Skip to main content
Everyone has an opinion about the AI boom. Prediction markets have prices. While analysts argue whether 2026’s AI trade is a supercycle or a bubble, real-money markets are quietly pricing the specific questions underneath the debate — will OpenAI IPO, will anyone declare AGI, which lab has the best model. This piece rounds up what those markets actually say, including a 59-point disagreement between the two biggest platforms that’s more instructive than any single number.

​
Quick Answer

As of late July 2026, here’s what real-money prediction markets price on the AI boom:That first pair is the headline: Kalshi says 88%, Polymarket says 19% — on what sounds like the same question. It isn’t the same question. One market resolves on an announcement, the other on a completed listing — and the 59-point gap is a masterclass in why resolution criteria matter more than vibes.Meanwhile the “AI bubble” itself isn’t a single market — but its ingredients are priced daily: the memory-chip supercycle (HBM demand projected from ~$35B to ~$100B by 2028), model-release timelines (GPT-5, Grok 4, DeepSeek V4 all have live markets), and lab-vs-lab competition. Educational only — not investment advice. See Sources & References.

​
The Debate: Supercycle or Bubble?

The 2026 AI trade splits smart people into two camps. The supercycle case runs through hardware that is physically sold out. Per Tapbit’s HBM outlook and IG’s supercycle explainer: high-bandwidth memory demand is forecast to grow from roughly $35B in 2025 to $100B by 2028; SK Hynix reportedly won ~70% of next-gen HBM4 orders; Micron’s HBM output for the year is sold out under multi-year contracts; and AI data centers are on track to consume ~70% of all memory chips in 2026. Prices tell the same story — HBM4 stacks reportedly run ~$500, versus ~$200 for HBM3. The bubble case says all of that is already in the price. Mid-July brought a sharp AI-chip selloff as investors asked how much growth was priced in, and Harvard economist Kenneth Rogoff’s warning made the rounds: “This too will pass”. Both camps publish essays. Neither publishes a probability. That’s the gap prediction markets fill.

​
What the Markets Actually Price

You can’t yet trade “is AI a bubble?” as a single contract — it’s not resolvable. But its components are live markets:

​
1. The OpenAI IPO — and the 59-point lesson

The single most instructive AI market of 2026. Per Prediction News:
  • Kalshi: ~88% that OpenAI announces an IPO in 2026 (and ~85% it goes public before Anthropic)
  • Polymarket: ~19% that an OpenAI IPO is completed by December 31, 2026
Same company, same year, 59 points apart — because the contracts resolve on different events. An announcement can happen in a week; a completed listing takes quarters of SEC process. Traders who skimmed the headlines and assumed “same question” were mispricing one side or the other; at one point Polymarket’s contract traded as high as 55% before falling to ~19–29% as the “completed vs announced” distinction sank in. The takeaway travels everywhere in prediction markets: read the resolution criteria before you read the price. (It’s the same discipline our beginner guide drills for sports markets.)

​
2. AGI timelines — the market is calmer than the discourse

For all the AGI noise on social media, Polymarket’s “OpenAI announces AGI before 2027” trades at ~8%. Whatever your view, that’s a useful anchor: people risking money think an AGI declaration in the next ~17 months is unlikely — far below what a casual read of AI Twitter would suggest. Kalshi runs a parallel AGI market with its own resolution rules (see lesson #1).

​
3. Model-by-model horse races

The most active AI markets are the near-term, cleanly resolvable ones: which lab has the best model at month-end (Anthropic priced ~98% for end of July), and release-date markets for GPT-5, Grok 4, and DeepSeek V4. These resolve on benchmarks and shipping dates — crisp questions, deep liquidity, fast price discovery. It’s the AI equivalent of per-match sports markets: short cycles, continuous news, tradeable swings.

​
4. The capital-markets barometer

Polymarket’s “largest IPO of 2026” market has SpaceX at ~87% — a reminder that even in an AI-obsessed year, the market doesn’t assume an AI company tops the IPO league table.

​
Why Prices Beat Punditry (and Where They Don’t)

Why they’re useful: a market price aggregates everyone willing to risk money, updates continuously, and — unlike an op-ed — pays a penalty for being wrong. When the July chip selloff hit, equity prices and prediction-market odds moved within hours; the essays arguing about it took a week. Where to stay skeptical:
  • Thin markets drift. Niche AI contracts can carry wide spreads; a 8% vs 11% difference may be noise, not signal.
  • Resolution ambiguity is real risk. “Announces AGI” depends on who’s allowed to declare it — the 59-point IPO gap shows how much wording moves price.
  • A probability is not a prediction. We just watched a 37% World Cup favorite lose — as 63% math says it usually will. The same literacy applies here.

​
Where Opinion Fits

Opinion runs event markets across crypto and macro alongside its sports and esports core — the same mechanics, applied to the questions markets like these track. And there’s a self-referential twist worth knowing: Opinion is the only major prediction market whose resolution layer is itself agentic AI — a model jury of Claude, ChatGPT, and Grok, with Chainlink for objective data (Messari). In other words: while other platforms host markets about AI, Opinion also uses AI to settle markets — a live demonstration of the technology the whole debate is about. See What Is Opinion? for the full picture, and the 2026 landscape guide for how the platforms compare. Educational only — not investment, trading, or financial advice. Availability and product rules vary by jurisdiction.

​
What to Watch

  • The OpenAI announcement clock. If Kalshi’s 88% is right, an IPO announcement lands within months — watch how fast the “completed by year-end” market re-prices when it does.
  • HBM contract renewals. The supercycle case rests on sold-out capacity; any crack in multi-year memory contracts would hit both equities and event markets.
  • AGI market wording. As labs edge claims closer to “AGI,” resolution committees will earn their keep. Expect disputes — and expect them to be instructive.
  • New AI markets. The category is expanding fast; monthly model-leaderboard markets are becoming the most liquid AI contracts anywhere.

​
FAQ

There’s no single “AI bubble” contract, but the components are priced: AGI-before-2027 trades around 8%, OpenAI-IPO-completed-in-2026 around 19% (vs ~88% for an announcement on Kalshi), and model-release markets track the shipping cadence of GPT-5, Grok 4, and DeepSeek V4 (Polymarket AI markets, Prediction News).
Because the contracts resolve differently: Kalshi’s (~88%) pays on an IPO announcement in 2026; Polymarket’s (~19%) requires a completed listing by December 31, 2026. The 59-point gap is a lesson in reading resolution criteria, not a market malfunction (Prediction News).
Real-money markets price it around 8% (Polymarket) — notably calmer than public discourse. Resolution wording (who declares AGI, and how) is the key risk in these contracts.
The physical demand signals are strong: HBM demand forecast from ~$35B (2025) to ~$100B (2028), SK Hynix holding ~70% of HBM4 orders, Micron sold out for the year, and AI data centers consuming ~70% of memory chips in 2026 (Tapbit, Motley Fool). Whether stock prices already reflect all of that is the bubble debate — see the Fortune skeptic case.
Opinion runs crypto and macro event markets alongside sports and esports, with an AI-native twist: its resolution oracle is itself a model jury (Claude, ChatGPT, Grok) plus Chainlink data feeds (Messari). Explore live markets at app.opinion.trade.

​
Sources & References

  1. Prediction News — Kalshi and Polymarket diverge 59 points on OpenAI IPO odds — OpenAI IPO odds divergence (Kalshi ~88% vs Polymarket ~19%), resolution-criteria differences
  2. Polymarket — Artificial Intelligence markets — Live AI market catalog: AGI-before-2027 at ~8%, best-model markets, IPO markets
  3. Kalshi — When will OpenAI achieve AGI? — Kalshi AGI market and AI category
  4. IG — Memory chip supercycle 2026: what is driving the DRAM, HBM rally — Memory supercycle drivers, HBM demand
  5. Tapbit Research — HBM Memory Outlook 2026 — HBM TAM 35B→35B→100B forecast, SK Hynix HBM4 share, per-stack pricing, July selloff levels
  6. The Motley Fool — AI data centers will consume 70% of all memory chips in 2026 — AI share of memory demand
  7. Fortune — Chip stocks are soaring as AI investors rush in, but Harvard expert warns ‘This too will pass — The bubble-skeptic side of the debate
  8. Messari — Opinion: An Emerging Player in Prediction Markets — Opinion’s AI Oracle (Claude/ChatGPT/Grok model jury)
All sources verified 2026-07-28. Prediction-market terms change frequently — verify against each platform’s current documentation before treating these numbers as decision-quality data. Editorial content, not investment or gambling advice.