FUSION-OS / TELEMETRY ONLINE / INDEX 2026.05聚变 OS · 遥测在线 · 索引 2026.05

The operating system of fusion civilization.聚变文明的操作系统。

An interactive command center for controlled nuclear fusion — every reactor, every approach, every number, every equation. From the Lawson criterion to Kardashev I.受控核聚变的交互指挥中心——每一座反应堆，每一条路线， 每一个数字，每一道方程。从劳逊判据，直至卡尔达肖夫一型文明。

Reactors tracked装置数
12+
Companies profiled公司数
12
Best Q (lab)实验最高 Q
1.54
NIF · 2022NIF · 2022
Longest plasma最长脉冲
1066 s
EAST · 2025EAST · 2025
→ ENTER GRID→ 进入电网RUN LAWSON SIM运行劳逊模拟VIEW COMMAND CENTER指挥中心
02/SECTION章节

Planetary Fusion Grid行星聚变网格

Every major device on one map. Click a node — read the device, its physics, its place in the race.所有主要装置在一张图上。点击节点，读出它的物理参数、它在竞赛中的位置。

PLANETARY FUSION GRID行星聚变网格
LIVE TELEMETRY实时遥测
ITER
Operating运行中
Construction建造中
Assembly组装
Concept概念
Decommissioned退役
DEVICE READOUT装置读数

ITER国际热核聚变实验堆

ITER Organization (35 nations) · France
Construction建造中Tokamak托卡马克D-T (planned)
R · major radius大半径
6.2 m
a · minor radius小半径
2 m
B · toroidal field环向场
5.3 T
V · plasma vol等离子体体积
840 m³
Iₚ · plasma current等离子体电流
15 MA
τpulse · pulse length脉冲长度
400 s
Tᵢ · best ion temp最高离子温度
20 keV
Q · achieved/target实测/目标
10
t₀ · first plasma首次等离子
2034

Largest tokamak ever built. Designed for Q=10 (500 MW out / 50 MW in). First D-T burn slipped to 2039 in the 2024 baseline reset.史上最大托卡马克。设计 Q=10（500 兆瓦输出 / 50 兆瓦输入）。2024 基线重设后首次氘氚燃烧推迟至 2039。

03/SECTION章节

Reactor Database装置数据库

Confinement type, fuel cycle, magnetic field, plasma current, pulse length, achieved Q. Sortable. Filterable. Sourced from public records (ITER, IAEA, EUROfusion, NIF, IPP, ASIPP, KFE, QST, FIA).约束方式、燃料循环、磁场、等离子体电流、脉冲长度、实测 Q。可排序、可筛选。数据取自公开档案（ITER、IAEA、EUROfusion、NIF、IPP、ASIPP、KFE、QST、FIA）。

Device装置↑Type类型Fuel燃料R (m)RB (T)BIₚ (MA)Iₚτpulse (s)τ脉冲Best QQFirst plasma首次Status状态
CFETR中国聚变工程实验堆
ASIPP · Chinese MOST
TokamakD-T7.26.514—252035Concept
EAST东方超环
ASIPP · Chinese Academy of Sciences
TokamakD-D1.853.511066 s—2006Operating
ITER国际热核聚变实验堆
ITER Organization (35 nations)
TokamakD-T (planned)6.25.315400 s102034Construction
JET联合欧洲环
EUROfusion · UKAEA
TokamakD-T2.963.454.85 s0.671983Decommissioned
JT-60SAJT-60SA
QST · F4E
TokamakD-D2.972.255.5100 s—2023-10Operating
KSTAR韩国超导托卡马克高级研究装置
Korea Fusion Energy Institute
TokamakD-D1.83.52102 s—2008Operating
MAST-UMAST 升级
UKAEA · STEP precursor
Spherical TokamakD-D0.850.7825 s—2020Operating
NIF国家点火装置
Lawrence Livermore National Lab
ICFD-T———10 ns1.52009Operating
PolarisPolaris
Helion Energy
FRCD-He3———1 ms—2025Assembly
SPARCSPARC
Commonwealth Fusion Systems · MIT PSFC
TokamakD-T1.8512.28.710 s112027Construction
TFTRTFTR
Princeton Plasma Physics Lab
TokamakD-T2.563—0.271982Decommissioned
Wendelstein 7-X文德尔施泰因 7-X
Max Planck IPP
StellaratorD-D5.53—480 s—2015Operating
Tokamak
Toroidal magnetic confinement, axisymmetric
Stellarator
Twisted helical magnetic confinement, no plasma current
ICF
Inertial confinement: laser/ion-driven implosion
FRC
Field-reversed configuration: closed-line plasmoid
Z-pinch
Sheared-flow stabilized linear pinch
Spherical Tokamak
Low-aspect-ratio tokamak (R/a < 2)
MTF
Magnetized target fusion (compress magnetized plasma)
04/SECTION章节

Companies & Capital公司与资本

The 12 private fusion programs that matter — founders, investors, devices, schedules, and the engineering bottleneck each one is actually fighting.12 家真正关键的私营聚变项目——创始人、投资人、装置序列、节奏，以及每一家在实际硬战的工程瓶颈。

Tracked here本表收录
12
of ~50 globally（全球约 50 家）
Capital raised累计融资
$5.97 B
across this list（本表内）
Industry total行业总额
$9.8 B
FIA 2025 figures（FIA 2025）
Targeting 2030s grid目标 2030s 入网
14
across the field（全行业）
Founders创始人
Bob Mumgaard, Brandon Sorbom, Dennis Whyte, Martin Greenwald
Key investors主要投资人
Breakthrough Energy, Tiger Global, Google, Equinor, Eni, JIMCO, Khosla, Future Ventures
Device line装置序列
SPARC (R=1.85m, B=12.2T) → ARC (≈400 MW net)
Status进展
SPARC under construction in Devens; first plasma targeted 2027, Q>2 burning plasma demo, ARC at Chesterfield VA targeting net-electric early 2030s.SPARC 在 Devens 建造中，2027 年首次等离子；目标 Q>2 燃烧等离子体演示。后继 ARC 选址弗吉尼亚 Chesterfield，目标 2030 年代初实现净电。
Engineering bottleneck工程瓶颈
Tritium fuel cycle at industrial scale; HTS magnet quench protection at 12T+; high-energy neutron tolerance of first wall.工业级氚燃料循环；12T+ HTS 磁体失超保护；第一壁高能中子耐受。
05/SECTION章节

Lawson Simulation Lab劳逊模拟实验室

Drive density, ion temperature, and confinement time. Watch the triple product cross the ignition line. Switch fuel from D-T to D-D to p-B11 and see why the aneutronic dream is 500× harder.拖动密度、离子温度、约束时间，看三乘积越过点火线。从氘氚切到氘氘再到氢硼，亲手验证「无中子化」的梦想为什么难 500 倍。

CONTROL SURFACE控制面
Fuel cycle燃料循环
Ignition target 点火门槛 nTτE≈3.00×1021 keV⋅s/m3nT\tau_E \approx 3.00 × 10^21\ \mathrm{keV \cdot s/m^3}
Ion temperature Tᵢ离子温度 Tᵢ
18.0 keV ≈ 209 M K
Density n密度 n
10^20.40 m⁻³
Energy confinement τE能量约束时间 τE
10^0.60 s
1 ps1 ms1 s10³ s
Real-device presets真实装置预设
PLASMA STATE等离子体状态
● IGNITION点火
nTτE=1.80×1022 keV⋅s/m3nT\tau_E = 1.80 × 10^22\ \mathrm{keV \cdot s / m^3}
vs. ignition target相对点火门槛600.00%
Q (fusion gain)
43.89
Pressure p ≈ nkT
7.23e-1 bar

Reading解读

For D-T fuel, ignition (Q→∞) requires a triple product of about 3×10²¹ keV·s/m³. Tokamaks raise nτE through magnetic confinement at moderate T; ICF crashes to enormous n at vanishingly small τE for a few hundred picoseconds — both legal solutions to Lawson.氘氚燃料的点火（Q→∞）要求三乘积约 3×10²¹ keV·s/m³。 托卡马克通过磁约束在中等温度下抬高 nτE； 惯性约束聚变则在数百皮秒内把 n 推到极高、τE 极小—— 两条路径都满足 Lawson 判据。

06/SECTION章节

AI Plasma Command CenterAI 等离子体指挥中心

A live-mocked autonomous reactor OS — discharge sequence, real-time signals, RL controller status, disruption warnings. The HUD a 2035 plant operator will read.模拟自主反应堆 OS——放电序列、实时信号、强化学习控制器状态、破裂预警。一台 2035 电站操控员日常面对的 HUD。

REACTOR MODE反应堆模式
STARTUP启动
phase阶段 0%
Discharge sequence放电序列
STARTUP启动
RAMP-UP上升
FLAT-TOP平顶
ALPHA-HEATINGα 加热
RAMP-DOWN下降
Sim time模拟时间 0.0 s
Cycle周期 0
RL agent — PPO, plasma-shape policy v3.4强化学习智能体 — PPO，等离子体形状策略 v3.4
PRIMARY SIGNALS主信号
Plasma current Iₚ等离子体电流 Iₚ
0.30 MA
Ion temperature Tᵢ离子温度 Tᵢ
0.40 keV
β (normalized)β（归一化）
0.00
Fusion gain Q聚变增益 Q
0.00
Disruption probability破裂概率
2.0%
AI ADVISORYAI 决策建议
All envelopes nominal.所有包络正常。
DIGITAL TWIN数字孪生
SOLPS-ITER edge model · CORSICA core · Ne ≈ 0.8 nGW · running on hot-spare GPU rack #B4 (latency 22 ms)SOLPS-ITER 边界模型 · CORSICA 核心 · Ne ≈ 0.8 nGW · 热备 GPU 机架 #B4（延迟 22 ms）
07/SECTION章节

Fusion University聚变大学

The eight foundational ideas, each in one card and one equation: Lawson, β, alpha-heating, q, TBR, DPA, RL plasma control, and why HTS magnets changed everything.八个基础概念，每个一卡一方程：劳逊判据、β、α 加热、q 因子、氚增殖比、DPA、强化学习等离子体控制，以及 HTS 磁体为何改变一切。

Plasma等离子体物理#lawson

The Lawson criterion劳逊判据

Lawson asked the simplest question of fusion: when does the reaction sustain itself? The answer balances three numbers — particle density n, temperature T, and the time τE the plasma can hold its energy before it leaks out. Below the line, you spend more energy than you make. Above it, the plasma feeds itself.Lawson 提出了聚变最朴素的问题：反应何时能自持？答案要让三件事达成平衡——粒子密度 n、温度 T、以及等离子体在能量泄漏出去之前能维持的时间 τE。线之下你输的比赚的多；线之上，等离子体自我供养。

nTτE≥3×1021 keV⋅s⋅m−3(D-T ignition)n T \tau_E \geq 3 \times 10^{21}\ \mathrm{keV \cdot s \cdot m^{-3}} \quad (\text{D-T ignition})
Triple product threshold for D-T ignition.氘氚点火的三乘积门槛。
MHD磁流体力学#beta

Plasma β: how much pressure your bottle holds等离子体 β：磁瓶能装多少压强

β measures plasma pressure against the magnetic pressure that holds it. β=1 means the plasma's own pressure equals the field's. Most tokamaks operate at β=2-5%. Push higher and instabilities devour the plasma. Spherical tokamaks reach β≈40% — the prize, and the danger.β 衡量等离子体压强相对于约束它的磁压。β=1 意味二者相等。大多数托卡马克在 β=2-5% 区间运行；再推高，不稳定性会吞噬等离子体。球形托卡马克可达 β≈40%——既是奖赏，也是危险。

β=pplasmapmagnetic=nkBTB2/2μ0\beta = \frac{p_{plasma}}{p_{magnetic}} = \frac{n k_B T}{B^2 / 2\mu_0}
Plasma beta — the efficiency of magnetic confinement.等离子体 β——磁约束效率。
Plasma等离子体物理#alpha

Alpha-particle self-heatingα 粒子自加热

When a D-T reaction fires, the 14.1 MeV neutron escapes — but the 3.5 MeV alpha (a helium-4 nucleus) carries 20% of the energy and is trapped by the magnetic field. Above Q≈5, alphas alone heat the plasma faster than it cools. This is the threshold called 'burning plasma'.氘氚反应触发时，14.1 MeV 中子飞走——但 3.5 MeV α 粒子（氦-4 核）带走 20% 的能量并被磁场俘获。Q≈5 以上时，α 粒子单凭自身加热的速度就超过冷却速度。这就是所谓「燃烧等离子体」的门槛。

Pα=14n2⟨σv⟩Eα,Eα=3.5 MeVP_{\alpha} = \frac{1}{4} n^2 \langle\sigma v\rangle E_{\alpha}, \quad E_{\alpha} = 3.5\ \mathrm{MeV}
Alpha heating power per unit volume.单位体积 α 加热功率。
MHD磁流体力学#safety

The safety factor q安全因子 q

q counts how many toroidal trips a magnetic field line takes per poloidal trip. q<1 in the core invites the sawtooth instability; q at the edge governs ELMs and disruptions. The whole magnetic shaping of a tokamak is a fight to keep q profiles where they don't murder the plasma.q 表示磁力线每绕一圈极向时绕了多少圈环向。芯部 q<1 会引发锯齿不稳定；边界 q 决定边界局域模与破裂。整个托卡马克的磁形塑造，都是在跟 q 廓线博弈，避免它把等离子体杀掉。

q(ψ)=12π∮BϕRBθ dlq(\psi) = \frac{1}{2\pi} \oint \frac{B_\phi}{R B_\theta}\,dl
Safety factor along a flux surface.沿磁面计算的安全因子。
Reactor反应堆工程#tritium

Tritium breeding ratio (TBR)氚增殖比（TBR）

Tritium has a 12.3-year half-life and no large natural source. Every D-T plant must breed its own from lithium in a blanket surrounding the plasma. TBR > 1 is the condition for fuel self-sufficiency. No power plant has yet demonstrated it; the materials, geometry, and neutron multiplication for TBR ≥ 1.05 are the central engineering bet.氚的半衰期 12.3 年，自然界无大规模来源。每座 D-T 电站都必须自行用锂在等离子体周围的包层中增殖氚。TBR>1 是燃料自给的条件。至今无一电站演示过它；要实现 TBR ≥ 1.05 所需的材料、几何与中子倍增，是核心工程赌注。

TBR=NT,bredNT,burnt≥1.05\mathrm{TBR} = \frac{N_{T,bred}}{N_{T,burnt}} \geq 1.05
TBR must exceed unity by ~5% to cover startup, decay, and losses.TBR 须超过 1 约 5%，覆盖启动、衰变与损耗。
Materials材料科学#neutron

Neutron damage in DPA中子损伤（DPA）

First-wall components in a D-T plant absorb 14 MeV neutrons constantly. Damage is measured in displacements per atom (DPA): how many times each atom in the lattice has been knocked off its site. Tungsten survives ~50 DPA. ITER will accumulate ~3 DPA in its lifetime; a power plant first wall must endure 100+ DPA. This drives the entire reduced-activation steel programme.D-T 电站第一壁组件不断吸收 14 MeV 中子。损伤以「每原子位移次数」（DPA）衡量：晶格中每个原子被撞离原位的次数。钨可承受约 50 DPA。ITER 全寿命累积约 3 DPA；电站第一壁须撑过 100+ DPA。整个低活化钢研发都是为此驱动。

DPA=∫0tσdisp(E) Φ(E,t) dE dt\mathrm{DPA} = \int_0^t \sigma_{disp}(E)\,\Phi(E,t)\,dE\,dt
Cumulative displacements per atom over fluence time.随中子注量累积的每原子位移数。
Control控制系统#rl-control

RL plasma shape control强化学习等离子体形状控制

DeepMind & EPFL/SPC showed in 2022 that a single RL policy can control the magnetic actuators of TCV to drive the plasma into shapes — diverted, snowflake, droplet — that classical controllers struggle with. This collapses the controller-tuning problem from human-months per shape to GPU-hours per shape. The next reactor OS layer is being written in PyTorch.2022 年 DeepMind 与 EPFL/SPC 证明：一个强化学习策略可控制 TCV 的磁致动器，把等离子体塑造成传统控制器难以驾驭的偏滤、雪花、液滴等形状。控制器调参问题由「每形状数人月」坍缩为「每形状数 GPU 小时」。下一层反应堆 OS 正在 PyTorch 里写。

π∗(s)=arg⁡max⁡aE[∑tγtr(st,at) ∣ s0=s]\pi^*(s) = \arg\max_a \mathbb{E}\left[\sum_t \gamma^t r(s_t,a_t) \,\big|\, s_0 = s\right]
Optimal policy under discounted return — the RL plasma controller objective.折扣回报下的最优策略——强化学习等离子体控制器目标。
Reactor反应堆工程#hts

Why HTS magnets changed everything为什么 HTS 磁体改变了一切

Tokamak fusion power scales as B⁴ × R³. The cost scales like R³. So smaller, higher-field machines win — IF the magnets exist. REBCO high-temperature superconductor tape can carry 20+ tesla at ~20 K, twice what NbTi-based ITER magnets achieve. CFS's 2021 demo of a 20-T HTS coil is what made compact tokamaks venture-fundable.托卡马克聚变功率正比于 B⁴ × R³，造价约正比于 R³。于是更小、更高场的机器更有竞争力——前提是磁体存在。REBCO 高温超导带可在约 20 K 下承载 20+ 特斯拉，是 ITER NbTi 磁体的两倍。CFS 2021 年 20 T HTS 线圈演示，让紧凑托卡马克具备风投可融资性。

Pfus∝β2B4V⇒Pfus/V∝B4P_{fus} \propto \beta^2 B^4 V \quad \Rightarrow \quad P_{fus}/V \propto B^4
Fusion power density scales with the fourth power of magnetic field.聚变功率密度与磁场强度的四次方成正比。
08/SECTION章节

Civilization Timeline文明时间轴

From Big Bang nucleosynthesis to Kardashev II. 24 inflection points across 13.8 billion years.从大爆炸核合成到卡尔达肖夫二型。138 亿年里的 24 个拐点。

−13.8 Gyr
Cosmic宇宙

Big Bang Nucleosynthesis大爆炸核合成

Within the first 20 minutes after the Big Bang, ⁴He, ²H and ⁷Li form by fusion. Sets the universe's hydrogen reservoir — the substrate of every star and every fusion plasma.大爆炸后头 20 分钟，由聚变生成 ⁴He、²H、⁷Li，奠定宇宙氢储量——所有恒星与聚变等离子体的母质。

−13 Gyr
Cosmic宇宙

First stars ignite p-p chain首代恒星点燃质子—质子链

Gravitational confinement begins natural fusion at 1.5×10⁷ K. The Sun still runs this same reaction at 600 million tons of H per second.引力约束让 1.5×10⁷ K 自然聚变启动。太阳至今仍以每秒 6 亿吨氢的速率运行这一反应。

1929
Theory理论

Atkinson & Houtermans propose stellar fusion阿特金森与胡特曼斯提出恒星聚变

First quantitative model of nuclear fusion as the source of stellar luminosity, predating Bethe's 1939 CNO/p-p paper.首个把核聚变定为恒星光度来源的定量模型，先于 Bethe 1939 年 CNO / 质子—质子链论文。

1952
Engineering工程

Ivy Mike: first thermonuclear detonationIvy Mike：人类首次热核爆炸

Liquid-deuterium device delivers 10.4 Mt yield. Proves D-T/D-D fusion releasable on Earth — but as weapon, not as power.液态氘装置释放 1040 万吨当量。证明氘氚 / 氘氘聚变可在地球释放——但是武器，不是电力。

1956
Theory理论

Lawson criterion published劳逊判据发表

John D. Lawson defines nτE > 10²⁰ s/m³ for ignition. The single inequality that has framed every fusion device for 70 years.Lawson 给出点火条件 nτE > 10²⁰ s/m³。这一不等式定义了 70 年来的每一台聚变装置。

1968
Experiment实验

Soviet T-3 reaches 10 million K苏联 T-3 达到一千万度

Kurchatov Institute's tokamak (Sakharov & Tamm concept, 1951) demonstrates electron temperatures 10× higher than any other device. Triggers global tokamak adoption.库尔恰托夫所的托卡马克（萨哈罗夫与塔姆 1951 概念）电子温度比任何其他装置高 10 倍。触发全球托卡马克转向。

1977
Experiment实验

Shiva laser at LLNLLLNL Shiva 激光器

20-beam, 10.2 kJ Nd:glass laser. First serious laser ICF program. Direct ancestor of NOVA → NIF.20 束、10.2 kJ Nd:玻璃激光器。首个严肃的激光 ICF 项目，NOVA → NIF 的直接祖先。

1994
Experiment实验

TFTR sets D-T fusion power record at 10.7 MWTFTR 创下 10.7 兆瓦氘氚聚变功率纪录

First reactor-relevant deuterium-tritium fusion at Princeton. Q≈0.27 — net loss but closing the gap.普林斯顿首次反应堆级氘氚聚变。Q≈0.27，仍是净损耗，但差距收敛。

2006
Engineering工程

ITER Agreement signed by 7 partiesITER 协议由 7 方签署

EU, US, China, Japan, Korea, Russia, India commit to building the largest tokamak ever in Cadarache, France. Originally costed $5B; current estimate $25B+.欧、美、中、日、韩、俄、印承诺在法国 Cadarache 建造史上最大托卡马克。原预算 50 亿美元，当前估算 250 亿美元以上。

2009
Engineering工程

NIF first laser shotNIF 首发激光

World's largest laser (1.8 MJ across 192 beams) commissioned at LLNL after $3.5 B and 12 years.世界最大激光器（192 束、合 1.8 MJ）在 LLNL 调试完毕，耗时 12 年、35 亿美元。

2018
Industrial产业

Commonwealth Fusion Systems spins out of MIT联邦聚变系统从 MIT 衍生

First serious private bet on REBCO HTS magnets enabling compact tokamaks. Triggers the 'fusion is now venture-fundable' phase change.首次严肃押注 REBCO HTS 磁体让托卡马克紧凑化的私营公司，触发「聚变可由风投资助」的相变。

2021-09-05
Engineering工程

CFS demonstrates 20-T HTS magnetCFS 演示 20 特斯拉 HTS 磁体

World-first 20-tesla, large-bore HTS magnet at MIT/CFS — the engineering precondition that makes SPARC possible.MIT/CFS 全球首台 20 特斯拉、大孔径 HTS 磁体——SPARC 得以成立的工程前提。

2022-12-05
Experiment实验

NIF achieves scientific ignitionNIF 实现科学点火

Q=1.54 — first time a fusion reaction releases more energy than the laser delivered. Replicated 4× since with steadily higher gain.Q=1.54——人类首次让聚变反应释放的能量大于激光输入能。此后已 4 次复现，增益逐次走高。

2023-10
Experiment实验

JET sets 69 MJ D-T energy recordJET 创下 69 MJ 氘氚能量纪录

5.2 s shot at the Joint European Torus on its decommissioning campaign. Validates ITER scaling laws.联合欧洲环退役运行最后一炮，5.2 秒。验证了 ITER 缩放律。

2025-04
Experiment实验

EAST sustains 1066 s of high-confinement plasmaEAST 维持 1066 秒高约束等离子体

Hefei tokamak demonstrates that the steady-state engineering envelope of a fusion power plant is reachable today.合肥托卡马克证明聚变电站的稳态工程包络如今已触手可及。

2025
Industrial产业

Helion Polaris assembled; Microsoft PPA activeHelion Polaris 组装完成，微软购电协议生效

First grid-attached fusion plant timeline begins: 50 MW from 2028 if Polaris demonstrates net electricity.首个并网聚变电站时间表启动：若 Polaris 演示净电，2028 年起向微软交付 50 MW。

2034
Engineering工程

ITER first plasma (revised baseline)ITER 首次等离子体（修订基线）

Following the 2024 baseline reset, first hydrogen plasma slips to 2034 with full D-T burn in 2039.依据 2024 年基线重设，首次氢等离子体推迟至 2034，完整氘氚燃烧 2039。

2030s
Industrial产业

First commercial fusion electrons首批商业聚变电子

Helion (FRC), CFS-ARC (HTS tokamak), Tokamak Energy ST-E1, and Type One INFINITY ONE compete to deliver the first commercially-billed fusion electron.Helion（FRC）、CFS-ARC（HTS 托卡马克）、Tokamak Energy ST-E1、Type One INFINITY ONE 竞速交付首个商业计费的聚变电子。

2050s
Industrial产业

DEMO-class plants on gridDEMO 级电站并网

GW-class fusion plants begin to displace baseload coal/gas. EU-DEMO, K-DEMO, CFETR, BEST. Fusion enters the energy mix proper.GW 级聚变电站开始替代基荷煤气电源。EU-DEMO、K-DEMO、CFETR、BEST。聚变进入能源结构主流。

2040–2070
Civilizational文明

AI-coupled fusion data centersAI 耦合聚变数据中心

Frontier-AI training shifts to behind-the-meter fusion plants. The coupling of AGI compute and fusion energy becomes a national-security category.前沿 AI 训练转向「自有聚变」电站。AGI 算力与聚变能源耦合，升格为国家安全门类。

2080s
Civilizational文明

Fusion-powered megacities, fully electrified shipping & aviation聚变城市群，全电化海空交通

Direct desalination at fusion plants makes coastal megacities of 50M+ feasible. Synthetic e-fuels for aviation are unit-cost-competitive with kerosene.聚变电站直驱海水淡化，5000 万级沿海城市群成为可能。合成 e-燃料航空与煤油已具单位成本竞争力。

2100+
Civilizational文明

Interplanetary fusion drives行星际聚变推进

Direct-fusion drives (Princeton-style) replace nuclear-electric. Earth–Mars in 30 days. Asteroid mining becomes economically self-sustaining.直接聚变推进（普林斯顿式）取代核电推进。地火 30 天可达，小行星采矿实现经济自循环。

2200
Civilizational文明

Kardashev I attained卡尔达肖夫一型文明

Civilization energy use ≈ 10¹⁶ W — full planetary power budget. Fusion + space-solar hybrid grid.文明能耗 ≈ 10¹⁶ 瓦——全行星能源预算。聚变 + 空间太阳能混合电网。

—
Civilizational文明

Toward Kardashev II: stellar engineering通向卡尔达肖夫二型：恒星工程

Beyond planetary fusion: stellified gas giants, Dyson swarms, controlled stellar atmospheres. Fusion physics as civilization toolkit.超越行星级聚变：恒星化巨行星、戴森群、受控恒星大气。聚变物理作为文明工具集。

09/SECTION章节

Civilizational Futures文明推演

Set a doubling time. Set a learning rate. Find the year fusion eats the grid. The model is illustrative — but the shape is what matters.设定翻倍周期与学习率，找出聚变吞下电网的那一年。模型只示意——但形状才是关键。

DEPLOYMENT KNOBS部署旋钮
Year年份
2050
Capacity doubling time装机翻倍周期
3.5yr
Learning rate per doubling每次翻倍成本下降
18%

Solar PV's 50-year learning rate is ~20% per doubling. Wind is ~12%. Nuclear fission, by contrast, has a negative learning rate. Fusion's number is unknown — assume it lies somewhere between solar and the worst nuclear reactor build-out.太阳能光伏 50 年学习率约 20% / 翻倍，风电约 12%， 核裂变反而出现负学习率。聚变学习率仍未知—— 在太阳能与最差的核电建造之间任你押注。

CIVILIZATIONAL READOUT文明读数
2050
phase相位
● Early commercial早期商业期
Installed fusion聚变装机
2.6 GW
Share of global electricity全球电力占比
0.43%
Cost per kW单 kW 造价
$3861
Kardashev Σ卡尔达肖夫 Σ
0.342
Δ from 2030较 2030×5.3e+1

At a 3.5-yr doubling time and 18% learning rate, fusion goes from curiosity to majority-share grid in roughly 60 years from first commercial plant — a faster transition than coal-to-gas, slower than mobile phones. Kardashev I requires civilization-wide 10¹⁶ W, dominating any single energy source — fusion is necessary, not sufficient.若装机翻倍周期 3.5 年、学习率 18%，聚变将在首座商业电站后约 60 年内 从奇观变为电网主力——比煤改气快，比移动电话慢。 卡尔达肖夫一型要求文明级10¹⁶ 瓦，超出任何单一能源—— 聚变是必要条件，不是充分条件。