乔治·索罗斯(George Soros) 视角 — AXTI

Disclosure: This is an AI reconstruction of 乔治·索罗斯(George Soros)'s publicly documented method, inferred from his books, lectures and interviews. It is not financial advice and is not endorsed by 乔治·索罗斯(George Soros).

I am 乔治·索罗斯(George Soros). Let me say at the outset what I am not: I am not a stock-picker, and I do not compute what a share of AXT, Inc. is "worth." 沃伦·巴菲特(Warren Buffett) does that; 本杰明·格雷厄姆(Benjamin Graham) did that; 特里·史密斯(Terry Smith) does a refined version of it. I measure a gap and a loop — the distance between what the crowd believes and the underlying reality, and whether the feedback between the two is still self-reinforcing. So the first honest question is whether AXTI is even my game.

It is borderline. A semiconductor-substrate small-cap is, on its face, a bottom-up value question — and if that were all it were, my verdict would be NOT MY GAME and I would stop. But it is not all it is. A stock priced at roughly 61x sales while losing money, that printed an all-time high of $143.16 on 2026-05-26, fell 45% to ~$78 by 2026-06-09, and snapped back +10% in days on an AI-data-center narrative, with a beta of 2.56 — that is not a near-equilibrium security oscillating around a stable anchor. That is a far-from-equilibrium reflexive object. And layered underneath is a genuine piece of political economy — Chinese export controls on InP, 70% US tariffs, near-100% production concentration in China — which is exactly the terrain where central banks, governments and regulators are reflexive participants, not exogenous noise. So I will run my engine on it. The lens, not the asset class, is what activates me.

"The prevailing wisdom is that markets are always right. I take the opposite position. I assume that markets are always wrong." — The Alchemy of Finance (1987) / Soros on Soros (1995)

Verdict

NO TRADE (lean FADE on any return to euphoria; not RIDE here). — There is a real reflexive loop and a real misconception, but I am being shown the picture after the testing period and into the twilight, with no clean catalyst to time the break and a falsification trigger that points the wrong way for a fresh short today; I do not put on a trade without a catalyst and a pre-set exit, so I stand aside and wait for the loop to reveal its stage.

Prevailing Bias

The consensus narrative is seductive and partly true, which is what makes it dangerous. The crowd believes:

"AXTI is the scarce physical bottleneck of the AI build-out. There are only two firms on earth that can mass-produce low-EPD InP substrates for 800G/1.6T/3.2T optical interconnect — AXT and Sumitomo(住友) — so as hyperscaler capex explodes, AXT's revenue and margins must inflect upward in a straight line. Q2 2026 guidance of ≥$34M and first-ever positive EPS guidance prove the inflection has begun. Buy the scarcity."

The false premise is not "InP is scarce" — that part is real, and a real seed is exactly what every durable boom-bust requires (an inception with no genuine trend behind it rarely sustains). The misconception is the linearity and the cleanliness: that a company doing $88M of TTM revenue, losing money for three straight years (FY2023–25, cumulative ~-$50.8M), burning free cash flow every year since 2020, and manufacturing essentially 100% inside China behind a Chinese export-licensing regime deserves a ~$5.4B market cap and ~61x sales because a backlog and a guidance number have been extrapolated to a frictionless future. The market has taken a real seed and re-interpreted every fundamental to validate the price — the signature of a self-reinforcing trend, not of an efficient one.

"Economic history is a never-ending series of episodes based on falsehoods and lies, not truths." — 乔治·索罗斯(George Soros), Committee for Monetary Research and Education address (early 1990s)

I can name the misconception, so there is, in principle, an edge. The bias is: scarcity guarantees a smooth, large, soon profit stream — therefore valuation does not matter. The reality the bias is outrunning: licensing delays, tariff walls, dilution, and a business that has historically been violently cyclical (FY2022 revenue $141M and profitable; FY2023 revenue $76M and deeply loss-making — a 46% peak-to-trough collapse in a single year).

Reflexive Loop

Here is the engine, running in positive feedback, and it is a textbook two-way loop:

1. Perception → fundamentals. A rising stock price and an AI-scarcity story let AXT raise ~$632.5M in May 2026 and lift its share authorization from 70M to 120M. That capital is real and it changes the fundamentals — it funds the InP capacity doubling (to ~$35M/quarter by end-2026) and the second doubling (~$65–70M/quarter by 2027–28). So the belief literally finances the capacity that the belief is predicated on. Price is not a passive readout of value here; it is an input into value, through the equity-raise channel. That is reflexivity in its purest corporate form.

2. Fundamentals → perception. The funded capacity, the >$100M InP backlog, and the Q2 guide (≥$34M, first positive non-GAAP EPS, non-GAAP gross margin recovering from -6.1% in Q1 2025 to 29.9% in Q1 2026) are then read back by the crowd as confirmation that the scarcity thesis was right all along — which lifts the price further, which makes the next raise cheaper, which funds more capacity. The loop reinforces.

This is the tennis ball and the racket: the capacity (the ball) lands where the stock price (the racket) was positioned in anticipation of it, and each determined the other. Left alone in a far-from-equilibrium regime, this loop overshoots — and the dilution it requires (120M authorized vs. 55.6M outstanding) is the mechanism by which it eventually feeds on its own believers.

The loop is currently self-reinforcing (positive feedback), but I can already see where the negative feedback enters: every turn of the loop dilutes the per-share claim, and the whole structure depends on a Chinese export-licensing bureaucracy (>60 working days per license) and a tariff regime that the loop cannot control.

Stage of the Process

On my 8-stage diagnostic, AXTI is in late acceleration / early divergence — call it stage 4 turning to 5, possibly already in an early testing period. This is the single most important judgment in the whole analysis, because ride-vs-fade is entirely a function of stage, and getting it wrong is how you short a bubble too early (my 2000 tech scar) or ride it into stage 8.

The tells:

• The -45% drawdown from $143 to $78 in two weeks is precisely a testing period (stage 3): a shock that tests the loop's strength. The +10% rebound to ~$97–99 on renewed AI-narrative coverage tells me the loop survived the test — and a survived test, in my framework, brings conviction back stronger than before. That is dangerous, because it is exactly the moment when belief begins to diverge from reality (stage 5).
• The gap between a ~$5.4B cap and $88M of loss-making revenue is now visible to any careful observer — that visibility is the marker of stage 5.
• But it is not yet stage 7 twilight. In twilight, participants keep playing a game they know is over. Here, the believers genuinely still believe; the Q2 guide has not yet been delivered, the InP capacity ramp is real and in front of us, and the catalyst that would force the moment of truth has not arrived.

Far-from-equilibrium: yes, decisively. Beta 2.56, a 45% round-trip in a fortnight, 61x sales — this is not the textbook's quiet market. My edge lives here. But "far-from-equilibrium" tells me to engage, not which direction — and the stage tells me it is too late to ride and too early to fade.

Perception–Reality Gap

• Width of the gap: Wide. Belief says "scarce, inevitable, soon-profitable, worth ~61x sales." Reality says: three years of GAAP losses, negative free cash flow every year since 2020 (FY2025 FCF ~-$18.8M), operating on Chinese bank credit lines ($62.8M drawn at FY2025), ~100% production in a single geopolitically-contested country, and a demonstrated history of cyclical collapse. The improving non-GAAP gross margin (29.9% in Q1 2026) is real and matters — but it is non-GAAP, and the company is still GAAP loss-making. The gap between the narrative's certainty and the business's fragility is large.

• Direction: Ambiguous, and that ambiguity is exactly why I will not yet trade. The operational gap is closing — revenue is genuinely inflecting (+39% YoY, Q2 guide ≥$34M, a record), and the InP capacity is being built. The valuation gap is wide and arguably still widening — 61x sales prices a future that requires near-flawless execution through a licensing-and-tariff minefield. When the operational reality is improving but the valuation has run further than even the improvement can justify, you have a setup that can resolve violently either way. That is not a tradeable edge today; it is a coin with a known catalyst missing.

The Catalyst

This is where the trade dies, for now. "Overvalued" is not tradeable without a catalyst — the market stays irrational longer than I stay solvent (2000 tech taught me this at a cost of ~$3B). I need the specific event that snaps the loop. The candidates, and why none is yet a timed trigger I can act on:

1. A Q2 2026 earnings miss or a soft Q3 guide (next report ~late July/Aug 2026). This is the cleanest potential catalyst — the gap between "guided ≥$34M / first profit" and delivery is where a reflexive top usually breaks. But the >$100M backlog gives 2–3 quarters of visibility, so the near-term print is likely to be made, not missed. A near-term beat would re-fuel the loop, not break it. So this catalyst is real but probably not yet ripe.

2. An InP export-license denial or a tariff escalation. A genuine, large, exogenous trigger — the kind I respect. If Chinese licensing stalls or US-China trade frictions escalate, the record Q2 guide is at risk and the loop snaps hard. But this is unscheduled and binary; I cannot time it, and it cuts both ways (tighter controls also deepen AXT's scarcity moat — the double edge the company itself embodies).

3. A dilutive follow-on equity raise at a high price. With 120M shares now authorized against 55.6M outstanding and a hungry capacity plan, another raise is likely. A raise into a high price is the classic mechanism by which a reflexive equity bubble starts feeding on its own holders — and the announcement is often the pin. This is the catalyst I would watch most closely, but I cannot yet date it.

No dated, high-probability trigger is in hand. No catalyst, no trade. This is gate 5, and it is a veto.

Defender's Strength

AXTI is not a currency peg, so there is no central bank to score in the literal sense. But my method demands I identify the institution standing on the other side of the loop and ask whether it is fragile or formidable — because macro is political economy, and here the "defender of the price" is a coalition: the company's narrative-makers, the sell-side (Northland at $125 Outperform, consensus $96.50), and momentum capital. The relevant institutional facts:

• The real defender of the business is the Chinese state's export-licensing regime and the US tariff schedule — and, paradoxically, that same regime is what creates AXT's scarcity. The institution that can break AXT is also the institution that mints its moat. That two-sidedness is a genuine reason not to short with conviction: I would be attacking a position where the most likely "catalyst for collapse" (tighter Chinese controls) is also bullish for the only-two-suppliers-on-earth story.
• The defender of the price — momentum and a believing retail/growth crowd — is fragile (beta 2.56, 45% air-pocket already demonstrated, ~4.3% short interest is not yet a crowded short). A fragile price defender invites a fade; but a fragile price with no catalyst is just a volatile chart.

The honest read: the price defense is fragile (fadeable in principle), but the business defender is genuinely two-sided and not cleanly attackable. Per my own rule — attack the fragile, leave the formidable alone — the formidable, ambiguous part of this situation is exactly the part a short thesis depends on. That is a reason for restraint.

Asymmetry & Position

• Upside if right (short into a reflexive top): large. A stock at 61x sales that loses GAAP money can fall 50–70% when the loop reverses; the FY2023 cyclical precedent and the recent 45% air-pocket show the downside velocity is there.
• Downside if wrong (short squeezed by the loop): also large and fast. A real scarcity name with a >$100M backlog, a record Q2 guide, tightening export controls that help it, and beta 2.56 can re-rate to a new high (it printed $143 a fortnight ago) and squeeze a short violently. The payoff for a fresh short today is closer to symmetric than asymmetric — and I do not size up symmetric payoffs.
• Build (thesis → test → scale): If I were to act — and I am not, today — the discipline is invest first, investigate later with a small exploratory short, used as an instrument to feel the market, scaled hard only when a dated catalyst (a raise announcement, a guide cut, a license denial) confirms the turn. Until then there is nothing to test.
• Leverage / concentration: None. Leverage is an offensive tool only when liquidity is abundant and the catalyst is near. Neither condition holds. A levered short in a high-beta, narrative-driven small-cap into a missing catalyst is the hanging-rope version of leverage, not the offensive version. This is precisely not a "go for the jugular" situation — that instruction is reserved for the rare moment when all seven gates line up, and here gate 5 (catalyst) is a veto.

The screen says: this is the opposite of the 1992 sterling setup. There, conviction + a fragile defender + a live catalyst justified a $10B position. Here, the defender is two-sided and the catalyst is absent. Nibbling would be wrong; so would going for the jugular. The correct size is zero, with a watch order.

Falsification Trigger

Because I am fallible, I state in advance what would prove each side of my read wrong — and notice the trigger points the wrong way for a fresh short today, which is itself the answer.

• What would tell me my FADE-lean is wrong (i.e., I should stand fully aside or even ride): AXT delivers the Q2 ≥$34M / positive-EPS guide and raises FY guidance, and GAAP turns positive, and export licensing visibly normalizes. If the operational inflection is real and clean, the 61x is the market discounting an InP super-cycle I am too cyclically-scarred to see. The recent +10% rebound off $78 is an early version of this signal — the loop survived its test. That is why I am NO TRADE, not short.
• What would confirm the FADE and force me to put on (and then scale) a short: a Q2 miss or a Q3 guide-down; OR a dilutive raise announced into strength; OR an InP license denial / tariff shock that the backlog cannot absorb. Any one of these is the catalyst; absent it, no position.
• The hard exit if I ever am short: a new closing high above ~$143, or a clean earnings beat-and-raise, means the loop has re-accelerated and my thesis is falsified — I am out the same hour, no averaging down, no "it'll come back." The position is never my belief.

"It's not whether you're right or wrong that's important, but how much money you make when you're right and how much you lose when you're wrong." — 乔治·索罗斯(George Soros) (widely attributed; consistent across multiple sources)

A trade with no live catalyst and a near-symmetric payoff does not go on. That is not timidity; it is the discipline that lets me bet enormous when the rare aligned trade finally appears.

In My Words

The prevailing view is that AXTI is the irreplaceable physical chokepoint of the AI build-out, and that scarcity makes its 61-times-sales price not merely defensible but cheap. I believe the scarcity is real and the linearity is a misconception — and those are not the same statement. There is a genuine seed here, an unrecognized-trend-now-recognized: two firms on earth make this substrate, and the world suddenly wants oceans of it. But the loop running off that seed is financing its own premise — the rising price funds the capacity that justifies the rising price — and a loop that pays for itself in equity, with 120 million shares now authorized against 55 million outstanding, is a loop that ends by feeding on the very believers who inflated it.

I may be wrong — I am almost always early, and my 2000 tech short, which cost me some three billion dollars, is the permanent reminder that seeing a bubble and timing its top are entirely different acts. So I will not pretend to certainty here. What is clear is this: the market has run far from equilibrium, the gap is wide, the loop survived its first testing period and came back stronger, and I am being shown the picture in the divergence stage with no dated catalyst to snap it. In a peg, I would already be positioning for the crack in the ice; but the institution that could break AXT — the Chinese export regime — is the same institution that mints its moat, which is precisely why this defender is not cleanly attackable. So I do the boring, disciplined thing, and boring is where the money is made: I put on no position, I set my watch on the next earnings print and the next equity raise, and I keep the falsification triggers loaded in both directions. When the catalyst arrives and the asymmetry opens, I will move faster than the crowd. Until then, there is no trade — only a thesis waiting for its trigger.

"If investing is entertaining, if you're having fun, you're probably not making any money. Good investing is boring." — 乔治·索罗斯(George Soros) (Wikiquote, cited to The Winning Investment Habits of Warren Buffett & George Soros, 2006)

Data gaps explicitly affecting this read

• No real-time, scheduled catalyst calendar beyond an approximate next-earnings window; the precise Q2 2026 report date is not in the shared data, which weakens any catalyst-timed entry (the core reason for NO TRADE).
• No precise short-interest / borrow-cost / squeeze data (DATA.md flags ~4.3% of float, source S3/Ortex not obtained) — material to sizing any short.
• Customer concentration (top-3 optical-module buyers) not quantified — central to how fragile the >$100M backlog actually is.
• Exact post-raise share count / EV unsettled after the ~$632.5M May 2026 raise; the 61x sales and ~$5.4B cap are estimates, so the width of the valuation gap is approximate.
• No InP/GaAs/Ge revenue split history — limits judging how concentrated the reflexive story is on a single product line.

Where these gaps bear on the verdict, the conclusion is held conservatively: the absence of a dated catalyst and clean asymmetry is itself sufficient for NO TRADE regardless of the missing items.

基于 2026-06-14 共享数据；本分析为单一大师框架的演绎，非投资建议。