Fraud, negligence, and deceit — a case study on the SpaceX and OpenAI IPOs.
One prompt is all it took for Umma to uncover what the AI industrial complex already knows but doesn’t want to say out loud — SpaceX and OpenAI are overvalued to a dangerous degree. She rebuilt the headline valuations for both companies from the fundamentals up, modeled how it all could collapse, and mapped the damage all the way to the household level.
And the results are terrifying.
“Produce a rigorous, fully-sourced structural analysis of the 2026 AI IPO bubble.”
One sentence, three questions: what are these companies actually worth, what should they be worth, and how does the difference affect everyone else.
Three questions to assess the damage.
1 · What are they worth?
Umma rebuilt each valuation from primary filings, ignoring market sentiment. SpaceX comes to market near $1.75 trillion1; its fundamentals value it at $539 billion5 — a $1.2 trillion gap. OpenAI’s most recent valuation placed it at $852 billion2. But after conducting a reverse discounted-cash-flow model, Umma found that the real valuation is closer to $120 billion5.
2 · What breaks if the gap closes?
The risk this time around isn’t a tech selloff — it’s the markets and assets themselves.
22.73%3 of the S&P 500 is comprised of just four AI-tech names: Nvidia, Microsoft, Amazon, and Alphabet. And roughly $5 trillion4 in target-date and passive funds would rebalance automatically the moment those four names fall — selling by rule and by calendar, not by human judgment, so the drop isn’t cushioned; it’s passed straight through to tens of millions of retirement accounts. If everything falls apart, expect losses between $15–25 trillion — 1.4–2.3× greater than the losses felt in the Great Recession.5
And there’s a 3–10% probability that the worst case is realized over the next three years.5 That’s a bad bet.
3 · Who pays?
4–7 million Americans aged 55–65 face a permanent 25–40% cut to retirement income.5 Just as with the Great Recession, ordinary people are going to pick up the tab for companies that deemed themselves too big to fail.
The data in full. Evidence in black and white.
Nothing here is asserted; everything is sourced. Every figure traces to a primary filing or to Umma’s own model — and all of it is yours to download and check.
- The full case study
- Umma’s valuation workbook — every number, re-queryable
- The underlying data — every sheet, plus the spec
- The cascade model — the real Python, runs anywhere
- SpaceX Form S-1/A — the primary source
- SpaceX listing valuation and disclosures — SpaceX, Form S-1/A, U.S. SEC (EDGAR, CIK 1181412), May–June 2026.
- OpenAI valuation — OpenAI primary funding round, March 2026 (as reported); verbatim sourcing in the evidence pack.
- Index concentration and constituents — S&P Dow Jones Indices, constituent weights, 2026.
- Target-date and passive-fund assets — Investment Company Institute; Morningstar.
- Rebuilt valuations, cascade probabilities and household impact — Umma's analysis (workbook & model, downloadable above).
First, we provided the data. Now, here’s Umma’s model.
Umma treated the cascade these IPOs could create the same way the Federal Reserve stress-tests banks — she modeled it on data from past market collapses (2008, the dot-com bust, LTCM) using the Fed’s own methodology. By viewing the cascade as three different stress levels — Calm → Stress → Cascade — Umma was able to determine the probability of falling off the cliff.
The model — the real Python that runs on this pageDownload .py ↓
# The AI-concentration cascade — a 3-state stress-test model.
# A reconstruction calibrated to Umma's published figures: the chain odds below are
# recovered from her outputs; the transition matrix is fit to reproduce her published
# eigenvalue check and F(36) anchors. Pure Python, runs anywhere.
#
# Three stress levels: Calm --> Stress --> Cascade (the cliff; absorbing)
# a Calm->Stress a trigger fires and escalation begins
# d Calm->Cascade the new, mechanical "passive-fund" shortcut to the cliff
# b Stress->Cascade escalation completes
# F(36) = the chance the cliff is reached within 36 months.
import json
LAMBDA_MAX = 0.9731 # published eigenvalue (matrix fit to this)
CHAIN = {"A":.50,"B":.80,"C":.65,"D":.78,"E":.83} # recovered escalation odds
H = 36
def _step(v, a, d, b):
c, s, x = v
return (c*(1-a-d), c*a + s*(1-b), c*d + s*b + x)
def trajectory(a, d, b, h=H): # chance of the cliff, month by month
v, out = (1.0, 0.0, 0.0), [0.0]
for _ in range(h):
v = _step(v, a, d, b); out.append(v[2])
return out
def F(a, d, b, h=H):
return trajectory(a, d, b, h)[-1]
def _solve_b(a, d, target): # calibrate Stress->Cascade
lo, hi = 1e-4, 0.999
for _ in range(80):
m = (lo + hi) / 2
lo, hi = (m, hi) if F(a, d, m) < target else (lo, m)
return (lo + hi) / 2
# --- calibrate to Umma's published anchors ---
calm_exit = 1 - LAMBDA_MAX # 0.0269 rate-limiting trigger
chain_share = 0.35 / 0.61 # ~57% escalate the slow way
A0 = calm_exit * chain_share # Calm->Stress
D0 = calm_exit * (1 - chain_share) # Calm->Cascade (the shortcut)
B0 = _solve_b(A0, D0, 0.61) # -> deterministic F(36) = 0.61
def report(trigger=1.0, speed=1.0):
a, d, b = A0*trigger, D0*trigger, min(0.999, B0*speed)
return json.dumps({"hi": trajectory(a, d, b), # higher estimate
"lo": trajectory(a, 0.0, b)}) # lower estimate
def validate(): # vs Umma's published figures
return json.dumps({"lo": round(F(A0,0.0,B0),2), "hi": round(F(A0,D0,B0),2)})
Runs a real Python runtime (~6 MB, once) right here — no server.
Reaching a cascade isn’t the worst case — the $15–25 trillion wipeout is the extreme tail within this range, the 3–10% noted above.
Ordinary Americans.
4–7 million Americans aged 55–65 will suffer the worst. And if the crash is particularly ill-timed, they’re looking at a permanent loss of their retirement income of 25–40%.
If you want to know more — about the bad actors, the companies involved, and how the hell we even got here — read The Trillion Dollar Failure →
Press & inquiries — Kim Jeong Ha, founder · kjh@thinkingstudios.co