← Back to Research
Artificial Intelligence / Macro & Capital

AGI Is Here, Capital Is Losing Its Bearings: Valuations Doubled Before Any Public Test

FutureX Research · AI Lab · 2026.05.18 · 26 pp · preview 6 pp · Revised 2026.11.02

🎧

Listen · Audio Summary

5-8 min · AI narration in English · abstract + all key findings

Bottom Line

AGI is here and capital is losing its bearings: Anthropic's valuation anchor doubled in four months, and public markets have not tested it.

Why now
Bloomberg reported on October 1 that Anthropic could list as soon as mid-November at a target of up to about $2 trillion, roughly twice its May post-money valuation.
What it means
For founders, private capital is concentrating in a few leaders; for investors, Anthropic's IPO pricing is this cycle's first real stress test.

Abstract

AGI is here and capital is losing its bearings: Anthropic's valuation anchor roughly doubled in under four months, and public markets have not yet tested it. Anthropic could list as soon as mid-November at a target of up to about $2 trillion, against a $965 billion post-money valuation in May (Reported). Public markets are already sensitive: on October 8, when the FT reported OpenAI's annualized revenue about $20 billion below earlier reports, the Nasdaq-100 fell 1.39%. Private money is concentrating in the leaders: DeepSeek's new round is reportedly at least RMB 80 billion. On the technology side, the open-weight Kimi K3 has overtaken closed flagships on a blind coding leaderboard. (Data updated through 2026-10-11)

Key Points

  • 01Bloomberg reported on October 1 that Anthropic could list as soon as mid-November at up to about $2 trillion, twice its $965 billion May post-money; its draft prospectus shows $11.5 billion of Q2 revenue (Reported).
  • 02On October 8 the FT reported OpenAI's annualized revenue near $50 billion, about $20 billion below earlier reports; that day the Nasdaq-100 fell 1.39% and Nvidia 2.94% (Reported; Public market data).
  • 03AI hedge fund Situational Awareness lost 67% in July, and Citadel took over most of its stock portfolio on July 30; on July 28 the Nasdaq-100 fell 10% from its high into a correction (Reported).
  • 04DeepSeek's new round is reportedly at least RMB 80 billion (about $12 billion), well above the roughly $7.5 billion first planned, with CATL and Tencent the largest investors (Bloomberg, Reuters, October 6; Reported).
  • 05Moonshot AI fully open-sourced the 2.8-trillion-parameter Kimi K3 on July 27; at 1679 Elo it topped the WebDev Arena blind frontend-coding leaderboard, the first open model to beat closed flagships there (Reported).

1. The Open-Closed Gap Is Vanishing: Kimi K3 Tops a Blind Coding Leaderboard at 1679 Elo

Moonshot AI released Kimi K3 on July 16: a 2.8-trillion-parameter MoE model with 896 experts, about 104B active parameters and a 1M-token context. Late on July 27 it open-sourced the full model. The weights carry a modified MIT license, and three training-infrastructure stacks shipped with them: MoonEP, FlashKDA and AgentEnv. The community got both the model and the method for training a 3-trillion-class model (Reported). Hugging Face CEO Clem Delangue said K3 passed 4,000 likes within 30 minutes and topped the trending chart, the fastest launch in the platform's history (Reported).

ModelWebDev Arena blind frontend-coding score (Elo)
Kimi K3 (open)1679, first
Claude Fable 51631
GPT-5.6 Sol1618

K3 is the first open model to overtake closed flagships in a hard specialty. On composite intelligence indices it ranks third to fourth worldwide, depending on the snapshot (Reported).

Support arrived on day one. Huawei Ascend 950, Alibaba Cloud's Zhenwu M890, Nebius, Baseten and Fireworks shipped Day-0 support, and Cursor and Cognition (Devin) announced integrations (Reported).

Rivals answered fast. Musk reportedly said on July 18 that xAI is training a 2-trillion-parameter model aimed at surpassing K3. On July 24 Jensen Huang shared an open letter initiated by a16z and co-signed by Nvidia and others, arguing that open and closed frontier models should coexist (Reported).

2. Private-Market Premia Keep Expanding, While the Nasdaq-100 Entered a Correction on July 28

Private-market premia are still expanding. Per National Business Daily, citing people familiar with the matter, Moonshot AI sent investors a listing proposal in mid-July and aims to list in Hong Kong within about six months. It is first closing a round at a $31.5 billion pre-money valuation, more than 7x higher than half a year earlier (per our prior issue), and plans pre-IPO talks in August targeting $50 billion pre-money (Reported).

DeepSeek now carries three marks at once, which puts the fuzziness of private pricing on the table:

MarkValuationSource
Planned new round~RMB 500B (~$74B)Reuters, July 15 (Reported)
June first outside round of $7.4B, post-moneynearly RMB 400BMedia reports (Reported)
Implied by shareholder filings~RMB 350.9B (~$52B)Shareholder filings (Per company disclosure)

The last two differ by more than 10%. Reuters also reported that DeepSeek is weighing a STAR Market listing.

Zhipu, listed in Hong Kong since January 8, completed a placement of about HK$31.4 billion on July 13, extending its research-first narrative (Reported).

The deceleration showed up in prices for the first time. On July 28 the Nasdaq-100 fell 10% from its record into correction territory (Bloomberg; Public market data). The Philadelphia Semiconductor Index logged its longest losing streak of the year, and Nvidia sat more than 15% below its May peak (NYT; Public market data). Bloomberg pointed to two causes: worries about circular funding, and Chinese chip competition. On July 27 CXMT debuted in Shanghai and rose as much as 470% on its first day (Public market data).

The fund side pulled back too. Per an investor letter reported by Reuters on July 31, AI hedge fund Situational Awareness, founded by Leopold Aschenbrenner, lost 67% in July. Its assets fell from a peak of about $45 billion (per CNBC) to about $10 billion, and on July 30 Citadel bought most of its stock portfolio (Reported).

3. Anthropic's IPO Is the First Public Pricing Anchor, and SpaceX's 40%+ Drawdown Shows It Cuts Both Ways

Anthropic filed its S-1 confidentially on June 1. Bloomberg reported on July 15 that Morgan Stanley, Goldman Sachs and JPMorgan were arranging investor meetings ahead of a Nasdaq listing as soon as October (Reported). Under the rules, the S-1 must go public at least 15 days before the roadshow. Its May round raised $65 billion at a $965 billion post-money valuation (Reported). The timetable has since slipped several times; the latest report points to mid-November at the earliest (see Latest developments below).

Invest-for-orders deals are spreading. On July 22 AMD announced it would invest up to $5 billion in Anthropic, which will buy up to 2GW of AMD's next-generation AI chips from the first half of 2027 (Reuters; Per company disclosure).

Public markets have already started pricing this private company. On July 30 Amazon reported Q2 net income of $62.6 billion, of which about $53.4 billion was pre-tax non-operating gains, mainly from revaluing its Anthropic stake (Per company disclosure).

The control group matters as much. OpenAI's listing has slipped toward 2027 (Bloomberg, Reported). SpaceX, which listed in June, trades more than 40% below its June peak (NYT, Public market data). A public anchor works both ways: it can validate a premium or puncture a narrative. Per Bloomberg, DeepSeek could file as soon as end-2026 and list in 2027 (Reported).

FutureX Position · Open Source Breaks the Deadlock: Six of Seven Signals Are Lit, and the Bubble Sits in the Layer That Must Keep Refinancing (Xiamen keynote, 2026-09-03)

The Xiamen keynote hardens this report's thesis that AGI has arrived and capital is losing its bearings. It also gives a measurable mechanism to our cycle call, which places top-of-cycle risk where hyperscaler capex growth peaks. The method for spotting the top is unchanged: model-layer revenue is an unreliable denominator, so watch only for the point when capex growth starts to slow.

The mechanism is a seven-signal framework. Since March 2026 FutureX has scored valuation, funding pace, new-entrant supply, narrative, fundamentals, liquidity and exits every day, and our August reading had six of the seven lit. Liquidity has triggered: on August 17 the 30-year Treasury yield touched 5.31%, crossing our threshold (Public market data). Exits have only diverged: no AI company listed overseas in Q1, while 155 Chinese companies went public in the first half (Reported).

Consensus takes three forms, the same three as in 2000:

FormFigureComparisonSource
ConcentrationTop 10 US stocks are ~37% of S&P 500 market value27% at the 1999–2000 peakGoldman Sachs
Circular financingSecuritized data-centre debt ~$26.5B in 2025$1.3B in 2022; JPMorgan expects $30–40B in 2026Reported
Narrative samenessOnly 22 model-layer deals in 2025Fewer than 10 of 100 companies funded; attrition above 90%36Kr / Zero2IPO

Our view: this cycle does not need a verdict on whether a bubble exists. The job is to find the layer where value sits. The bubble is in the middle layer, which carries the most debt and must keep refinancing. Value sits in two places: the application layer, which captures what falling costs release, and the open-source chokepoints every layer passes through.

Full deck: /reports/open-source-breakthrough (first five pages public).

Latest developments, newest first (data current as of 2026-10-11)

  • 2026-10-11 · As of this date, no public version of Anthropic's S-1 was on the SEC's EDGAR system. A company that filed confidentially must make its S-1 public at least 15 days before the roadshow; with a roadshow in the week of November 9, as reported, the public version would be due by late October · Verified (EDGAR)
  • 2026-10-09 · The Nasdaq-100 rose 0.51% · Public market data (Tencent Securities)
  • 2026-10-08 · The Financial Times, as relayed by Reuters, reported that OpenAI told investors its annualized revenue was approaching $50 billion at the end of September, about $20 billion below the roughly $70 billion reported in late September by outlets including Reuters. Citing a person familiar with the matter, the FT said the gap arose because some OpenAI investors tried to compare directly with Anthropic, and the two companies count differently: Anthropic counts sales made through cloud partners such as AWS and Google Cloud, and OpenAI does not · Reported
  • 2026-10-08 · On the day of the report, the Nasdaq-100 closed at 30,725.81, down 1.39%, and Nvidia closed at $230.48, down 2.94%. Reuters' market close also cited higher crude prices and weak chip stocks · Public market data (Tencent Securities; Reuters)
  • 2026-10-06 · DeepSeek is close to completing a round of at least RMB 80 billion (about $12 billion), which people familiar with the matter say could approach $15 billion, with CATL and Tencent the largest investors. The round originally aimed at about $7.5 billion, and the company plans to restructure for an IPO in early 2027 once it closes. Reuters, citing a source, put the round above $12 billion · Reported (Bloomberg, via The Decoder and others; Reuters)
  • 2026-10-06 · Moonshot AI closed its last pre-IPO private round at a valuation of about $50 billion and plans to list in Hong Kong in Q1 2027; the report did not say whether the $50 billion is pre- or post-money · Reported (Bloomberg)
  • 2026-10-05 · The 10-year US Treasury yield closed at 5.31%, its highest close since May 2002 · Public market data (Federal Reserve H.15, via FRED)
  • 2026-10-01 · Anthropic aims to list as soon as mid-November, could start marketing its IPO as early as the week of November 9, and targets a valuation of up to about $2 trillion. Our September 25 update recorded an October roadshow and a listing before the November 3 midterms, so the listing date has slipped by about two weeks · Reported (Bloomberg, via Yahoo Finance)
  • 2026-09-28 · Anthropic's draft prospectus, seen by Reuters and the Financial Times and relayed by Fortune and The Decoder on September 28–29: 2025 revenue of about $4.6 billion, roughly 12 times the prior year; Q1 2026 revenue of $4.73 billion and Q2 of $11.5 billion; a 2025 operating loss of $8.06 billion (prior year $2.98 billion) and a net loss of about $42 billion, of which about $34 billion was accounting charges; two customers made up nearly a quarter of 2025 revenue · Reported
  • 2026-09-24 · Anthropic's seven co-founders are seeking a combined 50.1% of the vote on most corporate matters through super-voting shares, as long as at least three keep a minimum stake; each holds an economic stake of about 2%. The company is asking shareholders to approve the structure, and the Long-Term Benefit Trust would still select most board members · Reported (first by The Information; relayed by Reuters and TechCrunch, September 24–25)
  • 2026-09-24 · Medical AI company OpenEvidence raised $250 million at a $15 billion valuation from several hospital systems and a16z (Sina Tech says the two led the round); it was valued at $12 billion in January and has raised more than $1 billion over the past year · Reported (first reported by Business Insider; Sina Tech)
  • 2026-09-15 · OpenAI is in early, investor-initiated talks on a pre-IPO round at a valuation above $1.2 trillion, up from $852 billion in its March round. An internal document that surfaced the same week projects $278 billion of cumulative negative free cash flow over 2026–2030 and about $856 billion of cumulative compute and infrastructure spending through 2030; at the current burn rate, existing funds run out in 2028. OpenAI has recently cut token prices to counter Anthropic and Chinese open-weight models · Reported (Bloomberg, September 15; Cailian Press, September 19)
  • 2026-09-14 · Anthropic CEO Dario Amodei published “We Must Pace the Frontier” over the weekend of September 12–13, urging the industry to slow the pace of frontier capability gains; Sam Altman and Elon Musk publicly echoed the call. On September 14 the Nasdaq Composite fell 147 points (0.6%) to 26,186, Nvidia fell 3.4% and Micron 5.3% · Verified (CBS News)
  • 2026-09-13 · Anthropic has selected Nasdaq for its IPO, with bankers discussing a target valuation of about $2 trillion; Goldman Sachs, JPMorgan and Morgan Stanley are the lead underwriters. Less than four months after the May Series H at $965 billion post-money, the pricing anchor has roughly doubled · Reported (Bloomberg; first disclosed by Business Insider)
  • 2026-09-08 · AI coding company Cognition raised $2 billion at a $48 billion post-money valuation, led by a16z, Accel, Founders Fund, General Catalyst and Avenir; annualized revenue rose from $492 million in May to $900 million, and the valuation nearly doubled from $26 billion in May. Bloomberg's September 2 report of about $1 billion at roughly $47 billion is superseded by this disclosure · Verified (TechCrunch)
  • 2026-09-04 · Anthropic pushed publication of its public prospectus to late September, from an earlier plan of as soon as the following week, with marketing no earlier than mid-October and a listing possibly days before the November US midterms; it is also finalizing a $15 billion revolving credit facility. The IPO banks are Morgan Stanley, Goldman Sachs, JPMorgan Chase and Citigroup; Anthropic declined to comment · Reported (Reuters)
  • 2026-09-02 · Moonshot AI confidentially filed a Form A1 with the Hong Kong stock exchange this week and is raising at a $50 billion pre-money valuation, possibly its last round before the IPO; in July it closed a Series F of more than $3.5 billion at a $35 billion post-money valuation. Moonshot said it does not comment on market rumors · Reported (LatePost)
  • 2026-08-10 · Nvidia, with Apollo, Blackstone, BlackRock, Brookfield, Goldman Sachs and KKR, plans an independent financing platform targeting more than $500 billion of third-party capital for AI infrastructure; Jensen Huang said compute is becoming an investable asset class. Nvidia closed down 2.86% that day · Verified (Nvidia company statement); share price is public market data

What This Means for Founders

Capital has not shrunk, it has concentrated: Cognition's valuation nearly doubled to $48 billion in four months, and Bloomberg reported on October 6 that Moonshot AI closed its last pre-IPO round at a valuation of about $50 billion (Reported). The frontier labs still need to raise on the same scale: OpenAI expects its existing funds to run out in 2028 at the current burn rate. The open-weight Kimi K3 has overtaken closed flagships on a blind frontend-coding leaderboard, and we think value sits in the application layer once model costs come down. For exit paths, watch Anthropic's public S-1 first: OpenAI's annualized revenue differs by about $20 billion between two counting methods, and the market is now sensitive to how AI revenue is counted.

What This Means for Investors

Money is concentrating in the leaders: Anthropic's valuation anchor went from $965 billion to about $2 trillion in under four months, and public markets have not validated it. The anchor's cost is now visible: the 2025 operating loss was about 1.75 times revenue, and if the public S-1 counts revenue more than 10% differently from earlier reports, the anchor gets weaker. Risk may shift from equity to credit: Nvidia and its partners aim to mobilize more than $500 billion of third-party capital for compute infrastructure, so watch loan-to-value ratios, tenor and residual-value assumptions in the first projects. China's private market is moving the other way: DeepSeek's new round is reportedly at least RMB 80 billion, well above the roughly $7.5 billion first planned. This is not investment advice.

Where We Could Be Wrong

  • 01If Anthropic lists by the end of November 2026 without its valuation being cut below $1.5 trillion, public markets will have accepted the private pricing anchor, and our call that capital is losing its bearings is too strong.
  • 02We think public markets read the frontier labs' call to pace the frontier as a signal of lower capex. If hyperscaler capex growth has still not slowed by the end of 2026, the September 14 sell-off was only sentiment, and that transmission chain does not hold.
  • 03We think China's leading model companies still carry a narrative premium. Bloomberg said on October 6 that DeepSeek's round of at least RMB 80 billion was close to completion; if it has still not closed by the end of 2026, that view is wrong.

Key Questions

How does the open-sourced Kimi K3 perform — did it really beat closed flagship models?

Yes. Moonshot AI fully open-sourced the 2.8-trillion-parameter MoE model Kimi K3 (bespoke MIT-style Kimi K3 License with revenue-tiered commercial terms) on July 27, 2026. It topped the WebDev Arena blind coding leaderboard at 1679 Elo, ahead of Claude Fable 5 (1631) and GPT-5.6 Sol (1618) — the first open model to beat closed flagships there. Moonshot also released three training-infrastructure stacks (MoonEP, FlashKDA, AgentEnv), with Day-0 support from Huawei Ascend, Alibaba Cloud, Nebius, Cursor and Cognition.

What exactly happened in the July 2026 AI capital deceleration, and what happened to the Situational Awareness fund?

In late July the deceleration turned tangible for the first time: per an investor letter reported by Reuters on July 31, Leopold Aschenbrenner's AI hedge fund Situational Awareness lost 67% in July, with assets falling from a reported ~$45B peak (CNBC) to ~$10B; Citadel bought the bulk of its stock portfolio on July 30. The Nasdaq 100 entered a 10% correction on July 28, and Nvidia sits more than 15% below its May peak.

What is the latest on Anthropic's IPO — when could it list and at what valuation?

Bloomberg reported on July 15 that Morgan Stanley, Goldman Sachs and JPMorgan are scheduling investor meetings for a Nasdaq listing as soon as October 2026 (Bloomberg reported on Oct 1 that the debut has slipped to mid-November at the earliest), following a $65B May round at a $965B post-money valuation. On July 22 AMD announced up to $5B of investment in Anthropic, which will purchase up to 2GW of AMD chips from H1 2027; Amazon's July 30 Q2 report booked $62.6B net income, including ~$53.4B pre-tax gains primarily from its Anthropic stake.

Watch & Listen

In China: search WeChat Channels for 「倩姐投AI」; full library → Qian on AI

Sourcing and standards

Compiled from public sources; data current as of 2026.10.11. The text separates four evidence grades: verified, reported, per company disclosure and public market data, and states the derivation behind every estimate. When we get something wrong, the correction is written into the report body with the original call left visible, and logged publicly.

Research standards & corrections →

📄 Full Report

Full report: 26 pages · provided to professional investors & partners only

This is the public preview. The full report includes the sections below. For compliance reasons it isn't posted publicly or offered as a free download. To request a copy, contact the FutureX team.

  • 🔒4. Big-Tech Conflict: Apple v. OpenAI and the Judicialization of the Hardware War (Late-July Escalation)
  • 🔒5. Compute & Energy: xAI's APR Energy Deal and the 'Buy Power' Arms Race (Memphis 1.2GW Timetable Set)
  • 🔒6. Commercialization: SoftBank × Sierra and the Distribution War of the Agent Economy
  • 🔒7. Value-Chain Rebalancing: Compute — Energy — Models — Agents — Devices
  • 🔒8. Risk Scenarios: Narrative-Premium Unwind (Partly Realized), Litigation Spillover, Power Constraints
  • 🔒9. Conclusions & Watchlist (2H 2026)
Request the full report →

Where we stand on this

Questions people ask next

Building in this space, or want to discuss this report? Write to us. We usually reply within 48 hours →

Related Research

Industry research from FutureX Capital's AI Lab, compiled from public information; not investment advice; contains no fund performance, AUM, or offer to raise capital.