Two celestial bodies orbiting each other trace stable, predictable paths. Add a third, and the system turns chaotic: each body is large enough to bend the others' paths, and a small nudge by one can swing the trajectory of all three. That is how the author of a Fortune commentary frames the AI economy in 2026 — as three bodies pulling on one another: the closed-source frontier labs, led by OpenAI and Anthropic; open-weight models, mostly out of China; and the application companies built on top of both. Each is powerful enough to reshape the others' orbit, but none can dictate where the system finally settles.

Several events over the past weeks and months have increased the instability in the system. The most dramatic shift is the momentum built — and the new obstacles faced — by the frontier labs. Led by Anthropic, they have seen unprecedented demand and, with it, extraordinary revenue growth. That expansion has helped spread AI tools more widely through the economy, but it has also sharpened the question many buyers are now asking: what is the return on all this spending? By the author's estimate, spending on AI now runs somewhere between 0.5 and 1 percent of all white-collar salaries in the United States — a scale that, in the author's view, deserves scrutiny.

In July, Palantir's Alex Karp told CNBC that "something has gone completely wrong" with how the labs sell their product: enterprises, he argued, are "tokenmaxxing" — spending furiously on tokens with no matching gain in productivity. Competition at the frontier has intensified over the same period, with Meta (Muse Spark 1.1) and xAI (Grok 4.5) both fielding increasingly capable models alongside Anthropic, OpenAI, and Google.

Open models, especially those from Chinese companies, have also improved. Zhipu's GLM 5.2 and Moonshot's Kimi K3 now perform at or near the frontier on several important benchmarks. Priced at a fraction of comparable closed models while sitting so close in capability, they have created strong momentum for the open-weight ecosystem. U.S. open-weight models are adding credibility of their own: led by Thinking Machines' Inkling and Nvidia's Nemotron 3 — highly capable, if not yet quite at the frontier — they offer a domestic alternative to the Chinese releases. For application companies, that broader menu matters because model choice affects both costs and control, which is why many of the leading ones have ramped up efforts to build on top of open-weight models.

Three-body systems are notoriously difficult to predict. Nevertheless, the author discerns several broad trajectories over the course of the second half of 2026.

First, the discomfort with frontier pricing will ease — partly because competition will push prices down, and partly because the returns on AI spending will begin to show. Much of today's anxiety, the argument goes, reflects a timing mismatch: adoption is running ahead of utility. For most prior technologies, such as cars and cell phones, mass adoption followed declines in prices. In the case of AI, adoption happened much faster, so the pressure to prove value is arriving while usage is still scaling. The payoff will come, the author writes, as faster growth for some companies, cost savings for others, and eventually higher productivity across the economy.

Second, the shift toward a multi-model world will continue, driven by competition and, ideally, by real differentiation in what each model does best.

Third, U.S. open-weight models will become genuine alternatives to the Chinese ones and win real adoption as a result. They will also have a clearer business model, making it easier for customers to make longer-term bets. The author expects the distinction between open and closed eventually to diminish, as the frontier labs themselves support model personalization for the specific needs of their customers.

In other ways, too, the players will converge: frontier labs going deeper into the product stack to widen their moats and sustain high margins, and application companies going deeper into the model stack to build moats of their own. That convergence is rational, the commentary argues: software companies typically enjoy 70%+ gross margins while customers feel they get their money's worth from the product.

Given these moves and countermoves, the system remains a three-body one, and its equilibrium is still unsettled. Much of today's noise — the debate over open versus closed, China panic, and the hand-wringing over returns — looks temporary. The genuinely interesting question, the author concludes, is not whether AI pays off, but who captures the value when it does: the labs at the frontier, the open models nipping at their heels, or the applications that own the customer.

The opinions expressed in Fortune.com commentary pieces are solely the views of their authors and do not necessarily reflect the opinions and beliefs of Fortune.

This story was originally featured on Fortune.com.