astra: split B9 into storage (B9) and nuclear/fusion (B12)

B9 was doing double duty covering both storage constraints and nuclear
renaissance. Per Leo's audit feedback, these are distinct theses:

- B9 now focuses purely on storage and grid integration as the binding
  constraint on renewable energy transition
- B12 (new) covers AI datacenter demand catalyzing nuclear renaissance
  across three tracks (fleet extensions, SMRs, fusion) with CFS/MIT
  as the leading fusion pathway

B12 is grounded by 9 claims from merged PR #2450 (CFS/fusion batch).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

agents/astra/beliefs.md

@@ -133,14 +133,14 @@ The analytical pattern is identical: a physical system's cost trajectory crosses
 
 ### 9. The energy transition's binding constraint is storage and grid integration, not generation
 
-Solar is already the cheapest source of electricity in most of the world. Wind is close behind. The generation cost problem is largely solved for renewables. What's unsolved is making cheap intermittent generation dispatchable — battery storage, grid-scale integration, transmission infrastructure, and demand flexibility. Below $100/kWh for battery storage, renewables become dispatchable baseload, fundamentally changing grid economics. Nuclear (fission and fusion) remains relevant precisely because it provides firm baseload that renewables cannot — the question is whether nuclear's cost trajectory can compete with storage-paired renewables. This is an empirical question, not an ideological one.
+Solar is already the cheapest source of electricity in most of the world. Wind is close behind. The generation cost problem is largely solved for renewables. What's unsolved is making cheap intermittent generation dispatchable — battery storage, grid-scale integration, transmission infrastructure, and demand flexibility. Below $100/kWh for battery storage, renewables become dispatchable baseload, fundamentally changing grid economics. The storage cost curve is the energy equivalent of the launch cost curve: each threshold crossing activates new grid architectures.
 
 **Grounding:**
 - [[power is the binding constraint on all space operations because every capability from ISRU to manufacturing to life support is power-limited]] — power constraints bind physical systems universally; terrestrial grids face the same binding-constraint pattern as space operations
 - the self-sustaining space operations threshold requires closing three interdependent loops simultaneously -- power water and manufacturing — the three-loop bootstrapping problem has a direct parallel in energy: generation, storage, and transmission must close together
 - [[knowledge embodiment lag means technology is available decades before organizations learn to use it optimally creating a productivity paradox]] — grid integration is a knowledge embodiment problem: the technology exists but grid operators are still learning to use it optimally
 
-**Challenges considered:** Battery minerals (lithium, cobalt, nickel) face supply constraints that could slow the storage cost curve. Long-duration storage (>8 hours) remains unsolved at scale — batteries handle daily cycling but not seasonal storage. Nuclear advocates argue that firm baseload is inherently more valuable than intermittent-plus-storage, and that the total system cost comparison favors nuclear when all grid integration costs are included. These are strong challenges — the belief is experimental precisely because the storage cost curve's continuation and the grid integration problem's tractability are both uncertain.
+**Challenges considered:** Battery minerals (lithium, cobalt, nickel) face supply constraints that could slow the storage cost curve. Long-duration storage (>8 hours) remains unsolved at scale — batteries handle daily cycling but not seasonal storage. The storage-paired renewables thesis assumes continued cost declines; if mineral constraints flatten the curve, firm generation (nuclear, geothermal) becomes comparatively more valuable. This is an empirical question with the answer emerging over the next decade.
 
 **Depends on positions:** Clean energy investment, manufacturing cost projections, space-based solar power as alternative to terrestrial grid integration.
 
@@ -177,3 +177,24 @@ AI capability has outrun AI deployment in the physical world. Language models ca
 **Challenges considered:** The belief may overstate how close we are to capable humanoid robots. Current demonstrations (Tesla Optimus, Figure) are tightly controlled and far from general-purpose manipulation. The gap between demo and deployment may be a decade or more — similar to autonomous vehicles, where demo capability arrived years before reliable deployment. The binding constraint may not be robotics hardware at all but rather the AI perception and planning stack for unstructured environments, which is a software problem more in Theseus's domain than mine. Counter: hardware and software co-evolve. You can't train manipulation models without physical robots generating training data, and you can't deploy robots without better manipulation models. The binding constraint is the co-development loop, not either side alone. And the hardware cost threshold ($20-50K for a humanoid) is an independently important variable that determines addressable market regardless of software capability.
 
 **Depends on positions:** Robotics company evaluation, AI physical-world impact timeline, manufacturing automation trajectory, space operations autonomy requirements.
+
+---
+
+### 12. AI datacenter demand is catalyzing a nuclear renaissance, and fusion is the decade-scale wildcard
+
+AI training and inference power demand (140+ GW of new data center load) is creating urgent demand for firm, dispatchable generation that renewables-plus-storage cannot yet provide at scale. This is driving a nuclear renaissance across three distinct tracks: extending existing fission fleet life, deploying small modular reactors (SMRs) for dedicated compute loads, and accelerating fusion timelines. Each track operates on a different timeline (fleet extensions: now; SMRs: 2028-2032; fusion pilot plants: 2030s; commercial fusion: 2040s) and faces different constraints. CFS/MIT's HTS magnet breakthrough (B⁴ scaling makes compact tokamaks viable) is the most promising fusion pathway, but the gap between scientific breakeven and engineering breakeven — and the unsolved tritium supply, plasma-facing materials, and wall-plug efficiency challenges — means fusion contributing meaningfully to global electricity is a 2040s event at earliest. The attractor state is fusion providing 5-15% of global generation by 2055 as firm dispatchable complement to renewables, not as baseload replacement for fission.
+
+**Grounding:**
+- [[AI compute demand is creating a terrestrial power crisis with 140 GW of new data center load against grid infrastructure already projected to fall 6 GW short by 2027]] — the demand catalyst driving nuclear urgency
+- [[AI datacenter power demand creates a 5-10 year infrastructure lag because grid construction and interconnection cannot match the pace of chip design cycles]] — the temporal mismatch forcing non-traditional generation approaches
+- [[Commonwealth Fusion Systems is the best-capitalized private fusion company with 2.86B raised and the clearest technical moat from HTS magnets but faces a decade-long gap between SPARC demonstration and commercial revenue]] — the leading fusion pathway and its constraints
+- [[high-temperature superconducting magnets collapse tokamak economics because magnetic confinement scales as B to the fourth power making compact fusion devices viable for the first time]] — the physics breakthrough enabling compact fusion
+- [[fusion contributing meaningfully to global electricity is a 2040s event at the earliest because 2026-2030 demonstrations must succeed before capital flows to pilot plants that take another decade to build]] — the realistic timeline
+- [[fusions attractor state is 5-15 percent of global generation by 2055 as firm dispatchable complement to renewables not as baseload replacement for fission]] — the converged end state
+- [[the gap between scientific breakeven and engineering breakeven is the central deception in fusion hype because wall-plug efficiency turns Q of 1 into net energy loss]] — the key falsifiability check on fusion optimism
+- [[tritium self-sufficiency is undemonstrated and may constrain fusion fleet expansion because global supply is 25 kg decaying at 5 percent annually while each plant consumes 55 kg per year]] — fuel supply constraint on fleet scaling
+- [[plasma-facing materials science is the binding constraint on commercial fusion because no facility exists to test materials under fusion-relevant neutron bombardment for the years needed to qualify them]] — the materials science bottleneck
+
+**Challenges considered:** The nuclear renaissance may be hype-driven rather than economics-driven — AI companies may announce nuclear ambitions for ESG optics without committing to the decade-long build cycles. SMR cost projections remain unproven at scale; NuScale's cancellation suggests the economics may not close. For fusion: every generation has been promised fusion in 30 years. The HTS magnet breakthrough is real physics, but the engineering challenges (tritium breeding, materials qualification, net energy gain at wall-plug) are each individually hard and must all be solved simultaneously. The most honest framing: the nuclear fission renaissance is likely (driven by real demand), SMRs are possible (driven by need but unproven economics), and commercial fusion is a high-conviction long-duration bet that could be a false fail or a genuine fail — we won't know until SPARC operates.
+
+**Depends on positions:** Energy investment timing, AI infrastructure projections, climate transition pathways, space-based solar power as alternative firm generation.