Goldman Sachs' Insane SpaceX AI Forecast Has One Clear Winner: Micron Technology

The piece argues AI infrastructure could expand far beyond today’s semiconductor bottlenecks, citing a Goldman Sachs scenario of 5,288 SpaceX Starship AI missions by 2031. It estimates each launch could carry 30–50 AI satellites, each with ~1 GB300-equivalent AI rack, implying potentially millions of Nvidia-class accelerators in orbit and demand stretching well beyond the current HBM (and DRAM/NAND) supply constraints. While the forecast is framed as highly execution-dependent (launch reliability, regulator approvals, and economic viability), the article concludes that even a fraction of the scenario would likely keep memory demand elevated for years, supporting a long-term demand tailwind for Micron.

Analysis

The investable takeaway is not the orbital AI concept itself; it is that scarcity migrates from compute to memory and packaging earlier than most models assume. If AI deployment broadens even modestly beyond terrestrial hyperscalers, the marginal dollar of capex flows disproportionately into HBM, advanced packaging, and wafer allocation, which is why MU has more convex upside than NVDA from this narrative. The second-order effect is that every new accelerator generation becomes more memory-intensive, so supply discipline by the HBM trio can sustain pricing power even if GPU unit growth normalizes.

The market risk is that this is a very long-dated optionality story masquerading as a near-term fundamental catalyst. The first real checkpoints are not launch counts but HBM contract pricing, MU’s supply commitments, and whether memory lead times stay extended through the next 2-3 quarters. If capital spending from SK hynix and Samsung catches up faster than expected, the thesis compresses into a normal cyclical memory upturn rather than a structural re-rating.

Contrarianly, the consensus may be overstating who captures the value if these systems ever scale. A future custom-chip architecture would likely reduce the incremental leverage to NVDA while preserving demand for memory and foundry capacity; however, it could also shift more economics into in-house silicon design and away from merchant suppliers. TSM is a beneficiary only if custom AI silicon stays on leading-edge nodes; if the workload migrates to lower-cost or vertically integrated chips, TSM’s share of the upside is smaller than headline demand implies.

Key Decisions for Investors

• Long MU on pullbacks for a 6-12 month horizon; thesis is sustained HBM tightness and memory pricing power. Falsify if MU guides to easing lead times or HBM pricing rolls over in the next two quarters.
• Pair trade: long MU / short NVDA for 3-6 months. This isolates the scarcer input (memory) against the most crowded AI beneficiary; risk is NVDA regains relative strength if it secures more HBM supply and re-accelerates margins.
• Avoid chasing TSM on this headline alone; treat it as a watch item, not a conviction long. Upgrade only if there is evidence of incremental advanced-node allocation tied to AI-specific demand rather than generic capex spillover.
• If looking for upside convexity, use MU call spreads rather than outright stock into the next earnings cycle. Risk/reward is best if the market continues to underprice multi-quarter HBM scarcity, but premium should stay limited given long-dated execution risk.
• Set an alert on MU guide and HBM commentary over the next 1-2 earnings calls; if management signals supply normalization, take profits aggressively because the market will re-rate this from structural scarcity to a cyclical memory trade.