Nuclear Energy Gets a Big Government Boost (Again)

Qualcomm’s Investor Day highlighted a data center push, targeting ~$15B of revenue by 2029 with its Dragonfly C1000 planned for production in 2028, signaling continued AI infrastructure demand despite existing supply bottlenecks. IBM also announced sub-1nm (0.7nm) NanoStack research aiming for ~50% higher performance and ~70% greater efficiency, potentially addressing data center power constraints. Separately, the Trump administration’s nuclear loan program proposes low-cost capital to accelerate large-reactor development, aiming for 10 reactors by 2030—supportive for nuclear-linked power infrastructure beneficiaries.

Analysis

The main market implication is not “AI is over” or “nuclear is back,” but that the bottleneck is shifting from raw compute to power-efficient compute and then to grid-qualified deployment. That favors businesses that can monetize the constraint immediately: transmission/EPC names like PWR, incumbent operators with balance-sheet access like CEG/CCJ, and efficiency enablers like IBM, rather than pre-revenue reactor concepts whose cash flows sit well beyond the current policy window. The public-market tendency will be to overcapitalize the most narrative-heavy nuclear names while underpricing the boring contractors that actually turn federal credit support into backlog.

QCOM’s move into data-center CPUs reads more like an option on an inference-cycle expansion than a near-term EPS driver; the important tell is the customer signal, not the 2029 revenue target. If agentic workloads really multiply CPU demand, QCOM can win share in a part of the stack where performance-per-watt matters more than absolute GPU dominance. But the stock should be treated as a long-duration execution story: if design wins do not convert into tape-out milestones and production visibility by 2027, the market will likely re-rate this as strategic theater.

Contrarian view: the consensus is underestimating how much of the nuclear upside gets captured by financing and installation, not by the reactor IP itself. Government support lowers project IRR hurdle rates, but it does not remove permitting, interconnect, or EPC risk, so the winners are likely to be the ones with existing assets, regulated cash flows, or installed execution capability. The stock-bubble risk sits in OKLO/SMR; the policy-backed, lower-volatility exposure is farther up the chain in CCJ, CEG, PWR, and, on the demand-side bridge, FSLR and NEE.

Key Decisions for Investors

• Go long QCOM on pullbacks over the next 1-3 months as a 12-18 month call on inference-driven CPU share gains; falsify if no meaningful design-win cadence emerges by 2027 or if data-center commentary gets pushed out again.
• Enter a small pair trade: long PWR / short OKLO or SMR for 3-6 months. Thesis: federal support and AI power demand monetize through contractors and transmission first, while SMR names remain funding/permitting-risk heavy. Cover the short if DOE-backed project awards start arriving faster than expected.
• Accumulate CCJ and/or CEG on weakness rather than chase the first headline spike; 6-12 month setup favors incumbents with real assets and financing access. Theses breaks if project timelines slip beyond the 2030 window or if policy support is diluted in budget negotiations.
• Treat IBM as a strategic hold, not a trade-for-today. Best entry is on any post-news fade; it is a 1-3 year efficiency option on power-constrained AI rather than a next-quarter revenue story.