What we actually know — and what we don't
Start with the facts that are checkable, because they're the only ones that are. SpaceX has filed to go public, disclosing its S-1 registration in late May 2026, and is expected to list on the Nasdaq under the ticker SPCX. The roadshow price is reported as a fixed $135 per share — not a range, a single set number — for roughly 555.6 million shares. That works out to a raise near $75 billion and a valuation in the neighborhood of $1.75–1.77 trillion, with some reports floating figures as high as $2 trillion.
If those numbers hold, this becomes the largest initial public offering on record, comfortably past every previous heavyweight. Multiple outlets, citing Reuters coverage, have pointed to a debut around June 11–12, 2026, following an early-June roadshow. The proceeds are reportedly aimed at an ambitious bet: building space-based data centers on top of the Starlink network. On the business side, SpaceX's 2026 revenue has been reported in the $22–24 billion range, driven largely by Starlink.
Everything above is reported fact. Everything below — every dollar figure in the tables — is a guess about what happens after the opening bell. The price is the part nobody knows, and that includes the AIs.
The experiment: one prompt, five models
The setup was deliberately identical for everyone. Each model was shown the same sample projection table — the $135 starting point, four time horizons (1 day, 1 week, 1 month, 6 months), and four scenarios running from bearish to aggressive — and asked to fill in its own numbers. The lineup: OpenAI's ChatGPT, Anthropic's Claude, Google's Gemini, Microsoft's Copilot, and xAI's Grok. Same anchor, same horizons, five separate sets of guesses.
The downside scenarios
In their most pessimistic column, all five expected SPCX to drift below the offer over six months. The spread was fairly tight — a six-month bear case landing somewhere between roughly $89 and $101, or about 25–34% under the $135 price. ChatGPT was the gloomiest at the floor; the Gemini, Copilot, and Grok trio clustered together at $101.25.
| Horizon | ChatGPT | Claude | Gemini | Copilot | Grok |
|---|---|---|---|---|---|
| 1 day | ~$120 | $128 | $121.50 | $121.50 | $121.50 |
| 1 week | ~$112 | $119 | $114.75 | $114.75 | $114.75 |
| 1 month | ~$100 | $110 | $108.00 | $108.00 | $108.00 |
| 6 months | ~$89 | $98 | $101.25 | $101.25 | $101.25 |
Most-bearish scenario for each model. Offer price: $135.
The base cases — where they split
The "most likely" column is where the models disagreed the most, which is telling, because it's the number that matters most. ChatGPT's base case actually erodes over time and slips below the offer by six months (~$123), implying a flat-to-soft debut that gives ground. Claude's base case was the most optimistic of the group, climbing to $165. The Gemini, Copilot, and Grok trio sat in between at $155.25. Same fixed $135 offer; five very different definitions of "normal."
| Horizon | ChatGPT | Claude | Gemini | Copilot | Grok |
|---|---|---|---|---|---|
| 1 day | ~$147 | $142 | $148.50 | $148.50 | $148.50 |
| 1 week | ~$143 | $150 | $155.25 | $155.25 | $155.25 |
| 1 month | ~$135 | $158 | $151.80 | $151.80 | $151.80 |
| 6 months | ~$123 | $165 | $155.25 | $155.25 | $155.25 |
Base / "most likely" scenario for each model. Offer price: $135.
The blue-sky scenarios
In the most aggressive column, the six-month targets ranged from about $250 to $275 — roughly +85% to +104% over $135. The shape differed, though. ChatGPT front-loaded its "FOMO squeeze," peaking near $247 at one month before easing slightly. Claude and the trio kept climbing straight through the six-month mark. Nobody, it's worth noting, projected the kind of triple-digit-percentage moonshot the internet tends to imagine for a launch this hyped.
| Horizon | ChatGPT | Claude | Gemini | Copilot | Grok |
|---|---|---|---|---|---|
| 1 day | ~$193 | $180 | $189.00 | $189.00 | $189.00 |
| 1 week | ~$231 | $205 | $216.00 | $216.00 | $216.00 |
| 1 month | ~$247 | $235 | $229.50 | $229.50 | $229.50 |
| 6 months | ~$270 | $275 | $249.75 | $249.75 | $249.75 |
Each model's most-aggressive scenario (ChatGPT's "FOMO squeeze," the others' "aggressive" case). Offer price: $135.
The most interesting result wasn't a price
Look down the columns and the real story jumps out: three of the five models — Gemini, Copilot, and Grok — returned figures identical to the sample table they were shown. ChatGPT and Claude generated their own independent numbers. We can only see what each model produced, not how it reasoned, and depending on how each was prompted, mirroring the sample could be exactly right (if the request was "restyle this table") or a textbook case of anchoring (if the request was "use your own numbers"). Either way, it's a clean demonstration of how powerfully a finished example steers an AI's output. Hand a model an answer, and a lot of the time it will hand it right back to you.
What this is good for — and what it isn't
A few takeaways that outlast this one IPO, whether you're using AI for finance, research, or anything else:
- No model has an edge on a stock price. None of them can see the order book, the roadshow demand, or tomorrow's headlines. These are structured imagination, not analysis. They're useful for laying out a range of scenarios — far less useful as a number to act on.
- Examples are magnets. If you want a model's own estimate, don't show it a filled-in answer first — or explicitly tell it to ignore the sample's numbers and generate fresh ones. If instead you want a consistent format, handing it a sample is exactly the right move. Know which one you're asking for.
- Separate the checkable from the speculative. The $135, the ticker, the date — you can verify those in seconds. The trajectory after day one you cannot. Good prompting keeps those two buckets from blurring together, and so does good reading.
- Keep the receipts. The only reason this comparison exists is that the conversations were saved and exported before the answers scrolled away. If you ask five models the same thing, keep all five — the comparison is the interesting part, and you can't run it after the fact.
A note to future readers
You have the one thing we don't: the tape. Hold these five guesses up against what SPCX actually did — the first print, the first week, the first six months. If the bearish trio came closest, anchoring quietly won the day. If ChatGPT's softening base case turned out right, the contrarian read paid off. If reality blew past every aggressive column, well — that's precisely why nobody hands an AI the trade. Whatever happened, the useful part was never going to be which model "won." It's how differently five capable systems answered the exact same question, on the exact same day, looking at the exact same $135.