The energy constraint has become the defining bottleneck in AI development. As models grow smarter—not through more data, but through more compute time—the race has shifted to securing reliable, cheap power. Microsoft, Amazon, and Google have already begun investing in nuclear plants to feed their training infrastructure. Álvaro's prediction isn't that nuclear becomes trendy again; it's that it becomes infrastructure, as boring and essential as the power grid itself. This shift has real consequences: the companies that secure nuclear capacity first will outpace those relying on conventional energy markets.
On artificial intelligence proper, the consensus view breaks down. While reasoning models like OpenAI's O3 represent genuine capability jumps, Álvaro expects the next frontier to be multimodal: better audio interfaces, video generation that evolves faster than text models, and Large Behavioral Models—the systems training robots in kitchens and teaching agents to navigate computer interfaces. The boring but consequential bet is that AI becomes a commodity. Google and Anthropic will match OpenAI's reasoning capabilities. The moat disappears. What remains is distribution and integration.
But 2025 may also be the year an AI incident becomes headline news—not because the technology is new, but because guardrails fail at scale. A leaked deepfake affecting an election, a chatbot spreading disinformation that gains traction, a harmful interaction that OpenAI's nine-month safety sprint couldn't prevent. The reputational damage could be significant enough to reshape how these companies operate, though Álvaro hedges: this is less prediction than inevitability dressed up as foresight.
Consumer tech tells a different story. The Vision Pro's failure to gain traction forces Apple's hand: a cheaper "Vision Air" will arrive, priced below $3,000, with an app ecosystem finally mature enough to justify the hardware. Smart home adoption accelerates—not because the technology changed, but because it matured. TikTok Shop and social commerce will drive meaningful e-commerce volumes in the West for the first time. And one of the struggling social platforms—Blue Sky, Threads, or Mastodon—will face forced change: acquisition, shutdown, or pivot. Twitter's dominance in discourse, Twitter's sheer network effects, may finally prove impossible to dislodge, even as the open social media narrative sounds noble.
The macro trends are harder to miss. Bitcoin hits new highs, not through cult dynamics but through institutional adoption and sheer weight of capital. Crypto undergoes quiet maturation—fewer pump-and-dump schemes, more actual use cases hiding behind fintech interfaces. VR hardware becomes cheaper and better. And quantum computing, after years of theoretical breakthroughs, finally solves a problem with economic value. Not a math puzzle. Not a milestone. A real problem: drug discovery, cryptography, optimization. Something that moves from the lab to the business page.
The workplace reorganizes. More companies adopt four-day weeks, paired with efficiency gains from AI agents. More mandate office returns—not fully, but enough to pressure remote workers. And at least one company, somewhere, will run primarily on AI agents, doing it not because it works but because the PR stunt works. The real shift happens quietly: companies stop hiring for specific, narrow tasks. Instead, they hire for judgment and the ability to work alongside automation.
Mobility contracts. One traditional automaker faces serious trouble—probably European, almost certainly because Chinese EV makers like BYD have underpriced and over-delivered. Scooters get banned in more countries. The industry consolidates further.
In the background: TikTok gets banned in more places, including the U.S., eventually. Nvidia's reign as most valuable company ends. And Shutdown, the podcast, reaches 1,500 subscribers by year-end. Some predictions are bold. Most are just the obvious next move, waiting for someone to call it.