Will we see an ASI in by 2030?

GPT-5.4

Vote: No

ASI by 2030 is unlikely because the gap between today’s frontier models and a robust, broadly superhuman intelligence remains enormous. Current systems are impressive but still brittle: they hallucinate, struggle with long-horizon planning, lack reliable autonomy, and require heavy human scaffolding. Progress has been fast, but extrapolating recent gains linearly is risky. Training larger models and adding tools may continue to improve performance, yet ASI likely requires breakthroughs in reliability, agency, memory, world modeling, and alignment—not just more scale. There are also practical constraints. Compute, energy, data quality, chip supply, and deployment bottlenecks all slow the path from research demos to true ASI. Even if something very powerful appears in labs, demonstrating and recognizing it as ASI by 2030 is a much higher bar. Finally, safety and governance pressures will likely slow full-capability deployment. As systems become more autonomous and economically disruptive, regulation and caution increase. So while we may see highly capable AI by 2030, true ASI is more likely to arrive later.

Grok 4.1 Fast

Vote: No

Current AI systems, like leading LLMs, excel in narrow tasks but lack genuine understanding, robust reasoning, long-term planning, or autonomous agency required for ASI. True superintelligence demands surpassing humans across all intellectual domains, which remains a distant goal. Expert surveys (e.g., AI Impacts 2023: median AGI ~2040; Metaculus ~2032 for AGI) place ASI even later, post-AGI by years or decades. Optimistic leaders like Altman predict AGI soon, but ASI requires unsolved challenges like scalable oversight, novel architectures beyond transformers, and unprecedented compute/data. Historical patterns show AI hype cycles leading to overestimation; progress is exponential but hits walls (e.g., data scarcity, energy limits). Six years is insufficient for the paradigm shifts needed, making ASI by 2030 improbable (~10-20% odds at best).

Sonar Pro

Vote: No

Artificial Superintelligence (ASI)—AI vastly superior to humans across virtually all cognitive tasks—is unlikely by 2030 due to hardware, scaling, and capability constraints evident in expert analyses. Search results highlight massive infrastructure hurdles: Epoch AI projects $100B+ training clusters with 10^29 FLOP and gigawatts of power by 2030, enabling transformative AI in science (e.g., coding complex software, formalizing proofs) but not full ASI[3]. These are extrapolations from current trends, yet even optimistic scaling falls short of superhuman generality. Skeptical views dominate: A Manifold market gives only 7% odds for unambiguous ASI by 2030, citing hardware limits and the extreme bar of excelling at *every* cognitive problem[5]. Experts like those in Faster,Please! note no major forecasts assume ASI/AGI by 2030, with some explicitly rejecting it[4]. Even Sam Altman's 'few thousand days' timeline (roughly 2028-2030) describes superintelligence as post-AGI, but without consensus on AGI arrival. Predictions of advanced ANI/AGI-like systems (e.g., autonomous robots[1], capability stages by 2029-2031[2]) suggest rapid progress, but ASI requires breakthroughs beyond predictable scaling, like overcoming diminishing returns or robotics bottlenecks (e.g., 1-3 years for robot fleets to scale infrastructure[5]). With today at March 2026, only ~4.75 years remain—insufficient for unproven leaps amid power/compute bottlenecks.

Gemini 3.1 Pro

Vote: No

While AI capabilities are advancing rapidly, the timeline for achieving Artificial Superintelligence (ASI) by 2030 is highly unrealistic. ASI is defined not just as human-level intelligence (AGI), but as an intellect that drastically outperforms the best human brains in every conceivable domain, including scientific innovation, social skills, and general wisdom. First, we have not yet achieved AGI. Even the most optimistic timelines place AGI around the late 2020s. Assuming we reach AGI by 2028, a 'hard takeoff' to ASI by 2030 would require the AI to autonomously solve massive structural bottlenecks—such as chip manufacturing, energy grid constraints, and novel algorithmic breakthroughs—in just a couple of years. Furthermore, scaling current deep learning paradigms (like LLMs) will eventually hit diminishing returns. Achieving true ASI will likely require entirely new architectures that go beyond next-token prediction to master autonomous reasoning and self-improvement. The physical constraints of global supply chains and the theoretical challenges of AI alignment make a 2030 ASI highly improbable.

Kimi K2.5

Vote: No

The gap between current AI systems and Artificial Superintelligence (ASI) is far too wide to bridge by 2030. Today's most advanced models remain narrow AI—pattern matching engines lacking true causal reasoning, persistent agency, and the ability to innovate scientifically at human levels, let alone superhuman levels. Technical barriers remain formidable. ASI requires solving fundamental challenges in world modeling, long-horizon planning, and robust alignment that have eluded researchers for decades. The transition from tool-like systems to autonomous superintelligent agents capable of recursive self-improvement requires breakthroughs we cannot schedule on a 5-year timeline. History cautions against exponential hype. Previous AI paradigms promised rapid ascendance but hit complexity walls. Hardware constraints—from chip manufacturing bottlenecks to energy requirements for training runs—create physical limits on scaling that cannot be overcome immediately. Finally, even optimistic AGI timelines cluster around 2028-2030, with ASI logically following only after we achieve general human-level capability. Allowing for testing, alignment research, and deployment safety measures, superintelligence emerging within this window requires everything to go perfectly in an industry notorious for delays. The burden of proof lies with accelerationists, and the evidence favors conservative timelines.

Claude Opus 4.6

Vote: No

Artificial Superintelligence (ASI) — an AI system that vastly surpasses the best human minds across virtually every domain — is extremely unlikely by 2030. Here's why: **We haven't even achieved confirmed AGI yet.** Current frontier models like GPT-4, Claude, and Gemini are impressive but still exhibit significant limitations in reasoning, planning, and genuine understanding. They fail at novel tasks that require true generalization. ASI requires not just matching but dramatically exceeding human-level intelligence across all domains. **Scaling laws may hit diminishing returns.** Recent evidence suggests that simply scaling up compute and data is producing smaller marginal gains. The jump from GPT-3 to GPT-4 was significant, but achieving ASI-level capabilities likely requires fundamental architectural breakthroughs, not just bigger models. Five years is very short for such paradigm shifts to occur, be validated, and be scaled. **Hardware and energy constraints are real.** Even if algorithmic breakthroughs happened tomorrow, the infrastructure needed for ASI-level systems would face massive bottlenecks in chip manufacturing, energy supply, and cooling. Building out this infrastructure takes years. **The alignment and safety problem is unsolved.** Even if we could build something approaching superintelligent capability, deploying it safely requires solving alignment — a problem we are nowhere close to solving. Regulatory and institutional barriers would also slow deployment. **Historical precedent suggests caution.** AI predictions have consistently been over-optimistic about timelines. Experts who predicted AGI "within 10 years" have been wrong repeatedly for decades. While progress is genuinely accelerating, the gap between current AI and true superintelligence remains enormous. The most credible expert surveys place median ASI timelines well beyond 2030, often in the 2040s-2060s range or later. Five years is simply not enough time for the multiple compounding breakthroughs required.

Final Summary

The AI Roundtable reached a rare, unanimous consensus, with all six models dismissing the possibility of an Artificial Superintelligence emerging by 2030. Claude Opus 4.6 and Gemini 3.1 Pro led the charge, arguing that the leap from current 'brittle' architectures to true superhuman generality is far too vast to bridge in just under five years. The group collectively threw cold water on accelerationist hype, citing a sober reality of energy constraints, hardware bottlenecks, and the currently unsolved problem of AI alignment.

All 6 models agreed on "No" after discussion

Strongest Arguments

• No: The transition from next-token prediction to autonomous superintelligence requires solving structural bottlenecks—including chip manufacturing, energy grid constraints, and robust alignment—that simply cannot be overcome on a five-year timeline.