Headlines this week - Oct 19, 2025
A look at how capital is being deployed across future opportunities
This week in the future:
1 - AI is moving the US economy, but still not having a significant impact on productivity
AI investment is “juicing” the economy, but productivity gains remain elusive. While the AI boom is boosting US economic growth through massive capital expenditure and wealth effects from rising stock markets, its actual impact on worker productivity is still unclear. Economists are debating whether the technology is truly making workers more efficient yet.
Instead of boosting productivity, today’s AI is often delivering “workslop”. The current output of AI tools in work environments is frequently subpar, leading to errors and requiring human correction. This phenomenon, termed “workslop,” can actually create more work (I.e. reduce productivity) and undermine trust in the technology.
Blind faith in future AI productivity could be a bad idea. Treating AI investment as a guaranteed path to productivity without understanding the necessary integration steps is dangerous. An analysis by G Tett at the FT warns against this “cargo cult” mentality, emphasizing the complex work required to truly harness AI’s potential benefits within organizations.
Furthermore, productivity gains, when they arrive, could exacerbate inequality. AI tools may disproportionately benefit top performers, amplifying their existing skills and widening the gap between “superstars” and average workers. A WSJ article this week explores how AI could lead to greater income inequality.
2 - No one questions that AI is now an investment bubble. The debate is about if it’s a good or a bad one
The AI boom exhibits traits of a “double bubble”. An FT analysis suggests the current AI investment frenzy is both a potentially beneficial industrial bubble, building essential future computing capacity like past tech booms, and a risky financial bubble characterized by speculative valuations resembling “expensive lottery tickets”.
GPU rental pricing offers clues on how the bubble might pop. Market signals, such as smaller AI cloud providers renting GPUs below estimated cost, point towards unsustainable, venture-capital-fueled pricing. This aggressive discounting could foreshadow a market shakeout as the bubble eventually bursts.
Meanwhile, massive data center investments continue, fueled by private equity. Despite bubble concerns, huge sums are still flowing into AI infrastructure, often driven by private equity and supported by debt, which increases risk. A BlackRock-led consortium, for instance, just agreed to acquire Aligned Data Centers in a deal valued at around $20bn (roughly $40bn including debt).
3 - The economics of AI are changing, making profitability more challenging. But future economies of scale could save the day
AI’s economics remain brutal, with costs rising despite cheaper components. The cost per AI query has increased significantly as models become more complex and require more processing for “reasoning,” even though the cost per individual token has fallen. Future software improvements and massive economies of scale from new capacity could eventually lower costs, but profitability hinges on whether strong, sustained demand materializes to justify the immense investments.
4 - OpenAI’s business plan is looking for revenues at (mostly) very traditional sources
OpenAI is crafting a five-year plan to justify its $1trn spending. The AI lab is developing a five-year business plan outlining new revenue streams, debt strategies, and further fundraising to cover its massive infrastructure spending pledges. Goals include bespoke government/business products, shopping tools, Sora monetization, and AI agents.
E-commerce integration, starting with Walmart, is a key pillar. OpenAI will soon allow users to shop directly from Walmart within ChatGPT, signaling a significant shift in online retail towards conversational commerce. This partnership aims to make purchasing seamless within the AI chat interface.
Early experiences suggest AI could indeed transform online shopping. Using AI for shopping can feel like having a personal shopper who understands nuanced preferences, potentially revolutionizing product discovery and purchasing decisions. One positive user experience highlights AI’s ability to find specific items quickly and effectively.
The company is also reportedly exploring the lucrative adult content market. OpenAI appears to be considering allowing erotic content on ChatGPT for verified adult users, potentially opening up a controversial but highly profitable revenue stream. This move follows the success other AI companies have found in erotic role-play chatbots.
5 - A second aspect of OpenAI’s business model under debate is their aggressive vertical integration. E.g. their deal with Broadcom to build their own chips
OpenAI plans a 10GW custom chip deployment with Broadcom over four years. OpenAI has forged a multibillion-dollar deal with Broadcom to co-develop and deploy custom AI chips, aiming for 10 GW of capacity over the next four years. This represents a massive bet on bespoke hardware to power its future models.
This adds another $350bn-$500bn to OpenAI’s massive spending plans .This custom chip initiative significantly increases OpenAI’s already huge investment plans, potentially adding up to $500bn to its spending spree. This further extends the company’s commitment to building out AI infrastructure.
The goals are hardware savings, energy efficiency, and supply chain control. By designing its own chips, OpenAI aims to achieve significant cost savings (potentially 20-30% vs. Nvidia), improve energy efficiency through custom design, and gain greater control over its critical supply chain.
Custom chips likely target inference, complementing Nvidia for training. Christopher Mims at the WSJ suggests OpenAI’s strategy involves using Nvidia GPUs for the intensive task of training models, while deploying these new, customized Broadcom chips for the higher-volume task of running inference. This “chocolate and peanut butter” approach aims to optimize performance and cost.
The deal sent Broadcom’s shares soaring. Investors reacted positively to the news, sending Broadcom’s shares significantly higher as the market recognized the scale of the potential revenue from the OpenAI partnership. The deal reinforces Broadcom’s position as a key player in the AI hardware ecosystem.
However, some analysts see the deal as a significant risk for Broadcom. Despite the market enthusiasm, some view Broadcom’s heavy reliance on a single, albeit massive, customer like OpenAI as a considerable risk. OpenAI’s own uncertain path to profitability makes this large bet potentially precarious for the chipmaker.
6 - Nuclear energy startups are making progress
The US Army is planning to deploy tiny nuclear reactors to power bases. Small Modular Reactors (SMRs) are being considered to provide resilient power for critical US Army bases, ensuring operational continuity even if main grids fail. The Janus Program aims to have reactors operational by 2028.
Nuclear startup Radiant Industries will build an SMR plant at a historic site. Radiant Industries plans to construct a factory for its small reactors in Oak Ridge, Tennessee, on land once part of the Manhattan Project. Production could begin by 2028, aiming to mass-produce portable nuclear generators.
Securing the fuel supply chain remains crucial for the SMR industry. Oklo is partnering with UK-based Newcleo to invest up to $2bn in US facilities to produce advanced nuclear fuel. This initiative aims to build a domestic supply chain for the higher-enriched fuels required by many next-generation reactor designs.
7 - China has started a war on rare earths, which does not look so good for the West
China’s new export controls on rare earths threaten supply chain chaos. Beijing now requires explicit authorization for exporting products, including magnets, containing Chinese rare-earth materials, even in trace amounts. Western companies, reliant on these materials for strategic goods like fighter jets and cars, warn of production delays and higher prices.
The US faces urgent pressure to achieve rare-earth independence. China’s actions highlight America’s vulnerability after decades of offshoring critical mineral production. Hastening the development of domestic supply chains, from mining to magnet production, is now seen as a critical national security imperative.
High-level talks are underway to de-escalate the dispute. US Treasury Secretary Scott Bessent is meeting with Chinese Vice-Premier He Lifeng in Malaysia in an attempt to resolve the spat over rare earths. The outcome of these talks could determine the trajectory of the broader US-China trade relationship.
Europe may be the biggest casualty in this geopolitical crossfire. Caught between US digital dominance and China’s control over critical minerals, Europe finds itself highly vulnerable. Lacking sufficient investment in its own high-tech industries, the EU risks becoming a permanent supplicant to either Washington or Beijing.
8 - Europe wants to be self sufficient in high tech. But this will require changes in our business culture
Europe is pushing for AI independence to avoid becoming a “tech colony”. Facing accusations of being too slow, European nations are fighting for AI sovereignty. They aim to build homegrown capabilities and avoid complete reliance on US and Chinese technology giants for this critical infrastructure.
The bigger obstacle may be Europe’s risk-averse investment culture. While regulation is often blamed, a lack of ambitious venture capital funding, especially from large institutions like pension funds, is a major hurdle. A Bloomberg article this week argues that unlocking more risk capital is key to nurturing world-class AI companies in Europe.
9 - New AI models have just been released that specifically aim to accelerate drug discovery
AI can now predict cellular changes before lab experiments are run. A new AI model predicts the visual changes in cells resulting from gene edits or drug treatments, based on gene expression data. This generative tool allows researchers to simulate potential outcomes before conducting physical experiments, potentially accelerating drug discovery.
10 - One more lab (xAI) confirms it’s working on “world models”, which are generally perceived as a key step for further progress in AI
Elon Musk’s xAI joins the race to build “world models” for physical spaces. xAI is developing “world models,” training them on videos and robot data to understand physical environments. An initial application will be in video games, generating interactive 3D environments, potentially expanding to robotics later.