real world” data collection. The WSJ highlights an emerging rivalry between Elon Musk’s humanoid robots and Mark Zuckerberg’s AI glasses. Both are seen as competing approaches to capture the vast amounts of real-world data needed to train the next generation of “world models”.
Humanoid robots remain a hot space, with Nvidia as a key driver. The humanoid robot field continues to heat up, with Nvidia and Fujitsu recently announcing a new collaboration to work on AI-powered robots. This partnership underscores the growing momentum and investment in the space, with Nvidia’s hardware playing a central role.
The rise of humanoid robots is also creating new geopolitical challenges. A Bloomberg analysis warns that the US needs to take control of its own humanoid robot future to avoid ceding a critical industry to China. The article argues that leadership in this field will be essential for both economic and national security. Among other things, as we’ve said, data from these devices is expected to become a key asset to build the next wave of AI models
3 - Consensus seems to grow about the AI race having turned into an investment bubble. But this might not be so bad as it seems…
We may be approaching the “capex endgame” of the current AI boom, but the “bubble” won’t necessarily be bad. An FT analysis argues that the AI boom is following a classic bubble pattern, where massive over-investment in infrastructure will lead to a bust. While this may not pay off for current investors, the resulting excess capacity will ultimately benefit society by making AI cheap and ubiquitous.
The recent Nvidia-OpenAI deal is being compared to the dot-com bubble. Veteran tech investor James Anderson has warned that Nvidia’s planned $100bn investment in OpenAI brings “uncomfortable echoes of the dotcom bubble”. His key concern is that the recent surge in AI valuations is becoming disconnected from fundamentals.
The investment boom itself may be driving a large part of current GDP growth. The FT’s “Unhedged” column questions whether economic growth today would be near zero without the massive capital expenditure on AI that we’re seeing. The article highlights that investment in AI infrastructure accounted for at least half of US GDP growth in the first half of the year.
The increasing use of debt to fuel the boom adds a new layer of risk. A Heard on the Street column at the WSJ warns that debt is “fueling the next wave of the AI boom,” a dangerous trend that combines financial and operational leverage. According to historical precedents, this could amplify the negative effects of a potential bust, creating a more severe downturn.
But the momentum continues, with even conservative investors making huge bets. Despite the bubble fears, major players are still making massive investments in AI infrastructure. An example this week was Bloomberg reporting that BlackRock’s GIP is nearing a deal to acquire Aligned Data Centers (a tech infrastructure company) for about $40bn. This shows that even the more conservative pools of capital are now entering the AI infrastructure race.
4 - In this context, AI labs are in an urgent need to monetize, both with consumers and businesses
OpenAI’s massive spending creates an urgent need for revenue. An FT analysis highlights OpenAI’s “era-defining money furnace,” with operating losses reaching $7.8bn in the first half of the year. This massive cash burn puts immense pressure on the company to find a viable path to profitability.
The company is now pursuing “classic” consumer monetization strategies:
A new Sora video app is positioned to compete with TikTok and YouTube. This week OpenAI has launched a new app for its Sora video generator, a direct challenge to established social media platforms. The move signals a clear ambition to capture a large consumer audience.
A dedicated social app for AI-generated content may be next. OpenAI would also be preparing to launch a standalone social app for AI-generated videos, which would likely be monetized through advertising. This would be another major step into the consumer social media space.
There are intellectual property challenges for this project. The company has actually announced that the new model will require content owners to “opt out” if they don’t want to participate. But this is (obviously) controversial
E-commerce features are now being integrated directly into ChatGPT. Also, OpenAI is now allowing users to buy products directly through ChatGPT, a significant move into e-commerce. This new feature lays the groundwork for a future of AI agent-based shopping (and also for trying to capture advertising revenue).
In the business segment, a talent shortage is the key bottleneck:
Despite a seemingly large pool of tech talent, companies are struggling to find workers. with the specific skills needed to implement AI. This is creating a major bottleneck for enterprise AI adoption.
Walmart, for example, needs more developers to build and deploy AI agents. The WSJ reports that Walmart is actively hiring more software engineers to support its push into AI-powered automation. The retail giant is not just adopting AI, but building the internal teams needed to create and manage a new generation of AI agents.
5 - New data points supporting the thesis of an “AI Energy Crisis”, while the “nuclear energy renaissance” accelerates (as part of the solution)
Power bills are soaring in the US, driven by a boom in AI data center demand. A new analysis shows that wholesale electricity prices have surged in areas with significant data center activity, passing the costs on to consumers. This trend is directly linked to the massive energy consumption of the rapidly expanding AI industry.
By 2030, AI’s energy needs could equal the output of 44 nuclear reactors. IEEE Spectrum estimates that by the end of the decade, AI’s electricity consumption could reach a staggering 337 terawatt-hours annually (equivalent to the power of 44 conventional nuclear reactors). This massive demand is creating a potential bottleneck for future progress in the field.
This is fueling a projected $350bn nuclear boom in the US alone. To meet this demand, investors are expected to pour $350bn into new nuclear power projects in the US by 2050. According to a Bloomberg article this week, this investment will be driven almost entirely by the power-hungry data centers running AI systems.
Next-generation nuclear technologies like Small Modular Reactors (SMRs) will be a key part of the solution. SMR startups are moving to meet this new demand. This week Oklo revealed they’re targeting a mid-2026 launch for their first US reactor. This would be part of a broader push to accelerate the deployment of advanced nuclear technologies.
As we’ve already discussed, a critical challenge will be securing the nuclear fuel supply chain. The US is also working to reduce its reliance on Russian nuclear fuel, with suppliers like Urenco now receiving approval to produce more powerful fuel domestically. This move is seen as a key step in building a secure, independent supply chain to support the nuclear renaissance.
6 - A new breakthrough in genomics: building embryos from human skin cells
Researchers have created early-stage human embryos using DNA from skin cells. In a major breakthrough, scientists have successfully fertilized a human egg using DNA taken from skin cells. The technique, which involves placing the skin cell’s nucleus into a donor egg stripped of its own genetic material, has so far produced a small number of early-stage embryos.
This could one day offer a new path to fertility for women without viable eggs. The new technique could eventually provide an alternative to IVF for women who cannot produce their own eggs, according to The Economist. While practical treatments are still likely a decade away, the research opens up new possibilities for overcoming certain forms of infertility.
In other genomics news, a massive project is underway to sequence all life on Earth. The Earth BioGenome Project is a monumental effort that looks to sequence the genome of all 1.8m known species on the planet. This vast library of genetic information is expected to revolutionize our understanding of life and could accelerate the discovery of new health treatments.
7 - On the negative side of biotechnology, bioweapon risks could be growing
A flaw in security software has exposed the growing risk of AI-powered bioweapons. This week we learned how researchers discovered a “striking vulnerability” in the software designed to prevent the creation of dangerous biological agents. The flaw, now patched, highlights the urgent need to address the rising threat of bioterrorism as AI accelerates advances in synthetic biology.
8 - There are big expectations about how AI could accelerate the discovery of new materials
A new “golden age” of materials science is being driven by AI and robotics. This week Andrew Cote described in a post how, according to his views, the combination of AI-powered prediction and robotic labs is creating a new golden age for materials science. This new paradigm is expected to dramatically accelerate the discovery of novel materials with revolutionary properties.
AI is now “dreaming up” millions of new materials, with promising early results. A report in Nature highlights the explosion of AI-generated materials, with some initiatives proposing millions of new compounds. While many of these are still theoretical, the sheer scale of this new discovery pipeline is a sign of rapid progress in the field.
A practical example is the AI-driven search for better battery materials. As an example, AI is being used to accelerate the search for new battery materials, a challenging problem with huge potential impact. By sifting through millions of candidates in a matter of hours, AI is drastically reducing the time needed for this critical research.
9 - China is winning the Electric Car race, and it could also win in Electric Trucks
Having already won the electric car race, China’s next target would be freight trucks. Press reports are highlighting the fact that Chinese automakers already account for 80% of global electric truck sales, leveraging their domestic scale to export aggressively. While Western rivals have stalled, Chinese giants like BYD are now shipping electric trucks across the world.
In the West, Tesla posted record sales, but they were a temporary blip, driven by the end of a key subsidy in he US. Tesla’s record-breaking sales quarter, announced on Friday, was actually fueled by a last-minute rush from consumers to take advantage of an expiring $7,500 tax credit. This suggests the sales boom is not really a sign of underlying strength, but a temporary pull-forward of demand.
At the same time, Western giants like GM are slamming the brakes on their EV ambitions. The Wall Street Journal reports that General Motors is scaling back its lofty EV ambitions, as consumer demand in the US has proven to be weaker than expected. The company is now re-evaluating its aggressive push into the electric market.
Europe’s truck industry is now calling for EU action to avoid being crushed by Chinese competition. European truck makers are warning they will lose out to China in the race to electrify heavy vehicles without urgent EU action. The industry is struggling with slow uptake and a lack of charging infrastructure, creating a major opening for Chinese rivals.
10 - The new space race continues to gain momentum
A crewed mission to Mars is the long-term goal, and a new geopolitical race is on. A new space race is heating up, with both the US and China aiming to be the first to land humans on Mars. This has created a new generation of “rocket chasers” who are following the developments with intense interest. The FT is launching a new “Tech Tonic” podcast series to discuss these projects.
NASA is still highly dependent on SpaceX, but is working to diversify its suppliers. While NASA’s ambitious plans still rely heavily on SpaceX, the agency is actively working to cultivate a more competitive landscape. As a key step in this direction, Jeff Bezos’ Blue Origin has been awarded a contract to deliver a NASA rover to the Moon.
Beyond Mars, a new generation of robotic deep-space exploration is being planned. An IEEE Spectrum report this week details an emerging vision for interstellar travel using swarms of tiny, high-speed probes. This new approach could make it possible to explore nearby star systems within a 25-year timeframe.