Radical Blog

In an annual end-of-year tradition, Radical Ventures Partner Rob Toews published 10 predictions for the world of AI in 2026 in his Forbes column. Five out of Rob’s 10 AI predictions for 2025 ended up coming true, including scaling laws being discovered beyond text in biology and robotics, the Trump-Musk falling out, and AI data centers in space project announcements. 

This week, we share excerpts from Rob’s 10 AI predictions for 2026:

1. Anthropic will go public. OpenAI will not.

AI research labs OpenAI and Anthropic are such unique organizations that it can be easy to forget that, ultimately, they are venture-backed businesses. And not just any venture-backed businesses — they are the fastest-growing and most capital-hungry venture-backed businesses in history.

According to its own projections, Anthropic will burn through close to $20 billion before it becomes profitable, making it one of the most capital-intensive businesses in history. But that figure is dwarfed by OpenAI, which according to its own projections will need roughly $150 billion before it starts generating cash.

Companies with such staggering capital needs have no choice but to tap public capital markets in order to raise larger sums of both equity and debt financing.

Both companies have raised a lot of money from a lot of institutional investors over the past few years, many of them crossover investors whose primary focus is public markets. Pressure will begin to mount from this investor base to go public sooner rather than later in order to provide liquidity.

We predict that, in 2026, Anthropic will debut on public markets. It will be one of the biggest and most highly anticipated IPOs of all time.

2. Details of SSI’s research and technology will leak to the public. The big labs will make meaningful adjustments to their research roadmaps as a result.

No technology company in the world is more shrouded in mystery than Ilya Sutskever’s Safe Superintelligence (SSI).

In his public statements since launching SSI, Sutskever has been clear that he believes that the research direction that the large incumbent labs are pursuing — including his former company OpenAI — are destined to plateau and are not the best path to building superintelligence. He says he and SSI are working on something entirely new.

“We’ve identified a mountain that’s different from what I was working on,” he told a reporter last year. “Once you climb to the top of this mountain, the paradigm will change … Everything we know about AI will change once again.”

But nobody outside of SSI has any idea what this new paradigm is.

What could Ilya and SSI’s big idea possibly be?

Two obvious answers would be recursive self-improvement (AI systems that can build stronger AI systems, that can build stronger AI systems, and so forth) or continual learning (AI systems that can learn on an ongoing basis as they interact with the world).

Both of these fields address fundamental shortcomings of today’s AI systems, and both have become buzzy frontier research areas in recent months.

But we speculate that it’s something less consensus and more “out there” than these. We can’t wait to find out.

3. China’s domestic AI chip sector will make significant strides, planting the seeds for the eventual decline of Nvidia’s global dominance.

Imposing strict export controls on AI chips to China was one of the most important decisions to come out of the entire Biden administration. It was a bold and decisive move with clear logic. Unfortunately, this strategy looks likely to backfire spectacularly.

The key failure of this strategy boils down to a tradeoff between short-term impact and long-term impact.

Losing access to the world’s advanced AI hardware provided a painful but unmistakable wakeup call to the CCP that this was a set of technologies and capabilities that was too important for China not to control itself. And so, China has set out to cultivate its own domestic AI chip industry and to wean itself off its reliance on the West.

Making cutting-edge semiconductors is one of the most complex activities in which humanity engages. The accumulated intellectual property and technical knowledge inside an organization like Nvidia or TSMC is unimaginably vast and nuanced. It cannot be replicated overnight.

But China has many talented engineers, vast resources and a deep national commitment to this effort. In 2026, we predict that China’s domestic chip industry will make concrete, meaningful progress toward closing the gap with the U.S. on AI hardware. It will not yet achieve parity with Nvidia’s most advanced chips; it will not even come close, but by the end of 2026, it will be evident that China’s chip industry is on a productive path and is making steady progress toward the frontier of AI chip production.

4. Discourse about AGI and superintelligence will become less fashionable and less common.

Coming into 2025, expectations about AI’s trajectory and the timeline to artificial general intelligence were sky-high. The discourse was breathless.

Over the course of 2025, gradually but unmistakably, this has changed.

The vibe is shifting. Across the AI ecosystem, a consensus is emerging that superintelligent AI is likely not around the corner — and more to the point, that it may not matter that much. This technology is already extremely powerful. Well before the arrival of AGI, trillions of dollars of value creation are up for grabs as AI reshapes every industry and organization.

In 2026, we predict that this vibe shift will translate to noticeably less discourse about and interest in the concepts of AGI and superintelligence. It’s not that people will challenge or reject these concepts outright; they will just be less focused on them. AI leaders like Sam Altman, Dario Amodei, Sundar Pichai and Satya Nadella will spend less time talking about superintelligent AI and more time talking about enterprise AI adoption. Commentators and thought leaders will choose to opine on more proximate topics, from the geopolitics of AI to AI-driven job displacement. Go-to discussion topics at cocktail parties and around the watercooler will shift in a similar direction.

Discourse about AGI will not go away altogether in 2026 — Eliezer Yudkowsky is not going anywhere! — but it will become far less common.

5. A mundane and esoteric accounting concept — depreciation schedules — will become critically important, especially as debt plays a growing role in the AI infrastructure buildout.

AI is an exciting and futuristic space. Accounting is not. Yet a seemingly boring and obscure accounting concept will become critically important for the field of AI in 2026. Get ready to start hearing a lot about depreciation schedules for AI chips.

Historically, it has been common to depreciate chips, servers and other computing resources over a five-year period. In general, this has served as a reasonable assumption about how long chips last before they need to be replaced.

But in today’s AI era, things move faster than they ever have before.

Should companies that own AI chips — cloud providers like Amazon and Microsoft, AI labs like OpenAI and Anthropic, data center companies like Equinix and Digital Realty, neoclouds like CoreWeave and Nebius, among others — use a meaningfully shorter depreciation schedule than they historically have when accounting for their chip investments (say, one or two years)?

This debate will have significant real-world implications. It will influence the broader discourse about the profitability of AI and the long-term financial sustainability of the entire field. Longer depreciation schedules will support narratives that AI margins are improving and that AI-based businesses can be highly profitable. Shorter depreciation schedules, on the other hand, will strengthen the perception that AI, while it may be a powerful technology, is so capital-hungry that its economic returns and viability remain unclear.

If companies choose long depreciation schedules (say, five years) and then reality moves much faster (say, demand for those chips falls off a cliff after two years), this can lead to massive impairment charges that can abruptly transform a company’s or even sector’s financial health. An “impairment bomb” scenario like this is exactly what played out with the fibre overbuild in the early days of the internet.

The potential pitfalls are further amplified when debt levels ramp up, which is exactly what has happened over the past year in the world of AI infrastructure.

Treating AI chips as long-lived assets with long depreciation schedules can encourage the introduction of greater leverage into the system — possibly too much leverage.

Heading into 2026, as concerns continue to mount about a potential AI bubble and an infrastructure overbuild, these otherwise arcane accounting details will suddenly become keenly interesting to many people.

6. Many more AI companies will begin building custom chips.

Today, nearly all of the world’s AI organizations, devices and products are powered by chips from a remarkably small number of companies: Nvidia, Google, AMD, Amazon, a handful of others.

What if it didn’t have to be this way?

What if it were possible for every company to design and deploy its own custom chips, optimally suited for the particular products and use cases that that company is pursuing? Tradeoffs between energy efficiency, compute power, cost, form factor and more could all be optimized for each particular application.

Among the first movers will be the large AI labs, for whom purpose-built chips will become one more part of the technology landscape on which to innovate and compete in order to continue advancing the frontier of AI models. A few months ago, OpenAI announced a partnership with Broadcom to develop its own in-house chips. It is not crazy to imagine that, eventually, OpenAI will develop a new purpose-built chip for every new AI model generation that it trains, with the two co-optimized for one another.

But the big AI labs will just be the start. Robotics companies, consumer hardware companies, autonomous vehicle companies, BCI companies and beyond will increasingly begin planning to design their own AI chips.

One important driver of this trend will be the fact that cutting-edge reinforcement learning systems will make it possible to automate more and more of the chip design process, dramatically reducing the time and cost to develop a custom chip and therefore making it a viable option for many more companies. Today, designing a new chip can take two to three years. Imagine if that could be reduced to two to three weeks. Ricursive Intelligence, a buzzy new startup that came out of stealth  last month, is tackling this exact challenge.

7. Sam Altman will step aside as CEO of OpenAI.

In 2026, we predict that OpenAI’s board and leadership, including Altman himself, will come to the conclusion that Altman is no longer the best person to lead the organization as it prepares for life as a public company. Will OpenAI’s board fire Altman? Will Altman choose to step down and move on to his next chapter? The reality is that the public won’t know, and it won’t really matter. Unlike the dramatic events of November 2023, in which OpenAI’s board shocked the world by abruptly firing Altman, this transition will be tightly choreographed. It will be portrayed as Altman’s decision. He will announce the news on his terms.

The most logical choice as OpenAI’s next CEO will be Fidji Simo, who previously took Instacart public.

8. AI will be one of the central issues in the 2026 U.S. midterm elections. The politics will get complex, especially when it comes to AI-driven job loss.

In 2026, especially as we get further into the year, the news cycle in the United States will be dominated by the midterm elections. And these midterm elections will be dominated by the topic of artificial intelligence.

The political calculus around AI is fascinating and nuanced — and we may see it change as the year goes on.

AI-driven job loss, long discussed as a theoretical possibility, is rapidly becoming a present-day reality. Graduating college students are suddenly finding it difficult to get jobs because entry-level roles can increasingly be carried out by AI. Companies across sectors laid off tens of thousands of workers this year as a direct result of AI. And this is just the tip of the iceberg. According to an MIT study released last month, AI can now replace 11.7% of the U.S. workforce, representing over $1 trillion in wages.

Being on the wrong side of a job loss narrative is one of the cardinal sins of U.S. politics. And President Trump, though he was elected as a member of the Republican Party, has a serious populist streak in him. Trump will increasingly find himself compelled to side with, and to speak out in favor of protecting the jobs of, everyday working Americans against the encroachment of AI. This will complicate his and the Republican Party’s full-throated support of the technology and will muddy the waters in terms of their narrative around AI.

For their part, Democratic candidates will find it natural to advocate for policies that rein in AI’s deployment in order to mitigate AI-driven job loss. But Democrats, too, will face a complex balancing act. They cannot come out too unequivocally against the technology and risk being painted as blindly anti-innovation, anti-economic growth, and/or insufficiently focused on national security considerations and the competition with China.

Another policy issue that will present a tricky juggling act for many Democrat candidates: balancing their commitment to combat climate change with the pressure to unlock more domestic energy capacity for AI development, including from fossil fuel sources like natural gas.

9. One of the large global pharma companies will acquire one of the leading protein AI startups.

The big pharma companies are no strangers to partnering with AI startups on AI-powered drug development. Typically, this has taken the form of commercial deals whereby the startups use their AI to generate some candidates for the pharma company and in return receive a mix of upfront payments, milestone-based payments and future royalties. To this point, pharma companies have generally not acquired AI companies outright, preferring to only do big acquisitions when a startup has a specific therapeutic asset that the pharma company believes is worth paying for.

Why will this change in 2026?

Because AI’s promise is no longer merely hypothetical in this field — it is now really starting to work — and things will therefore start moving much more quickly. It will now become more compelling and even necessary for pharma companies to bring these AI platforms in-house, integrate them more tightly with their broader development and clinical pipelines, enable a faster flywheel of research and development, and preclude these startups from working with other pharma competitors.

As is the case in frontier AI more broadly, world-class talent in AI for protein design is incredibly rare. Only a handful of people on earth are capable of developing cutting-edge AI systems for de novo antibody design. And those people generally don’t choose to work at big pharma companies like Merck or Pfizer. Many of them are concentrated in a small group of leading protein AI startups. M&A will present a path for the big pharma players to bring this key talent in-house.

10. Brain-computer interfaces (BCI) will transition from a fringe frontier field to a mainstream technology and startup category. Neuralink’s position as the clear category leader will become shakier.

To most people, brain-computer interfaces sound like science fiction. People may be loosely familiar with Elon Musk’s Neuralink, but most generally assume that the technology is many years or even decades away from the real world.

In fact, this technology is rapidly nearing an inflection point in terms of real-world functionality. 2026 will be the year that that becomes broadly understood and that interest in BCI goes mainstream. Expect to see a wave of new BCI startups, a surge in venture capital dollars invested in BCI, meaningful clinical progress (though no FDA approvals yet) and a step-change increase in public discourse about the technology.

Read Rob’s past predictions for 2026 in full on Forbes. You can read Rob’s past predictions for 2025 and his end-of-year grades in his regular column on artificial intelligence for Forbes.