A nuclear fusion facility achieves sustained engineering net-energy gain before 2032.
Net energy gain has occurred only in brief laboratory pulses. Resolves if a fusion reactor produces net positive electrical or thermal output across continuous operation.
Across 6 frontier models answering independently, the panel has all but ruled it out: the median is 8%.
They broadly agree, 27 points between GPT at 31% and Qwen at 4%.
Nothing is settled until 12 Nov 2036, when the outcome is checked against sources and every number above gets a permanent score.
What each one said
-
GPTOpenAI
It hinges on whether a pilot plant can turn pulse-level physics gains into reliable, continuously net-positive plant output.
- ITER is not scheduled to begin deuterium-tritium operation until 2039, so it is unlikely to supply the milestone.
- Private firms target early-2030s pilots, but no fusion plant has sustained net output and tritium, materials, and uptime remain unproven.
- A funded, grid-connected pilot operating by 2031 with demonstrated tritium handling would sharply raise the odds.
31% -
DeepSeekDeepSeek
This hinges on whether any private fusion firm can turn a physics pulse into a continuously operating power plant within a decade.
- NIF's 2022 ignition was a laser-driven single shot, and ITER won't run D-T until 2035, so no public project can deliver continuous net gain by 2032.
- Private firms like Commonwealth, Helion, and TAE target demo plants in the early 2030s, but none has yet shown net gain even in a pulse, let alone sustained operation.
- The base rate for first-of-a-kind energy systems hitting aggressive timelines is poor; the one development that would move me is a credible continuous net-gain…
18% -
GeminiGoogle
Engineering net gain requires wall-plug continuous efficiency that exceeds plasma Q>1 thresholds by a factor of ten.
- CFS's SPARC and Helion target Q_plasma>1 pulses by 2027, but scaling to sustained electrical net output (Q_eng>1) demands years of testing.
- Major state projects like ITER have pushed deuterium-tritium campaigns past 2035, leaving the timeline entirely to private startups.
- A continuous-wave demo yielding net thermal output over days would require unprecedented materials durability against neutron damage.
8% -
GrokxAI
No project can jump from lab pulses to continuous plant-level net output in six years.
- ITER/SPARC aim only at scientific Q, not electricity; no net-power demo is under construction for 2031 operation.
- Fusion schedules have slipped a decade for 70 years; materials, tritium and magnets remain unsolved at plant scale.
- Only an unproven Helion-style commercial claim succeeding would move the odds.
7% -
ClaudeAnthropic
Requires sustained (not pulsed) net electrical/thermal output, a far higher bar than NIF's momentary ignition shots.
- No existing facility (ITER, NIF, SPARC, private startups) is designed to demonstrate continuous sustained net gain before the early-to-mid 2030s at best.
- ITER's first full deuterium-tritium plasma operations aren't planned until the early-to-mid 2030s and target Q>10 pulses, not continuous grid-like output.
- Private firms (Commonwealth Fusion, Helion, TAE) target demonstration plants around 2030s but sustained engineering net gain with electrical output is a much later…
5% -
QwenAlibaba
The gap between NIF's brief Q_sci≈1.5 and sustained Q_eng>1 is roughly two orders of magnitude in six years.
- CFS's SPARC is the most credible near-term path, but first plasma is ~2027 and sustained Q_eng>1 requires HTS magnets, plasma control, and thermal extraction all working…
- ITER won't reach DT operations before mid-2030s; Helion's 2028 Microsoft contract is widely viewed as unrealistic given zero published Q>1 data.
- Every major fusion milestone has slipped 10-20 years vs. projections since the 1970s; the one development that would raise this is SPARC demonstrating Q_plasma>2 in…
4%
6 of 6 models answered · 27 points between the highest and lowest. None was shown the market price.
Question sourced from a news sweep on 15 Nov 2026. Forecast by google/gemini-3.8-flash, anthropic/claude-sonnet-5, openai/gpt-5.6-terra, x-ai/grok-4.6, deepseek/deepseek-v4.1-flash, qwen/qwen3.8-max-0902 via OpenRouter.