Headlines this week - Dec 21, 2025
A look at how capital is being deployed across future opportunities
This week in the future:
1 - Fusion energy consolidates as an investment trend, driven by AI’s “energy thirst” (and now also by Donald Trump)
Trump Media bets big on nuclear fusion. This week we learned that Donald Trump’s media company had agreed to a $6bn merger with Google-backed TAE Technologies (a nuclear fusion energy startup). The all-stock deal aims to leverage Trump’s access to capital to accelerate fusion energy development, positioning it to power the energy-hungry AI boom.
This deal could help supercharge fusion funding. The merger highlights the synergy between speculative finance and high-stakes science. By bringing fusion into the “meme-stock” orbit, Trump could unlock vast retail capital for an industry that desperately needs billions to bridge the gap between experimental breakthroughs and commercial reality.
AI’s energy thirst is a key driver of fusion investment. Despite nuclear fusion energy remaining years away from commercial viability, tech giants and billionaires are pouring money into startups in the field. The driving force is the desperate need for clean, 24/7 power for AI data centers, with investors betting that the massive potential payoff outweighs the significant technical risks.
But not everyone agrees: Elon Musk has publicly dismissed “terrestrial” fusion. The Tesla CEO mocked the rush for fusion reactors, calling it “super dumb” to build “tiny” versions on Earth when the Sun already functions as a massive, free fusion reactor. Musk argues resources are better spent harvesting solar energy than chasing complex nuclear physics.
2 - “Traditional” nuclear energy (aka fission energy) is also gaining momentum
In the US, the nuclear industry is set for a rapid expansion after 2025 slump. After a year where global capacity actually shrank, the sector is poised to add 15 new reactors in 2026 and over 50 by 2030. This major resurgence is driven by climate goals and the desperate need to power AI data centers.
The US government has pledged $80bn to revitalize the nuclear landscape. The massive federal plan aims to drastically expand capacity, but critics warn execution remains a hurdle. Given the industry’s history of cost overruns and delays, delivering on this ambitious roadmap will require overcoming significant logistical and regulatory challenges.
A potential solution might be Small Nuclear Rectors, which would radically simplify building capacity, and which are attracting investors attention, as seen with two startups this week:
Last Energy raises $100 million for off-the-shelf reactors. The startup has secured Series C funding to commercialize its 20-megawatt micro-reactors. By using standard components to reduce costs, Last Energy aims to swiftly deploy power plants for industrial customers, bypassing the delays typical of traditional nuclear builds.
Radiant hits near-$2 billion valuation for portable nuclear. This second startup has raised $300m to build 1-megawatt micro-reactors small enough to fit in a shipping container. With plans to power military bases and data centers, Radiant’s “nuclear-in-a-box” approach has attracted major venture capital to mass-produce the units.
3 - Data center energy needs are also triggering a political backlash against AI infrastructure deployment. The noise is growing
US Senate investigation targets AI’s impact on utility bills. Three Democratic senators have launched an inquiry into whether tech giants like Amazon and Google are driving up residential electricity costs. They fear that billions spent upgrading grids for power-hungry AI data centers are being unfairly passed on to ordinary consumers.
Bernie Sanders is demanding a moratorium on AI infrastructure. The senator is pushing to halt data center construction, arguing the “unregulated sprint” to deploy AI prioritizes the 1% over the public. Sanders insists on pausing development until democratic oversight can ensure the technology’s benefits are shared equitably.
But other AI skeptics reject the construction ban. Daniel Kokotajlo (a co-author of the AI2027 “doomer” report) dismissed the moratorium idea as “NIMBYism” that would fail to regulate AI behavior. He argues that blocking domestic infrastructure will simply force tech companies to build elsewhere, avoiding US oversight without solving the core safety problems.
Data centers in hot climates might exacerbate energy costs. A new analysis reveals that AI infrastructure is increasingly being built in tropical locations like Singapore. These hot environments require massive amounts of additional energy for cooling, deepening the trade-offs between digital progress and environmental sustainability.
Moving data centers to space could be a solution, but it faces different challenges, e.g. high costs and debris risks. Google’s proposal to launch solar-powered server farms into orbit aims to bypass earthly energy limits but faces steep challenges. Experts warn that the astronomical costs of launch and maintenance, coupled with worsening space debris, make this sci-fi solution fraught with practical peril.
4 - A second source of political backlash against AI is the threat on jobs. White-collar workers are supposed to be at risk, particularly in some creative functions (previously safe from automation)
White-collar workers face 2026 with deepening job insecurity. As AI tools proliferate and corporate layoffs mount, anxiety is spreading across the economy. Even where adoption is still nascent, the expectation of automation is driving workforce reductions, leaving employees wondering how long their roles will last.
A high-profile example: Hollywood struggles to balance AI productivity with survival. Studios face a critical dilemma: using AI to slash costs and boost output risks decimating the workforce. This shift threatens to fundamentally alter the industry’s economics, pitting efficiency against the livelihoods of the human talent that built it, and potentially creating opportunities for new entrants to disrupt the traditionally dominant companies.
A signal of what is coming for the content industry: YouTube bans channels showing AI-generated fake trailers. The platform has terminated prominent accounts like Screen Culture for misleading millions with AI-made movie teasers. This enforcement marks a significant escalation in the battle against deceptive synthetic content, signaling tighter policing of AI in entertainment.
In the UK, actors are refusing digital scanning to avoid being used to train AI models. In a decisive pushback, 99% of the Equity performing arts union members voted to reject on-set scanning. They fear these digital replicas will be used to train video-generating models, effectively allowing studios to replace them with AI clones without consent.
Meanwhile, the debate over “Made by AI” labels intensifies. While the EU’s AI Act mandates labeling, the gaming industry’s experience suggests transparency is complex. Validating human authorship is becoming a key battleground to preserve a market for real creators amidst a flood of indistinguishable synthetic content.
A second example: AI threatens to decimate advertising agencies. Brands are rapidly adopting AI tools to generate ads faster and cheaper, bypassing traditional agencies. This shift targets the “low-hanging fruit” in creative production, threatening to empty Madison Avenue and displace thousands of workers in a once-safe industry.
A more optimistic view: Junior lawyer pay is rising despite AI displacement fears. Contrary to predictions that AI would automate legal grunt work and kill jobs, salaries for junior associates are increasing. The FT’s “AI Shift” section, continuing with its optimistic narrative about AI and jobs, suggests AI may be acting as a complement rather than a substitute, shifting duties rather than eliminating roles.
5 - Meanwhile, the “brute force” approach to AI progress, a key driver of infrastructure deployments, is increasingly under question (by top researchers in the field…)
Fei-Fei Li bets on “spatial intelligence” to advance AI. The “godmother of AI” (ex Stanford AI labs, ex Google) claims language models are insufficient and has founded a new company, World Labs, to create AI that understands the physical world. Her “world models” aim to generate 3D environments, moving beyond simple text processing to true spatial reasoning.
Yann LeCun leaves Meta to build “World Models.” The AI pioneer (ex Meta) is raising €500m for his new startup, Advanced Machine Intelligence, targeting a €3bn valuation. Like Li, LeCun aims to move beyond LLMs by creating systems that understand the physical world, unlocking applications in robotics and autonomous transport.
Meta’s new models will test how much to expect from the traditional approach. The tech giant is developing a new image and video model code-named “Mango” and a next-gen LLM dubbed “Avocado.” These “brute force” models, championed by new AI chief Alex Wang, are expected to launch in early 2026 to regain dominance.
Meta is doing this in a context of turbulent internal culture wars. Meta’s aggressive hiring of “super-researchers” with massive pay packages has alienated existing staff. This “turbulent” bet has created a two-tier workforce, causing friction as the company scrambles to integrate expensive new talent with its established engineering teams.
6 - So more uncertainties are being added to the already risky “AI Infrastructure” investment theme. Investors are increasingly concerned
AI does exhibit classic signs of a financial bubble. Analyst Ruchir Sharma at the FT identifies four key warning signs (over-valuation, over-ownership, over-investment, and over-leverage) that currently characterize the AI market. With tech stocks meeting historical bubble thresholds, the risk of a crash rises significantly if the underlying industry falters (and we’ve already discussed some potential reasons for that, in the previous points…)
At the very least, a market correction looms as AI returns lag expectations. WSJ columnist Andy Kessler warns that while AI capabilities are improving, stock prices are dangerously detached from reality, pricing in years of future growth. With massive expectations but relatively poor immediate returns, the sector is “due for a dip” as reality sets in.
Apparently conscious of all this, Tech giants are engineering financial schemes to hide AI risks. Microsoft, Amazon, and Google are increasingly using complex financial leases and separate corporate entities to keep billions in AI infrastructure spending off their balance sheets. These maneuvers shield their profits from depreciation costs while offloading risk to lenders and partners.
At the same time, “circular deals” that augment the risks are still happening: E.g. Amazon and OpenAI this week. OpenAI is in talks to raise $10bn funding from Amazon in exchange for adopting its proprietary chips and cloud services. This potential partnership mirrors criticized “circular” investment structures, boosting Amazon’s hardware ambitions while inflating the startup’s valuation to over $500bn.
Scenarios where the bubble bursts are being openly discussed: OpenAI is not “too big to fail,” says top economist. Jason Furman from Harvard argues that an OpenAI collapse would be painful for investors but wouldn’t trigger a 2008-style systemic crisis. Unlike banks, AI firms don’t hold federally insured deposits, meaning a failure would destroy wealth without necessitating a government rescue.
7 - An extra, powerful incentive for the massive investments is the “AI war” against China, that is not showing any signs of slowing down
China is building a secret “Manhattan Project” for AI chips. Reuters reveals a clandestine effort to develop local Extreme Ultraviolet (EUV) lithography machines. By recruiting former ASML engineers and reverse-engineering tech, Beijing aims to break the Western monopoly and achieve semiconductor independence by 2030. This is being compared to the Manhattan Project during the Second World War.
Old ASML machines are being upgraded in China, to bypass sanctions. Chinese manufacturers are retrofitting older “deep ultraviolet” (DUV) lithography tools with new components to produce advanced AI chips. This strategy exploits gaps in export controls, allowing firms like Huawei to boost output despite Western restrictions.
8 - On the application side, the shift from traditional taxis to robotaxis is accelerating, with valuations of Waymo and Tesla soaring
Waymo targets a $100bn valuation to fuel global expansion. Alphabet’s self-driving unit is discussing a massive funding round led by its parent company. The fresh capital would support deploying the company’s robotaxi service in new markets like London and New York, solidifying its lead over rivals.
Also, Tesla stock hits record high on robotaxi optimism. Shares rallied to an all-time peak as investors bet that converting existing EVs into autonomous taxis will drive future growth. This hype has overshadowed a decline in vehicle sales, fueling a massive valuation surge.
Tesla itself keeps fueling the narrative: removing safety drivers for robotaxi tests in Austin. In a critical step toward a commercial service, the company has begun testing autonomous vehicles without human monitors. This move signals confidence in its software, aiming to compete directly with Waymo’s established driverless fleet.
A dominant design for autonomous vehicles (that could catalyze growth) may be emerging. Despite perceived differences, Waymo and Tesla seem to be converging on similar “end-to-end” foundation models for driving. This architectural alignment suggests the industry is moving toward a standardized, data-driven approach that could accelerate overall progress.
But the news from China are not so good, with local robotaxi stocks slumping while US rivals soar (a warning sign?). Shares in newly listed firms like Pony.ai have fallen since their debuts, highlighting a sharp divergence. While US investors view autonomy as a high-margin software platform, markets treat Chinese players as capital-intensive hardware services.
9 - A startup wants to build “digital brains”, as a new approach to AI
Princeton professor Sebastian Seung has launched Memazing (a startup) to create “digital minds.” The company aims to reverse engineer the fruit fly’s brain using its (already produced) complete wiring diagram / “connectome”. By simulating biological intelligence in software, Memazing seeks to build efficient AI models and ultimately pave the way for human brain emulation
10 - As AI adoption increases cybersecurity threats, valuations soar for startups addressing these threats
Blackstone leads $400 million bet on AI-native security. The investment giant’s round values Cyera at $3bn, signaling strong confidence in startups addressing AI-driven vulnerabilities. Founded by former Israeli military officers, Cyera uses AI to discover and protect cloud data, capitalizing on the urgent demand for security in the generative AI era.