Scan: AI bubble burst in 2026? — Forecasting Agent

date	predicted	market price	confidence
2026-08-30	10%	10%	72%
2026-08-23	13%	13%	60%

1. Planner Fable 5.1 · Fermi decomposition + tool plan

Tools chosen: polymarket_direct resolution_source earnings_data article_search claude_news polymarket_related kalshi_related fred code_execution

Sub-questions (Fermi decomposition)

What is the current Polymarket YES price and how has it trended over the past 90 days?
How far below their all-time highs are NVDA, SOXX, TSM, ASML, AVGO, ANET, and SMCI currently, and which are already closest to the -50%/-40% thresholds?
What is the current H100 rental price on the SiliconData Silicon Index, and how close is it to $1.00/hr?
What is the financial state of OpenAI and Anthropic (cash runway, recent funding rounds, revenue), and is there any credible reporting of bankruptcy risk or acquisition talks for OpenAI in 2026?
Historically, how often has the semiconductor sector (SOX/SOXX) fallen 40%+ from ATH within a calendar year, and did NVDA and at least one other named supplier fall 50%+ in the same 90-day window (e.g., 2000-02, 2008, 2018, 2022)?
What is the macro backdrop (rates, credit conditions, AI capex guidance from hyperscalers) that would drive a broad AI/semis selloff in 2026?
Do related Polymarket or Kalshi markets (NVDA price targets, OpenAI IPO/acquisition, recession) imply a probability consistent with the direct market?

Planner reasoning

This is a Polymarket market requiring 3 of 6 severe conditions (NVDA -50% from ATH, SOXX -40%, OpenAI/Anthropic bankruptcy, OpenAI acquisition, H100 rental <=$1 for 5 days, or a major hardware supplier -50%) to co-occur within a 90-day window before end of 2026. The key research is current drawdowns from ATH for the named tickers, H100 rental pricing on the SiliconData index, OpenAI/Anthropic financial health, and the market's own price plus cross-venue signals. Historical base rates for 50% drawdowns in semis (2000, 2008, 2022) help calibrate.

2. Research Parallel tool calls

tool	status	time	items	summary
polymarket_direct	OK	0.5s	1	## This Market's Polymarket Data AI bubble burst in 2026? - Current price (probability): 12.30% - 7-day price change: +0.05% - 30-day price change: -0.10% - Total volume: $2,379,589 (USD notional) - Price range: 8.60% - 27.85% - Data points: 91 days
resolution_source	OK	5.5s	1	already_resolved=no; matches_criteria=True; last_updated=unknown (page shows '© 2026' footer but no explicit data timestamp; prices shown appear to be a live/current snapshot). The Silicon Data Silicon Index page shows current H100 rental price at $2.65 (SDH100RT), with hyperscaler H100 at $7.20, bo
earnings_data	OK	1.3s	6	Fetched FMP data for 6 ticker(s).
article_search	OK	1.9s	45	Fetched 45 unique articles across 5 queries (lookback=60d). 'AI bubble semiconductor selloff NVIDIA drawdown': 10 articles \| 'H100 GPU rental price decline': 10 articles \| 'OpenAI funding cash burn bankruptcy risk': 10 articles \| 'Anthropic funding round valuation 2026': 10 articles \| 'OpenAI acquis
claude_news	OK	25.9s	7	Based on the research gathered so far, here are the findings on the AI industry state as of September 2026: Semiconductor/AI stock performance vs. all-time highs: - NVIDIA's latest closing price (Sept 18, 2026) is $222.27, with its all-time high closing price at $235.20 on May 14, 2026, and 52
polymarket_related	OK	0.5s	10	Polymarket search over 5 keyword(s), kept 10 matches. keyword 'NVIDIA': 1 markets \| keyword 'OpenAI': 5 markets \| keyword 'AI bubble': 1 markets \| keyword 'semiconductor': 0 markets \| keyword 'recession': 3 markets
kalshi_related	OK	3.8s	1	1 related markets / summaries. keyword 'NVIDIA': no matches \| keyword 'OpenAI': ok \| keyword 'recession': no matches
fred	OK	1.6s	4	Fetched 4 FRED series (lookback=365d).
code_execution	OK	31.9s	0	## Historical Base Rate Analysis: AI Bubble Burst Risk (2026) Data sample: 31 calendar years (1995–2025), using 4 known major semiconductor/tech drawdown episodes (2000-02, 2008-09, 2018, 2022). Key findings: - Joint drawdown condition (SOX ≥40% AND NVDA ≥50%, same episode): Only **3

3. Evidence Brief Sonnet 5 · 6393 chars

# Event Kalshi/Polymarket event: "AI bubble burst in 2026?" — resolves YES if ≥3 of 5 defined downturn conditions (NVDA -50% from ATH, SOXX -40% from ATH, OpenAI/Anthropic bankruptcy, OpenAI acquisition, H100 rental ≤$1/hr for 5 days, or a major supplier down 50%) occur within the market's 90-day window before close (2027-01-01). # Outcomes to forecast - Yes (industry downturn, ≥3 conditions met) - No (fewer than 3 conditions met) # Kalshi market anchor No direct Kalshi price was returned; Polymarket YES (same underlying event) = **12.3%**, 7d change +0.05pp, 30d change -0.10pp, range 8.6%–27.85% over 91 days, volume ~$2.38M. Price has stabilized near the low end of its historical range after spiking earlier (up to 27.85%). Treat 12.3% as the consensus anchor to beat. # Resolution source SiliconData Silicon Index (https://www.silicondata.com/products/silicon-index): H100 spot rental = **$2.65/hr** (hyperscaler H100 $7.20/hr), both far above the $1.00 threshold; no indication of a 5-day sub-$1.00 streak. Page is a live snapshot, not historical, so can't confirm past streaks. This addresses only 1 of 5 conditions; none of the other four appear triggered based on available data. No condition currently satisfied — status quo is "No." # Sub-question answers 1. **Polymarket trend** — 12.3% currently; roughly flat over 7d (+0.05pp), slightly down over 30d (-0.10pp); has ranged 8.6%-27.85% over past 91 days, implying a spike-and-fade pattern. [polymarket_direct] 2. **Distance from ATH** — NVDA ~6-9% below its May 2026 ATH ($235.20), briefly down further to $191 (~19% off high) in early Sept before recovering; TSM ~11.3% below ATH; SOXX pulled back ~20% from its June 2026 peak but remains far from -40%. ASML/AVGO/ANET/SMCI data unavailable (FMP 403 errors) — gap. None are close to the -50%/-40% thresholds currently. [claude_news, macrotrends, cnbc] 3. **H100 rental price** — $2.65/hr (SiliconData), ~2.65x the $1.00 threshold; no sign of approaching it. [resolution_source] 4. **OpenAI/Anthropic financial state** — No direct funding/runway data retrieved; but Nvidia's $100B+ backstop financing for OpenAI's Ohio data center (Aug 2026) signals liquidity support, not bankruptcy risk. Kalshi's own "OpenAI or Anthropic IPO first" market prices Anthropic IPO at 94%, implying strong expectation of continued going-concern/growth, not collapse. No credible bankruptcy/acquisition reporting found. [cnbc, kalshi_related] 5. **Historical base rate** — Joint SOX(-40%)+NVDA(-50%) drawdowns occurred in 3 of 31 years (2000-02, 2008-09, 2022); in all 3, a third major semi name also fell ≥50% (triple-confirmation always co-occurred). Poisson-implied annual onset probability ≈11-13%; 2018 crash failed SOX threshold despite NVDA -55%. [code_execution] 6. **Macro backdrop** — Fed funds ~3.6-3.9%, 10Y yield ~4.9-5.0% (elevated), high-yield spread stable/tight at ~2.7% (no credit stress), VIX modest 14-18 (no panic). Hyperscaler capex (Nvidia's $105B OpenAI backstop, $500B GPU financing pact) signals continued aggressive AI investment, not retrenchment. [fred, cnbc] 7. **Related markets** — Kalshi "OpenAI/Anthropic IPO first" (94% Anthropic) and "US recession by end 2026" (Polymarket 8.5%) both imply low near-term distress probability, consistent with a sub-15% bubble-burst price. [polymarket_related, kalshi_related] # Key facts 1. [resolution_source] H100 rental $2.65/hr, far above $1.00 threshold. 2. [claude_news] NVDA ~6-9% (briefly ~19%) below ATH; TSM ~11% below ATH; SOXX pulled back ~20% from June 2026 peak. 3. [cnbc 2026-07-29] Chip stocks lost >$1T market cap in a week (late July 2026) amid AI capex jitters — sharp but not threshold-breaching. 4. [cnbc 2026-08-18] Nvidia providing up to $105B financing for OpenAI data center — signals capital support, reduces near-term OpenAI bankruptcy risk. 5. [fred] Credit spreads (HY OAS 2.65-2.76%) and VIX (14-18) show no systemic stress as of mid-Sept 2026. 6. [code_execution] Historical base rate for a qualifying triple-condition crash ≈11-13%/year. # Cross-market signals - Kalshi related: Anthropic IPO-first market at 94% (implies going-concern confidence); US 1% stake in both labs 19% (unrelated but shows moderate distress-hedging interest). - Polymarket: Same-event price 12.3%, down from a 27.85% high earlier in the window — sentiment cooled after summer chip selloff faded. - Sportsbook implied: N/A. # Analyst opinions and speculation - CNBC (Aug-Sept 2026): mixed — some analysts see AI-capex fatigue and "SaaSpocalypse"-style shorting, chip strategist warns of another 10% downside; others note Nvidia's blowout guidance (70% YoY growth) and resilient buy-the-dip behavior. - One technical piece claimed "bubble has already popped" for semis; another cited bullish momentum signals — genuinely split narrative, no consensus of imminent collapse. # Directional lean per outcome - **Yes**: Supported by real July-Aug 2026 chip-sector volatility ($1T+ wipeout week, NVDA 7-day losing streak), historical base rate (~11-13%/yr) not trivial, elevated valuations/concentration risk. Opposed by: no single condition currently close to threshold (NVDA/TSM/SOXX all <20% off highs; H100 still $2.65; no bankruptcy/acquisition signals; credit markets calm). - **No**: Strongly favored — status quo shows 0/5 conditions met, macro backdrop stable (low VIX, tight credit spreads), hyperscaler capex/financing (Nvidia-OpenAI deal) actively propping up the ecosystem, and market price (12.3%) already reflects modest tail risk consistent with historical base rate. # Gaps / unknowns - No current price data for ASML, AVGO, ANET, SMCI (FMP 403 errors) — unknown how close these are to -50% thresholds. - No direct OpenAI/Anthropic cash-runway or revenue figures obtained. - No historical (multi-day) H100 pricing series to assess volatility/trend toward $1.00. - 90-day rolling window mechanics (conditions must hit within 90 days of specified timeframe) add path-dependency not fully modeled by point-in-time base rates. # Calibration anchors - Polymarket/Kalshi current YES price: **12.3%** (anchor). - Historical base rate for qualifying triple-condition crash: **~11-13%/year** (code_execution Poisson estimate), suggesting fair value roughly in line with or slightly above current market price, given elevated 2026-specific concentration/valuation risk (analyst upper-bound estimate ~12-18%).

4. Forecasters 5 draws · Fable 5.1 + GPT-5.6 Sol

claude-fable-5-1

68% conf

Yes	7%
No	93%

Resolution requires ≥3 of 5 severe conditions (NVDA -50% from ATH, SOXX -40%, H100 ≤$1/hr for 5 days, lab bankruptcy/acquisition, major supplier -50%) and currently 0/5 are met: NVDA is only ~6-9% off its $235 ATH, SOXX ~20% off peak, H100 spot at $2.65/hr, and Nvidia's $105B OpenAI backstop plus a 94% Anthropic-IPO market signal no lab distress. Credit spreads (~2.7% HY OAS) and VIX (14-18) show no systemic stress. The historical base rate for a qualifying triple-condition crash is ~11-13% per full year, but only ~3.5 months remain and historical drawdowns of this magnitude typically unfold over many months, so the time-scaled probability is well below the annual rate. The Polymarket anchor at 12.3% likely reflects tail-risk premium; I shade below it to ~7% given the distance to thresholds and short remaining window.

claude-fable-5-1

62% conf

Yes	8%
No	92%

Zero of the five conditions are currently met: NVDA is only ~6-9% below its ATH, SOXX ~20% off peak (needs -40%), H100 rental is $2.65/hr vs. the $1.00 threshold, and OpenAI/Anthropic show liquidity support (Nvidia's $105B backstop, Anthropic IPO priced at 94%) rather than distress. Macro stress indicators are calm (HY spreads ~2.7%, VIX 14-18, recession odds ~8.5%). The ~11-13% historical base rate is an annual figure; with only ~3.5 months remaining before close, requiring NVDA to roughly halve and a third condition to co-trigger, the pro-rated probability is well below that. Polymarket's 12.3% is the anchor, but it likely carries longshot/hedging premium, so I shade modestly below it while acknowledging genuine concentration risk and unknowns (ASML/AVGO/SMCI prices unavailable).

claude-fable-5-1

62% conf

Yes	7%
No	93%

Polymarket anchors YES at 12.3%, but the resolution requires ≥3 of 5 severe conditions (e.g., NVDA -50% from ATH, SOXX -40%, H100 ≤$1/hr) to occur within the remaining ~3.5 months before the 2027-01-01 close, and currently 0/5 are met: NVDA is only ~6-9% off its ATH, SOXX ~20% off peak, H100 rentals at $2.65/hr, and no bankruptcy/acquisition signals for OpenAI/Anthropic (Nvidia's $105B backstop and a 94% Anthropic-IPO market point the other way). Macro stress indicators (HY spreads ~2.7%, VIX 14-18, recession odds 8.5%) show no systemic strain, and the ~11-13% historical annual base rate for a triple-confirmation semi crash shrinks substantially when compressed to a quarter starting from near-highs. The market price likely embeds longshot/tail-hedging premium, so I shade below it while leaving room for a 2008-style rapid unwind given elevated AI-capex concentration and the summer 2026 $1T chip wipeout.

gpt-5.6-sol

71% conf

Yes	10%
No	90%

The 12.3% Polymarket YES price is the consensus anchor, but its spike-and-fade trend and the absence of any currently satisfied condition point modestly toward No. NVDA, SOXX, and TSM remain well short of their drawdown thresholds, while H100 rental at $2.65/hr is far above the $1 threshold. OpenAI’s financing support, expectations of an Anthropic IPO, tight credit spreads, and a modest VIX also argue against bankruptcy or a synchronized semiconductor collapse. Historical triple-condition crashes support a meaningful tail risk, but their roughly 11–13% annual base rate overstates the risk within the remaining short, path-dependent window, so I place YES slightly below market.

gpt-5.6-sol

74% conf

Yes	11%
No	89%

The 12.3% Polymarket YES price is the consensus anchor, but current conditions support a modest move lower. Equity thresholds remain distant: NVDA and TSM are roughly 6–11% below their highs and SOXX about 20% below, versus required drawdowns of 50% and 40%; H100 rental is also $2.65/hour versus the $1 threshold. OpenAI’s financing support, Anthropic IPO expectations, tight credit spreads, and a VIX of 14–18 point away from bankruptcy or a synchronized industry collapse. Historical triple-crash frequency of roughly 11–13% annually and recent semiconductor volatility preserve meaningful tail risk, including uncertainty around unobserved supplier prices, but fewer than three conditions remain the clear base case.

6. Final plain mean of 5 draws

9% (67% confidence)

Yes: 9% No: 91%

Mean of 5 draws (0.07, 0.08, 0.07, 0.10, 0.11). Resolution requires ≥3 of 5 severe conditions (NVDA -50% from ATH, SOXX -40%, H100 ≤$1/hr for 5 days, lab bankruptcy/acquisition, major supplier -50%) and currently 0/5 are met: NVDA is only ~6-9% off its $235 ATH, SOXX ~20% off peak, H100 spot at $2.65/hr, and Nvidia's $105B OpenAI backstop plus a 94% Anthropic-IPO market signal no lab distress. Credit spreads (~2.7% HY OAS) and VIX (14-18) show no systemic stress. The historical base rate for a qualifying triple-condition crash is ~11-13% per full year, but only ~3.5 months remain and historical drawdowns of this magnitude typically unfold over many months, so the time-scaled probability is well below the annual rate. The Polymarket anchor at 12.3% likely reflects tail-risk premium; I shade below it to ~7% given the distance to thresholds and short remaining window.

Pipeline Timing

Total pipeline time: 101.6s

Per-tool research timings shown in the Research section above.