I. AI's Prove-It Year

We witnessed a mini-bear cycle at the end of 2025. Clearly, investors are nervous about hundred-billion-dollar valuations on companies with opaque revenue, circular deals, and "strategic partnerships" that look more like accounting tricks than real business. These investors have been around since the dot-com boom and housing market, and likely sense cracks easier than most, and can distinguish complex financial engineering. Historically, investors who know it's a shell game and are playing anyways.

My prediction: OpenAI or Anthropic files for IPO by Q4 2026. Not by choice, but by necessity. The private capital patience runs out. When they do, we'll finally see the real numbers. This will be the inflection point.

Who wins: I think Anthropic. Being an early user, I've witnessed their rise from niche to cult following, taking the stance on AI safety and being the premier coding model. Their smaller valuation means less to prove, and their developer loyalty is almost religious; walk around SF shops and you'll see what I mean: everyone uses Claude. They've positioned themselves as the "responsible" lab, which has become synonymous with "pro-human".

II. The Fracturing

We're barely into this presidency and we've witnessed shock after shock. "Independence Day" tariffs, inflammatory tweets, deliberate dollar devaluation being called strategy, the message is clear: America is pulling back.

By December 2026, at least two major economies announce bilateral trade deals that explicitly route around the US. EU-China deepens, and China-India deals will be likely if the Trump administration pushes their hands. Deglobalization means the USD becomes weaker as well.

The truth is, American soft power depends on other nations needing us. Our debt, our currency, our military, it all works because we're the anchor. When we unravel this, we're not "independent" but isolated. And if we can't scale up manufacturing, mining, and energy production before the unravelling completes, we absolutely will not be in a good position.

Who wins: Countries with manufacturing depth and resource access. China, obviously, but also Vietnam, Indonesia, India, which is basically: anyone positioned as the "not China" alternative while China remains the "not America" alternative. The lack of trust between the East and West takes more than an American absence to repair.

III. Hardware Becomes the Main Character

For the last two years, software ate the headlines. Foundation models, agents, chatbots, wrappers. But 2026 is the year everyone admits the real bottleneck: hardware. This is a hill I am willing to die on. Slowdowns in AI software cannot stop the rushing river that is technological revolution, but a diversion to robotic hardware and inference is the logical next step. Robotics is the missing link that connects the latent power of software to the physical world; you are a fool if you don't believe this. If nothing else, AI has only strengthened my belief in how robotics can change physical labor industries for the better.

My prediction: In 2026, we'll see a shift towards the robotic sector investment; it'll be like a new AI or SaaS revolution. We see demand beyond just GPUs, like we did in 2024, but: memory bandwidth and energy constraints become mainstream talking points, not just engineer complaints. We see the first credible "inference chip" company gain real value.

Why this matters: Software progress is hitting a wall. Scaling laws are flattening. The next leap isn't a better architecture (marginal), but it's better chips, better memory, better power efficiency. The AI race becomes a hardware race, and suddenly the US-China decoupling story intersects directly with AI. TSMC becomes the most important company in the world, and everyone knows it.

Who wins is a complicated question. Similar to AI labs, frontier labs are likely not going to be the foremost winners in this competition. When robotics hits a hardware equivalent of 'Turing complete (or human-equivalent dexterity), it's the companies that can use this tech effectively, whether in factories or the military, that will outpace the original genius. On a macro scale, this will have implications for the entire workforce. Who needs factory jobs if we have Turing complete robots? Who needs HVAC? Who needs to drive cars or cook or do chores, or even set foot on a battlefield? Well…. Who needs humans in anything anymore? What a depressing scenario.

Coda: The Open Source Wildcard

One more thing I'm watching: the "DeepSeek shock" proved you don't need $100B to build a frontier model. Whether this model is created through distillation or other 'catch-up' techniques, a $10 million model is still $10 million. As a Formula 1 fan, it's like drawing a comparison to Verstappen catching a slipstream from Hamilton and taking pole. Can we blame either, or is it just part of the game?

My prediction: By mid-2026, we'll see average open-source model performance reach the GPT-5 level, with flagship open models being on par with closed models, if only slightly lagged. Investors need to walk outside more; the hub of technology of the world is San Francisco; yes, developers will use Claude or oAI to help build their software; but when it comes time to serve an AI service in production, 99% of developers will pick a $1 open source model over a $10 closed source model. Deep-thinking and AIMO math capabilities are rarely needed for non-specialized services. This extends to corporate use cases as well. This is the wildcard that could flip everything. If intelligence becomes free, what's left to sell?

And, on a side note, the open source problem becomes a China - US problem. We need open source developed on American soil. National security implications aside, as I said, the real winners are likely not the frontier AI labs developing the technology, but those that can use the technology effectively. And, if those users choose the cheapest version that happens to be Chinese, when end-users pay for said product, we won't like where the money trail ends.