Headlines this week - Dec 28, 2025
A look at how capital is being deployed across future opportunities
This week in the future:
1. Amid bubble concerns and growing risks, Vertical Integration emerges as a key advantage for AI labs
“AI Bubble” talk continues, in the context of 2026 predictions, but the tech elite remains not worried. While skeptics fear a crash, tech leaders have not altered their massive investment plans. However, as we’ve already discussed here, volatility in Meta, Nvidia and other AI-centric stocks suggests investors are increasingly wary of excesses.
With massive capacity reducing token prices, Vertical Integration is becoming a shield for labs against commoditization. With token prices crashing and spending skyrocketing, advantage (and the ability to get returns on the massive investments) may be shifting to “full stack” giants like Alphabet that control chips, data centers, and applications, as this protects them from severe price deflation.
The massive build-out is stressing Big Tech’s financials. Corporate bond sales are near historic highs. AI-related borrowing now drives 30% of investment-grade issuance, with companies issuing $1.7tn in 2025 to fund infrastructure. This is sparking fears of a “debt glut” if returns stall.
Also, more people are commenting about how visible debt is only half the story, as billions move into opaque shadow financing Tech giants like Oracle, xAI, CoreWeave or Meta have moved approx. $120bn of data center spending into off-balance sheet Special Purpose Vehicles, shielding credit ratings but obscuring systemic risks if demand wavers.
2. Energy is being confirmed as a key resource, and Alphabet is doing Vertical Integration into this, too
To solve the energy bottleneck (and control the costs), Alphabet is moving to own the electricity supply directly. Breaking with the industry standard of mere purchase agreements, Google is paying $4.75bn to acquire Intersect Power, a clean energy developer with a 15GW solar and storage pipeline. This “self-powering” strategy bypasses grid constraints to fuel data centers directly, and reinforces Alphabet’s position as a fully integrated AI lab (an advantage in a deflationary AI world -see above)
3. Volatility is increasing in the energy industry, driven by uncertainty on who will be able to deliver on AI needs, and by geopolitical factors
All / most energy stocks are soaring on AI hype, but eventually there will be a selection. Investors indiscriminately bid up energy stocks, but reality is starting to set in. The market is realizing that not every power source, from solar to nuclear, can actually scale and deliver the gigawatts AI demands.
Desperate for power, US data centers are bypassing the grid with jet engines fed by fossil fuels. Facing seven-year grid delays, developers are installing jet engines and diesel generators on-site. This “dirty fix“ prioritizes immediate compute power over emissions goals, locking in fossil fuel reliance.
This bet on hydrocarbons offers speed today but risks strategic failure tomorrow. Relying on gas provides immediate speed but risks long-term competitiveness. Analysts warn that higher electricity prices and water stress from fossil-fuel-powered AI could eventually throttle the US position in the AI race against China.
Meanwhile, China seems to be playing the “long game”, betting its AI future on green tech dominance. Conversely, China would be accepting higher initial costs to scale green fuels like ammonia. This strategy would be looking to secure supply chain dominance, promising stable, low-cost power for its future AI infrastructure.
4. Nvidia confirms that the chip game is shifting to inference, where they look weaker, and absorbs a key challenger’s talent and tech
Nvidia is licensing Groq’s IP to cement its lead as the market shifts toward inference. In a strategic pivot, Nvidia has licensed technology from Groq (a key emerging competitor) to integrate specialized architecture for “inference”. For this purpose, Groq claims that its chips can be produced and deployed faster and use less power than Nvidia’s GPUs. The move signals Nvidia’s urgency to dominate not just AI training, but the rapidly growing market for running models efficiently.
The deal is an effective “acquihire,” through which Nvidia “poaches” a key executive to neutralize Groq’s threat. By hiring Groq founder Jonathan Ross and his top engineers, Nvidia removes a competitor and gains key talent, including the guy who designed Google’s TPU in the past. Analysts suggest the “licensing” structure helps bypass antitrust hurdles while effectively absorbing Groq’s core capabilities.
5. Meanwhile, in China four startups are working to turn the country’s ambition of chips self-sufficiency into a reality
The “Four Little Dragons” emerge as China’s answer to Nvidia’s dominance. Four startups, Biren, Moore Threads, MetaX, and Enflame, are spearheading China’s push for GPU self-sufficiency. Founded by veterans from Nvidia and AMD, they are racing to build domestic alternatives for AI training and inference, aiming to break the US stranglehold on advanced chips.
To fuel this capital-intensive battle, Biren and others turn to public markets. Facing high R&D costs and US sanctions, these “dragons” are tapping public markets. Biren is now planning a $623m Hong Kong IPO to fund its next-gen chips, following explosive market debuts by Moore Threads and MetaX that saw shares surge 400-700%.
This rise of domestic champions triggers new US trade threats, though action is delayed. Washington is accusing China of using unfair practices to dominate the global chip sector but has paused new tariffs until 2027. This delay offers a critical window for these startups to scale before facing the next wave of trade restrictions.
6. After major progress in 2025, expectations are growing for quantum computing. But not all players are equally well positioned
After a breakthrough year, 2026 promises the first leap into fault-tolerant, error-corrected computing. With hardware finally meeting 30-year-old theoretical thresholds, companies like Microsoft or QuEra are deploying machines with “logical qubits,” moving the industry from noisy experiments to reliable, error-corrected utility.
Yet this maturity threatens to expose the “snake oil” salesmen relying on hype over hardware. In a podcast interview this week, top Quantum Computing expert Scott Aaronson warned of a sharp divergence: while engineering-first firms like Alphabet, IBM, QuEra or Quantinuum are hitting milestones, others prioritize marketing narratives and IPOs, risking obsolescence as the market demands proof of scientific validity.
As we enter the “second quantum century,” scientific utility, not just hype, will define the winners Caltech’s John Preskill (also a top expert in the field) agrees with Aaronson’s views and argues that true value lies beyond today’s noisy systems. For Preskill (currently advising Amazon) error correction and neutral atom platforms will be the essential drivers for the “second century” of quantum utility.
7. People keep discussing the risks of AI: (a) Cyberattacks & Fraud
There is a consensus that AI lowers the barrier for cybercrime, allowing novices to launch sophisticated attacks at scale. Generative AI is democratizing cybercrime, enabling non-technical actors to automate attacks (remember the use of Anthropic agents by Chinese hackers some weeks ago) and craft highly convincing social engineering campaigns. This shift will force defenders to radically rethink their security strategies.
There is also a threat to the art world, where chatbots now forge convincing provenance documents to authenticate fakes. Fraudsters are using AI models to generate fake invoices and certificates, creating a “new dimension” of fraud. This automated documentation makes verifying ownership and authenticity increasingly difficult, even for experts.
8. People keep discussing the risks of AI: (b) Job substitution
AI is expected to automate the “grunt work,” breaking the traditional corporate ladder. By eliminating the routine tasks that once served as training grounds, companies risk severing the pipeline of future leaders, forcing a complete reinvention of how junior talent is developed.
This forces a radical shift (and an opportunity): juniors must now deliver senior-level value on day one. E.g. in software development roles, with basic coding commoditized, the bar for entry-level engineers has skyrocketed. However, according to an article at IEEE Spectrum this week, the shift also offers a unique opportunity: by using AI to bypass “grunt work,” young professionals can build high-quality software faster and accelerate their growth into strategic, higher-order roles
9. A short-term application of (nascent) AI “World Models” may be gaming landscapes
Video-games emerge as a critical application for AI ”World models”. As we discussed here last week, “World models” are moving AI beyond text to generate physics-compliant 3D environments. As an initial application they are targeting the gaming sector. Both Google DeepMind and Fei-Fei Li’s World Labs are developing systems to instantly create interactive worlds. This promises to revolutionize game production, slashing the years and billions currently required to build “triple-A” titles.
10. Higher uplink capacity and networks becoming sensors emerge as big innovation themes in telecoms, according to IEEE Spectrum’s views on 2026 perspectives
6G will shift the focus to the massive uplink capacity needed for XR. While 5G prioritized download speeds, the next generation of infrastructure is being designed to handle the massive “uplink” data demands of XR glasses and immersive tech, solving the bottleneck that currently limits these devices.
Simultaneously, networks might evolve from “dumb pipes” into vast, distributed sensing fabrics. Innovations are transforming telecom infrastructure into active sensors, enabling optical fiber lines to detect environmental changes. This turns the network itself into a global monitoring tool, offering applications far beyond simple connectivity.
Beyond the grid, innovation expands to “terahertz” chips, space lasers, and quantum security The sector is also advancing via terahertz chips for massive bandwidth, hollow-core fibers for speed, and free-space optical lasers, alongside new protocols for deep-space and quantum encrypted messaging.