Headlines this week - Dec 7, 2025
A look at how capital is being deployed across future opportunities
This week in the future:
1 - OpenAI is now officially under pressure (mostly from Google…) And this could hurt their innovation capabilities
Under pressure from Google, OpenAI has declared a “Code Red” to defend its lead. CEO Sam Altman has ordered staff to pause peripheral experiments like the “Pulse” personal assistant to focus exclusively on improving ChatGPT’s quality, reacting to the intense pressure from a resurgent Google.
Most people believe that Google’s Gemini 3.0 has erased the technical gap. Google’s latest model now surpasses GPT-5 on key industry benchmarks, leveraging a “full stack” advantage—owning everything from custom chips to data centers—to train models faster and cheaper than its rival.
In fact, Google’s infrastructure “moat” has the potential to create a long-term edge. Analyst Ben Thompson highlights Google’s unique advantage in possessing its own TPU infrastructure; unlike OpenAI, which is dependent on Nvidia, Google controls its supply chain, allowing for superior long-term economics.
Even Geoffrey Hinton (the “Godfather of AI”, and also an ex-Google employee…) bets on Google to win. Hinton now predicts that Google will ultimately win the AI race, arguing that despite a cautious start, its massive advantage in computing power is allowing it to rapidly overtake OpenAI.
OpenAI also has strengths: Their market share / user base could help them maintain traffic dominance. OpenAI’s chatbot retains billions of monthly visits compared to Gemini’s millions, relying on “sticky” features like persistent memory and personality. These could be a powerful tool to keep users loyal even in a scenario where Google leads in performance
However, the “Code Red” pivot risks stifling future “accidental hits.” By shelving promising “moonshots”—like the “Pulse” personal assistant and a new shopping agent—to defend ChatGPT, Altman prioritizes immediate survival over the open-ended experimentation that created his core product. This defensive retreat might solve today’s focus crisis but risks choking off the serendipitous discovery of the next breakthrough consumer application, effectively placing product innovation in the back seat
2 - The whole chip supply chain is getting hot. And Nvidia (like OpenAI) is under pressure
Nvidia’s high margins invite fierce competition. The company’s massive profits have created a “margin umbrella,” incentivizing rivals like Google and AMD to undercut pricing. Customers are increasingly seeking cheaper alternatives to Nvidia’s expensive hardware, creating an opening for competitors to erode its dominance.
Rivals seem to be rapidly catching up, and challenging Nvidia’s performance. Major tech players are no longer just customers but active competitors, deploying proprietary silicon that challenges Nvidia’s performance. This shift threatens to loosen Nvidia’s stranglehold on the AI hardware market as custom alternatives become viable for high-performance workloads.
Alphabet is the most serious threat. Their custom silicon could be a massive business. Analysts predict that if Alphabet sells its “Tensor Processing Units” to third parties, this could make it possible for the company to capture up to 20% of the AI market. These efficient chips are viewed as a “secret sauce” capable of rivaling Google Cloud’s own value.
Amazon is next, and has aggressively started selling its own AI chips. AWS has started selling access to its Trainium processors to startups like Decart, positioning them as a lower-cost alternative to Nvidia. This move allows Amazon to diversify its data center infrastructure while directly challenging Nvidia’s market supremacy.
Meanwhile, in China, investors are excited with local Nvidia alternatives. Chinese Nvidia rival Moore Threads’ stock rose +425% on its first day, fueled by Beijing’s push for semiconductor self-reliance. Chinese investors are betting the firm can be one of the local companies that replaces Nvidia in China, even though they currently lag domestic competitors like Huawei. At least the CEO is fully patriotic in his public comments (“computing power is state power”)
Other parts of the supply chain are also getting hot:
Trump administration takes equity in chip startup XLight. The US government will invest up to $150m in the photonics startup led by former Intel CEO Pat Gelsinger, which aims to increase the density of compute by improving the process of lithography, currently dominated by ASML.
AI startup Ricursive aims to automate chip design. Founded by ex-Google researchers, the company has raised $35m to build software that designs chips from scratch. By automating complex architectures, they aim to democratize custom silicon creation, potentially disrupting the $800 billion semiconductor industry
3 - Zuckerberg is re-focusing Meta on AI, at the expense of the Metaverse
Meta slashes the Metaverse budget to fund the AI race. The company is planning to cut its metaverse division’s budget by up to 30%, signaling a major retreat from the immersive worlds that have lost over $70bn since 2021. Resources are being aggressively redirected to win the battle for artificial general intelligence.
Smart glasses are now the vehicle for “superintelligence” (and not an Augmented Reality device anymore). Meta poached top Apple designer Alan Dye to lead a new studio focused on AI-powered eyewear, pivoting the product away from augmented reality. Zuckerberg believes these devices will eventually replace smartphones as the primary computing platform for AI.
Meta acquires Limitless to dominate AI wearables. Consistently with the new priorities, Meta has just bought Limitless, a startup building an “AI pendant” device, to accelerate its push into hardware designed specifically for AI assistants. This acquisition reinforces the company’s strategy to own the physical interface for consumer AI
4 - Anthropic might be preparing its IPO. It would test the market’s beliefs on its “safe AI” equity story
Anthropic is preparing for its IPO. The company has hired a law to prepare for a potential 2026 IPO, racing rival OpenAI to public markets. The listing could value the startup at over $300bn, testing investor appetite for massive, loss-making AI research labs.
This would be a test for the attractiveness of the “Safe AI” pitch. Investors must decide if Anthropic’s narrower, “safe” AI focus justifies a $350 billion valuation, roughly five times its projected 2028 revenue. Unlike OpenAI’s product sprawl, Anthropic bets its “helpful, honest, and harmless“ principles will command a premium from enterprise clients.
CEO Dario Amodei is at the center of the equity story. Amodei, who left OpenAI over safety concerns, positions himself as the only leader capable of responsibly stewarding all-powerful AI. While critics call his safety focus “fear-mongering,” his “cult” status helps retain top talent in the fierce race for AGI.
5 - Meanwhile, jobs emerge as the #1 concern about AI deployment. The substitute vs. complement debate is still open
On the “complement” side:
Radiologists defy predictions of extinction as AI fuels demand. Contrary to Geoffrey Hinton’s 2016 forecast that deep learning would make them obsolete, the number of radiologists has surged—up 40% in the UK since 2016. AI has not replaced them but acts as a “co-pilot,” handling pattern detection while radiologists focus on complex judgments and auditing AI outputs.
The human touch remains essential in customer support strategies. Major firms like Allianz and easyJet are finding that fully automated service is “unlikely and undesirable,” preferring hybrid models where humans handle sensitive issues. Gartner predicts that by 2027, half of the companies planning to replace staff with AI will abandon those plans due to the technology’s struggle with nuance.
On the “substitute” side:
Tech executives warn that AI job displacement is inevitable. At a recent WSJ event, industry leaders argued that AI will “flatten” organizations, specifically threatening middle management roles. The debate among these executives has shifted from if jobs will be lost to who is most at risk and how quickly the transition will occur.
6 - A significant impact of AI on productivity (and maybe jobs) might only come after deep transformations in the way companies work
Businesses must adapt to reach AI’s promised Plateau of Productivity. According to Gartner’s 2025 “Hype Cycle”, Generative AI could be sliding into the “Trough of Disillusionment” as hype has started to fade. However, firms like Mimecast, a global cyber security company that has successfully used AI to increase productivity, show that the actual reason why many AI projects don’t deliver is not so much linked to intrinsic technology weaknesses, but to the fact that true productivity gains are only possible after deep cultural shifts and workforce training
Smaller, cheaper models could also help accelerate adoption (this is China’s approach). While the US bets on massive, closed AI models, China is prioritizing efficient, open-source alternatives like DeepSeek. Driven by chip restrictions and a focus on rapid diffusion, this “fast and cheap” strategy is already overtaking US models in global adoption
7 - What’s next for AI: Interaction with “the real world” and new architectures (beyond “Transformers”)
Scale alone is hitting a wall. Future progress won’t come just from bigger datasets, but from “world models” and reinforcement learning environments. These approaches require digital “classrooms” for AI models / agents to interact, fail, and learn from experience rather than just mimicking static text.
Indeed, startups are building fake internet sites to train agents. Companies are creating perfect replicas of platforms like Amazon and Uber to serve as safe “sandboxes”. These mock-ups allow AI agents to practice complex tasks—like booking rides or shopping—without spending real money or triggering security bans.
New architectures aim to supersede Transformers. Startups like Pathway are betting on “post-Transformer” designs that process data streams continuously rather than in batches. This shift aims to solve the inefficiency of current models, enabling AI to learn in real-time and handle vast, changing contexts.
8 - The race is open to catch up with SpaceX in rockets. But investors don’t look too concerned
Blue Origin moves ahead in the lunar race with New Glenn, in direct rivalry with SpaceX. Jeff Bezos’ space company is successfully testing its heavy-lift rockets, aggressively positioning itself as the primary rival to SpaceX for NASA’s Artemis moon missions. This progress signals a serious challenge to Musk’s long-standing dominance in orbital logistics.
China is also testing reusable rockets (with some setbacks). Startup LandSpace’s Zhuque-3 rocket crash-landed during a vertical recovery test, highlighting the immense technical gap with SpaceX. Despite the failure, the mission provided critical data as Beijing aggressively pushes to develop its own reusable launchers to break the US monopoly.
Meanwhile, SpaceX investors don’t look too concerned. The company is targeting a record $800bn valuation. Elon Musk’s rocket giant is discussing a share sale that would double its value and surpass OpenAI as the world’s most valuable startup. Investors remain bullish on its launch monopoly and Starlink’s dominance, seemingly unfazed by the emerging competition from the US and China
9 - Some industry news this week support the idea that small reactors might be the future of nuclear energy
US invests $800m to boost small reactor deployment. The Department of Energy awarded funding to Holtec and the Tennessee Valley Authority to build develop nuclear projects with so called Generation III+ reactors, smaller versions of the technology that’s widely used now in conventional fission plants. The investment is driven by the need to supply power for AI data centers.
Bill Gates-backed nuclear project nears construction milestone. TerraPower (a Small Modular Reactor startup) expects to break ground on its next-generation Natrium reactor in Wyoming by mid-2026. The company finalized key design contracts, aiming to prove its molten salt technology can deliver cheaper, safer nuclear energy faster than traditional plants.
10 - Does Europe still have an opportunity in AI? News from a couple of startups are a source of optimism
German image-gen startup is competing with US and Chinese giants. Black Forest Labs raised over $450m to compete with Google and ByteDance, tripling its valuation to $3.25 bn. The Freiburg-based creator of the popular “Flux” text-to-image models aims to upend industries from advertising to Hollywood with its open-source technology.
HSBC taps Mistral for major AI rollout. The banking giant signed a deal with the French AI champion to deploy generative models for financial analysis and translation across its workforce. This partnership highlights Mistral’s growing role as Europe’s primary alternative to OpenAI for enterprise clients.