TikTok CEO Shou Zi Chew told reporters on October 12, 2026, that he believes artificial intelligence could unlock major scientific breakthroughs in physics and biotechnology within the next two to three years, according to CNBC. The comments, which quickly became one of the day’s most discussed technology stories, position Chew alongside a growing list of AI and tech executives willing to put a specific timeline on AI-driven scientific progress rather than treating it as a distant hypothetical.
Chew said the long-term potential of AI is still underestimated, even after several years of breathless headlines about chatbots and generative models. He argued that AI could accelerate the removal of technological barriers that have historically slowed down scientific research, letting labs test more hypotheses, iterate faster, and tackle problems that would have taken human researchers decades to work through unaided. At the same time, he was careful to add a caveat that has become something of a running theme among AI executives this year: human judgment, creativity, and authenticity still matter, and increasingly capable AI systems need to operate within clear ethical standards.
What Chew Actually Said About AI and Science
The TikTok CEO’s remarks, as reported by CNBC, centered on two specific scientific fields: physics and biotechnology. Chew reportedly said both disciplines are positioned to see major breakthroughs within two to three years as AI tools mature, a timeline that is notably aggressive given how incremental progress in both fields has historically been. He did not name a specific AI product, lab, or research result tied to this prediction, and no further technical detail from the original remarks has been independently verified beyond what CNBC reported.
What makes Chew’s comments notable is less the specific number of years and more the fact that he’s willing to put a number on it at all. TikTok is not an AI research lab in the way DeepMind or Anthropic are, but its parent company ByteDance has been steadily building out generative AI infrastructure, including creative tools folded into TikTok’s advertising stack. Chew’s framing suggests that even executives running consumer platforms, rather than frontier model labs, are now comfortable publicly forecasting breakthrough science on a short timeline.
Chew also used the moment to reiterate a point he has made in prior public appearances: that human creativity and judgment remain essential even as AI systems get more capable. That is consistent with past remarks he made at the Qatar Economic Forum in Doha, where he called the current state of AI “truly very exciting and fascinating” and said it “could profoundly improve productivity across many things,” while also stressing that “we should invest in understanding the risks” as the technology evolves quickly, according to The Peninsula Qatar.
Why Physics and Biotechnology Specifically
Physics and biotechnology are an interesting pairing for a breakthrough forecast, because they represent two very different kinds of scientific bottleneck. Physics research, particularly in areas like materials science and particle physics, is often limited by the sheer computational cost of simulating complex systems. Biotechnology, by contrast, is frequently limited by the cost and speed of real-world experimentation: growing cells, running trials, and waiting for biological processes to play out in a lab. AI has been pitched as a solution to both bottlenecks, but through different mechanisms: faster simulation and modeling on the physics side, and faster hypothesis generation plus lab automation on the biotech side.
That dual framing echoes a theme that has come up repeatedly over the past year in statements from frontier AI lab leaders. Anthropic co-founder and CEO Dario Amodei argued at the 2025 World Economic Forum in Davos that AI could become “smarter than a Nobel Prize winner across most relevant fields” within two to three years, a timeline nearly identical to the one Chew cited for physics and biotech breakthroughs. Amodei went further on biology specifically, predicting that AI-enabled biology and medicine could compress 50 to 100 years of progress into just 5 to 10 years, a claim from a widely circulated summary of his Davos remarks that also covered comments from Demis Hassabis and Jensen Huang.
Google DeepMind CEO Demis Hassabis has staked out a somewhat longer and more structured version of the same bet. Hassabis has pointed to physics-aware world models and systems capable of running their own laboratory experiments as prerequisites for major scientific advances, and has talked about a “new renaissance” in discovery spanning materials science, drug design, and mathematics over a 10-to-15-year window, a timeline reported by Fortune in February 2026. DeepMind’s own track record lends some weight to that framing: Hassabis and collaborator John Jumper shared half of the 2024 Nobel Prize in Chemistry for AlphaFold’s protein-structure prediction work, with David Baker receiving the other half for computational protein design, an award the Royal Swedish Academy of Sciences announced on October 9, 2024.
How the CEO Predictions Stack Up
Chew’s two-to-three-year window for physics and biotech breakthroughs doesn’t exist in a vacuum. It lands in the middle of an unusually crowded field of public AI timeline predictions from executives who, in most cases, have direct financial and reputational stakes in AI living up to the hype. The table below lines up the major public forecasts from the past year, drawn from named outlets and events.
| Executive | Prediction | Timeline | Source / Venue |
|---|---|---|---|
| Shou Zi Chew (TikTok) | Major breakthroughs in physics and biotechnology | 2-3 years | CNBC, October 12, 2026 |
| Dario Amodei (Anthropic) | AI “smarter than a Nobel Prize winner” across most fields | 2-3 years | World Economic Forum, Davos, 2025 |
| Dario Amodei (Anthropic) | 50-100 years of biology/medicine progress compressed | 5-10 years | World Economic Forum, Davos, 2025 |
| Demis Hassabis (Google DeepMind) | “New renaissance” in materials science, drug design, math | 10-15 years | Fortune, February 2026 |
| Demis Hassabis (Google DeepMind) | Roughly 50% chance of AGI | Within the decade | World Economic Forum, Davos, January 2026 |
| Sam Altman (OpenAI) | Early versions of true superintelligence | “A couple of years” | India AI Impact Summit, February 2026 |
| Jensen Huang (Nvidia) | AI infrastructure for simulation and scientific deployment | Ongoing, no single date | CBS News / industry analysis, 2026 |
The pattern is striking once it’s laid out this way. Nearly every major AI-adjacent executive making a public forecast in the past 18 months has converged on a similar two-to-three-year window for the first wave of visible scientific payoff, even though their companies occupy very different positions in the AI stack. Chew runs a consumer platform. Amodei runs a frontier model lab focused on safety research. Huang runs the chipmaker that supplies nearly everyone else’s compute. That convergence doesn’t prove the predictions are correct, but it does suggest a shared industry narrative has taken hold: AI’s scientific payoff is framed as imminent rather than distant, a sharp break from the decades-long timelines scientists themselves have traditionally used for fields like physics and drug discovery.
The Track Record So Far: What AI Has Actually Delivered
Skepticism about these timelines is reasonable given how slowly some of AI’s most publicized scientific claims have held up under scrutiny. OpenAI’s experience earlier this year offers a useful cautionary data point. The company released 722 math manuscripts from an unreleased model, a dump that initially drew comparisons to a breakthrough on the scale of a “country of geniuses,” but independent verification told a more complicated story. When mathematicians ran the claimed results through the Lean formal-verification system, only about 42% of the papers cleared the checks, and a group of Fields Medalists pushed back publicly on the broader claims. The Tao Group, among others, scrutinized the 722 claims directly, finding that the initial framing overstated what had actually been proven.
That episode matters here because it shows the gap between an executive’s forecast and a verified scientific result can be wide, even when the underlying model genuinely represents progress. AlphaFold’s Nobel Prize is the clearest counterexample: a case where an AI system’s output was independently validated, peer-reviewed, and recognized by one of science’s most conservative institutions. The difference between AlphaFold’s trajectory and OpenAI’s math dump is instructive. AlphaFold took years of iterative publication and external validation before the Nobel committee recognized it in 2024. The math papers, by contrast, arrived all at once with limited independent verification baked in, and the subsequent scrutiny deflated some of the initial hype.
Anthropic’s own research offers a middle case. The company’s Claude model was reported to have reached 67.2% on a Riemann zeta bound problem, a genuine mathematical result that fell short of solving the Riemann Hypothesis itself but represented measurable progress on a problem mathematicians have worked on for more than 150 years. That kind of incremental, independently checkable progress is arguably a better predictor of where “two to three years” of AI-assisted science actually lands than a flashy announcement of hundreds of simultaneous results.
Market and Industry Context: Why Everyone’s Talking Timelines Now
Chew’s comments arrive during a period of enormous capital commitment to AI infrastructure, which gives these timeline predictions real financial stakes beyond press-cycle bragging rights. Microsoft, Meta, and Google have collectively pushed AI capital expenditure past $550 billion in 2026, spending that only makes long-term sense if the technology eventually delivers returns well beyond chatbots and ad targeting. Scientific breakthroughs, especially in high-value fields like drug discovery and materials science, are one of the few categories of payoff large enough to justify that level of spend. When a CEO as prominent as Chew puts a number on how soon that payoff might arrive, it reinforces the investment thesis that’s currently justifying hundreds of billions of dollars in data center buildouts.
There’s also a competitive dimension. Nvidia CEO Jensen Huang has framed his company’s role as supplying the training, simulation, networking, and industrial deployment capacity that physics and biotech research would need to benefit from AI at scale, according to an analysis published by Educational Technology Journal in September 2026. Huang has also pushed back on doom-focused framing of AI risk, telling CBS News there’s “zero percent chance” of AI causing civilizational collapse, a framing that sits in tension with the industry’s simultaneous push for safeguards. Microsoft CEO Satya Nadella, for instance, has called for AI systems to include six specific safeguards, including an “emergency brake” mechanism, reflecting a split in how different executives balance optimism about breakthroughs against caution about losing control of increasingly autonomous systems.
Sam Altman has arguably gone furthest on the optimism side, previously declaring that the “AI singularity” has effectively arrived and calling it “awesome for the world.” Altman’s framing, combined with Chew’s and Amodei’s specific two-to-three-year windows, paints a picture of an industry leadership class that has largely stopped hedging on when AI’s biggest scientific payoffs will show up, even as independent verification of specific claims, like the OpenAI math papers, continues to lag behind the rhetoric.
TikTok’s Own AI Push Provides Some Context
Chew’s comments about scientific breakthroughs didn’t come out of nowhere in terms of his own company’s AI trajectory. TikTok and parent company ByteDance have spent 2026 steadily expanding AI tooling across the platform’s advertising and content ecosystem. TikTok’s Symphony suite has integrated ByteDance’s Dreamina Seedance 2.5 video model to support AI-powered creative production for advertisers, and the company introduced a Business Model Context Protocol that lets developers build AI agents that interact directly with TikTok’s advertising systems. Those are commercial products, not scientific research tools, but they illustrate that Chew is overseeing a company actively investing in applied AI across its own business, giving his broader comments about AI’s scientific potential some grounding in direct operational experience with how fast the underlying models have been improving. ByteDance’s leadership has made similar points about the pace of change before: the company’s chief has previously warned against mediocrity as AI brings disruption across the industry, according to WTAQ News Talk.
The Ethics Caveat: A Recurring Theme From Chew
One part of Chew’s October 12 remarks deserves particular attention: his insistence that human judgment, creativity, and authenticity remain important even as AI systems become more capable, paired with a call to ensure increasingly powerful AI follows ethical standards. This isn’t a new position for him. At a prior appearance tied to Nanyang Technological University, Chew told graduates that “this is a pivotal and exciting point in history, where change is not constant, but exponential and compounding,” a remark catalogued on his Wikiquote page. He has also previously said that “some form of thoughtful, careful regulation is necessary to ensure that we’re building guardrails, but at the same time not killing off the innovation.”
That consistent pairing of enthusiasm with a call for guardrails separates Chew’s public rhetoric somewhat from the more unambiguously bullish framing used by some AI lab executives. Where Altman talks about an arrived singularity and Huang dismisses doom scenarios outright, Chew’s comments on October 12 kept the ethical caveat front and center, suggesting TikTok’s CEO is trying to occupy a middle position between AI boosterism and AI caution as he makes his own breakthrough prediction.
Competitive Comparison: How the Major AI Labs Are Positioned for Scientific AI
Separate from the executive rhetoric, it’s worth looking at how the labs actually most likely to deliver on a two-to-three-year scientific breakthrough timeline are positioned today. The comparison below focuses on verified capabilities and public commitments rather than forecasts.
| Organization | Scientific AI Focus | Verified Result or Commitment | Status |
|---|---|---|---|
| Google DeepMind | Protein structure prediction, drug design, automated labs | 2024 Nobel Prize in Chemistry (Hassabis, Jumper); UK automated lab agreement targeted for 2026 | Independently validated |
| Anthropic | Mathematics, model safety research | Claude reached 67.2% on a Riemann zeta bound problem | Partial, independently checkable |
| OpenAI | Mathematics research dump | 722 manuscripts released; only about 42% cleared Lean formal verification | Disputed by mathematicians |
| Nvidia | Compute infrastructure for simulation and training | Supplies training, simulation, and industrial deployment hardware broadly used across the sector | Infrastructure, not a direct scientific claim |
| TikTok / ByteDance | Applied generative AI for commercial products | Symphony/Seedance 2.5 integration, Business Model Context Protocol for ad tooling | Commercial, not scientific research |
The contrast is useful context for evaluating Chew’s prediction. DeepMind is the only organization on this list with an actual Nobel Prize tied to AI-assisted science, and even that took years of published, peer-reviewed work before the recognition arrived. Anthropic’s math result is real but partial. OpenAI’s splashiest scientific claim of the year has faced the most public pushback. TikTok, meanwhile, has no direct scientific research output of its own; Chew’s prediction is a forecast about the broader field, not a preview of something ByteDance is building internally.
Historical Context: AI Timeline Predictions Have a Mixed History
It’s worth remembering that specific AI timeline predictions from industry leaders have not always aged well. Hassabis himself offers a useful internal example: he had originally targeted the end of 2025 for the first AI-designed drug to enter human trials, but acknowledged at the January 2026 World Economic Forum that the milestone had slipped to the end of 2026, with oncology candidates leading the pipeline. That kind of slippage, from a lab with one of the strongest track records in the field, is a reminder that even well-resourced, scientifically grounded predictions routinely run into real-world delays: regulatory review, experimental setbacks, and the simple difficulty of biology not cooperating on a press-release schedule.
Broader AGI and superintelligence timelines carry the same pattern of optimistic executive forecasts running ahead of more conservative outside assessments. A survey of 2,778 AI researchers gave human-level AI only a 50% chance by 2047, a timeline decades longer than the “couple of years” Altman has floated for early superintelligence. That gap between researcher consensus and executive rhetoric is one of the more consistent features of the AI timeline conversation over the past several years, and it’s a useful frame for evaluating Chew’s two-to-three-year window on physics and biotech breakthroughs as well.
What a “Breakthrough” Would Actually Need to Look Like
Part of the difficulty in evaluating Chew’s prediction, or any of the comparable executive forecasts, is that “breakthrough” is doing a lot of undefined work. A Nobel-caliber discovery is different from a useful but incremental advance, and the AI industry has repeatedly blurred that distinction in its own announcements. The OpenAI math papers episode is the clearest recent example: the initial framing implied breakthrough-level results, but independent Lean verification found that well under half of the claimed proofs actually checked out, and prominent mathematicians said the framing had outpaced the substance.
For physics specifically, a genuine breakthrough attributable to AI would likely need to take the form of either a materials discovery that gets independently synthesized and tested, or a simulation-driven insight that leads to a testable, falsifiable prediction later confirmed experimentally. For biotechnology, the bar is somewhat clearer because there’s already a template: an AI-designed drug candidate that clears preclinical testing and enters human trials, the exact milestone Hassabis pushed from late 2025 into late 2026. Whether Chew’s two-to-three-year window produces something at that level of rigor, rather than another contested announcement, will be the real test of whether his prediction holds up better than OpenAI’s math claims did.
Predictions: What Happens Next
- Expect more non-lab executives, not just AI company founders, to start publicly forecasting scientific AI timelines over the next year, as the narrative shifts from “AI is coming for science” to specific, quotable year counts.
- Independent verification efforts, like the Lean formal-checking process applied to OpenAI’s math papers, will likely become a standard part of how major AI scientific claims are evaluated, reducing the gap between announcement hype and confirmed results over time.
- DeepMind’s automated UK laboratory, targeted for 2026, is likely to become a reference point other labs and executives cite when making their own breakthrough timeline claims, simply because it’s one of the few concrete, named infrastructure commitments in this space.
- Biotechnology will likely produce more verifiable near-term progress than physics, given that the drug-trial pipeline offers clearer, externally tracked milestones (preclinical to human trials) compared to open-ended physics simulation claims.
- Expect continued friction between AI optimism (Altman’s “singularity,” Huang’s dismissal of doom scenarios) and AI caution (Nadella’s safeguards, Chew’s repeated calls for ethical guardrails), with neither camp fully winning the public narrative through 2027.
Frequently Asked Questions
What exactly did TikTok CEO Shou Zi Chew say about AI and science?
Chew said AI’s long-term potential is underestimated and that it could drive major breakthroughs in physics and biotechnology within the next two to three years, according to CNBC’s October 12, 2026 report. He also said AI may accelerate the removal of technological barriers in scientific research.
Did Chew name a specific AI model or research result?
No. The CNBC report attributed a general timeline and framing to Chew but did not identify a specific product, lab, or scientific paper tied to the prediction.
How does Chew’s prediction compare to other AI executives?
It’s close to Anthropic CEO Dario Amodei’s 2025 Davos prediction that AI could be “smarter than a Nobel Prize winner” within two to three years. It’s shorter than Google DeepMind CEO Demis Hassabis’s 10-to-15-year “new renaissance” framing for broader scientific discovery.
Has AI already produced any Nobel-level scientific results?
Yes. Demis Hassabis and John Jumper shared half of the 2024 Nobel Prize in Chemistry for AlphaFold’s protein-structure prediction work, with David Baker receiving the other half for computational protein design, as announced by the Royal Swedish Academy of Sciences on October 9, 2024.
Why did OpenAI’s math breakthrough claims face pushback?
OpenAI released 722 math manuscripts from an unreleased model, but when checked against the Lean formal-verification system, only about 42% of the papers cleared verification, prompting criticism from mathematicians, including scrutiny from the Tao Group.
Does Chew’s prediction have anything to do with TikTok’s own AI products?
Not directly. TikTok and ByteDance have been expanding applied AI tools for advertising and content creation, including the Symphony suite’s integration of the Dreamina Seedance 2.5 video model, but those are commercial products rather than scientific research.
What ethical concerns has Chew raised alongside his AI predictions?
Chew has repeatedly said human judgment, creativity, and authenticity remain important as AI develops, and that increasingly capable AI systems need to follow ethical standards. He has previously called for “thoughtful, careful regulation” that builds guardrails without stifling innovation.
Is a two-to-three-year AI breakthrough timeline realistic?
It’s contested. DeepMind’s own drug-trial milestone slipped from late 2025 to late 2026, and a survey of 1,580 AI researchers put only a 50% chance on human-level AI by 2042, suggesting executive timelines often run well ahead of broader research consensus.
