It’s finally happened: Apple has chosen Google’s A.I. to power the big new Siri upgrade that’s apparently coming later this year. The two juggernauts announced on Monday that they’ll be entering into a multiyear partnership wherein the next generation of Apple’s foundation models will be based on Google’s technology. No details yet on how much money will be changing hands, though it’s worth noting that this is not the first time the pair have partnered: Google pays Apple around $20 billion a year to be the default search engine on Safari (a deal that survived a major antitrust suit last year).

The partnership is yet another indication of Google’s increasingly dominant position relative to its A.I. competitors—specifically OpenAI, which currently has a Siri-related partnership with Apple. (We’ll see whether that one sticks around.) Either way, Apple’s failure to build its own A.I. may end up working in the company’s favor: L.L.M.s are fast becoming commodities; better to let someone else absorb the costs and risk.

Runner-up: Andreessen Horowitz just raised another $15 billion across a number of funds. American V.C. seems to be alive and well—for now.

And now for the main event…

Today, we’re going to take one more look at the months to come. I’ve talked to sources, gazed into the crystal ball, and communed with the ghost of Alan Turing in an attempt to distill all the information running through my brain into quantifiable predictions. And so begins an annual tradition here at The Hidden Layer. (As my partner Bill Cohan says, this is not investment advice!)

In the FOMO-fueled rush to get ahead, A.I. tools are advancing far more quickly than the industry’s security safeguards—and I fully expect this to be the year that bad actors find a way to exploit the gap. Indeed, according to a recent survey of more than 1,100 developers, a full 96 percent said that they don’t fully trust the code that their A.I. tools generate. Sixty-one percent agreed that “A.I. often produces code that looks correct but isn’t reliable.” But despite growing concerns about sensitive data and security vulnerabilities, they’re still shipping A.I.-generated code. (Is this a bad time to mention that the U.S. military will begin integrating Grok into Pentagon networks?)

Cybersecurity researchers have been warning me for years about the risks and multiplying potential points of attack—from bad code and non-coders accidentally leaking data, to prompt-injection attacks and even software engineers downloading third-party A.I. coding tools with malicious backdoors. With so much code now being generated by A.I., it’s just a matter of time

A number of major lawsuits filed at the tail end of 2025—including some that media companies hoped would fundamentally alter the course of A.I.—resulted in settlements. I expect the same fate will befall The New York Times Company’s lawsuit against OpenAI, which has long been heralded as the most significant A.I. copyright case. The discovery phase closes in February, but I don’t think this case will ever see 2027. More likely, it will end with some sort of partnership between the two companies (and all the other publishers attached to the case), similar to what happened with UMG and Udio, and with Warner Music and Suno.

It’s likely that many other lawsuits, copyright and otherwise, will come to similar conclusions. If they’re smart, plaintiffs will settle soon, before the bubble pops and the money goes away. (One of the few cases that won’t settle, however, is Elon Musk v. Sam Altman, which is headed for what promises to be a very juicy trial.)

I said this more than a year ago, and I stand by it now: 2026 will be the year the A.I. investment cycle begins to unwind, marked by a bear market for tech stocks. Will investors keep funneling money into A.I. cash incinerators? I don’t think so. Google will be fine. Private, unprofitable companies like OpenAI will fare worse.

These types of downturns usually require a catalyst—an enterprise pullback, for example, or the type of cybersecurity disaster I predict above. Or the correction could start with another debt load for the major startups to buy a million more chips that don’t pay off. Whatever ends up precipitating it, I think it’s inevitable, though I don’t think it will be comparable to the dot-com crash. Instead, it will resemble the 2022 correction, when the Nasdaq fell more than 30 percent. The bull market that’s been raging since 2023 has been relentless, but trees don’t grow to the sky, and A.I. is not exempt from historic financial trends. This year, the bull will stutter, and the Nasdaq will take a 20 percent-plus hit.

In the end, there will be only a couple of winners in A.I., and Apple will likely be the biggest. Among the eight largest companies in the world, Apple is the only one that has not poured hundreds of billions of dollars into data center infrastructure. Nor has it attempted to radically reconstruct its offerings to take advantage of the A.I. hype. Apple’s $3.8 trillion valuation, in my rough estimation, has almost nothing to do with A.I. Yet a not-insignificant portion of the world is wholly reliant on its suite of products, through which most people access A.I. technology.

Apple’s just-announced partnership with Google, integrating Gemini models into Siri, will be worth far more than whatever it ends up paying because the company will be more insulated than its peers when the market contracts. Bonus prediction: In the wake of this contraction, Apple will seize the chance to acquire a few discounted A.I. startups to boost its internal efforts to get a reliable, safe, and secure A.I.-powered Siri up and running. I’m quite sure that the company’s own work in A.I. isn’t yet done, just as I’m positive that it would rather not be reliant on another company’s technology in the long term.

Despite all the anecdotal stories, there are still no signs of major A.I.-related job losses, according to research by Oxford Economics Global. As the Oxford group wrote in a recent report: “While a rising number of firms are pinning job losses on A.I., other more traditional drivers of job layoffs are far more commonly cited.”

Economists have made all sorts of nuanced attempts to figure out what drives labor market behaviors, but the report’s authors boiled things down with a simple logic: “If A.I. were already replacing labor at scale, productivity growth should be accelerating. Generally, it isn’t.” Indeed, a recent survey by Apollo found that “the vast majority of C.F.O.s are seeing no impact from A.I. on labor productivity, decision-making speed, customer satisfaction, or time spent on high-value-added tasks.” I don’t expect this story to change this year.

In 2026, I suspect the industry will move away from the fiction of artificial superintelligence since the hype is no longer doing much to get enterprises to adopt the technology. Instead, marketing will shift to utility: what works, what doesn’t, and what it takes to make things work reliably. “A.G.I. is such a funny term,” Anthropic co-founder Daniela Amodei told CNBC recently. “Many years ago, it was kind of a useful concept to say, ‘When will artificial intelligence be as capable as a human?’ I think maybe the construct itself is now wrong—or maybe not wrong, but just outdated.”

I predict we’re going to see a lot more interest in novel architectures, creative systems, and approaches that take us away from L.L.M.s and transformers. Fundamentally, interest in small, scalable, highly efficient systems with a clear R.O.I. will grow out of necessity as much as anything else. Neurosymbolic systems, hybrid approaches, KANs, small language models—and some new thing that no one has heard about yet—are the approaches of the future.

Yann LeCun, who recently left Meta to launch his own startup, famously declared that L.L.M.s are a dead end, and that the future of A.I. development will revolve around other approaches, such as world models. I’m not sure that’s exactly right—L.L.M.s have their utility, though they are fundamentally inefficient. Perhaps their greatest value will be in reorienting the industry toward techniques that do the same thing, but with greater reliability and way more efficiency.

Perhaps the holy grail of the A.I. industry is medical discovery, as we saw recently with OpenAI’s and Anthropic’s back-to-back launches of healthcare-focused applications within ChatGPT and Claude, respectively. There’s been talk of cures for cancer and a long list of rare diseases; predictive, personalized healthcare; and a dramatically reduced timeline between drug discovery and mass production. The potential here is undeniable—but it’s still long-term. The reality is that biology is enormously complicated, much of the relevant data is difficult to access or produce, and there are systems in place (clinical protocols, the F.D.A., etcetera) that prevent breakthroughs at speed. We’re just not there yet, and we won’t be this year.

Since 2023, Washington has faced the question of how best to legislate a fast-moving technology that it doesn’t seem to understand. But, far from acting quickly, Congress has behaved much like a deer in the headlights—effectively paralyzed by the need to act. Meanwhile, of course, the Trump administration is trying to keep the sector as unregulated as possible, in line with the industry’s convenient narrative that regulating A.I. will ensure we’re all speaking Chinese by the end of the decade.

But even as A.I. becomes a more salient political issue, Congress is unlikely to pass any meaningful legislation to address it. That will come down to the states, whose efforts to regulate the technology in the year to come will draw them into what are likely to be intense, yearslong legal battles with Trump’s D.O.J.

While the gains offered by A.I. are—and will remain—both jagged and inconsistent with the hype, this year they’ll simply become expected. First, exposure to A.I.-generated content will become a consistent reality of the internet. A.I.-generated music has already infiltrated playlists and topped the charts, and will continue to do so; A.I.-generated videos and shorts are proliferating on YouTube and other social sites; emails and posts produced by generative A.I. are all over the place (and none too difficult to spot); and internet search results are likewise A.I.-generated, whether you search inside ChatGPT or go to Google. And Google’s share of the search market hasn’t dipped, reinforcing the idea that people are comfortable with—or at least tolerant of—receiving A.I.-generated results.

The term “A.I.” will remain, but fewer and fewer people will use it to refer to the fantastical seed that might become Skynet, or the digital realization of cognition; it’ll just be another app layer that does a thing. Not everyone needs it, not everyone wants it, and most people won’t really understand it… but it will be everywhere—a perpetual sidebar to our new, more energy-intensive internet.

That’s all for today. I’ll see you on Thursday.

Ian

Need help? Review our FAQ page or contact us for assistance. For brand partnerships, email ads@puck.news.

You received this email because you signed up to receive emails from Puck, or as part of your Puck account associated with {{customer.email}}. To stop receiving this newsletter and/or manage all your email preferences, click here.

Puck is published by Heat Media LLC. 107 Greenwich St., New York, NY 10006