🚨 Alert #371 Tue Jul-21 17:43:24
Google is developing a new custom AI server chip codenamed Frozen V2 that reportedly offers significantly higher efficiency by hard-coding parts of the Gemini AI system directly into the circuitry.
🎯 Tickers & verdicts
| ticker | direction | verdict | since close | 30m | day lo | day hi | intraday px |
|---|---|---|---|---|---|---|---|
| primary GOOGL | ▲ bullish | FRESH | -0.63% | +0.08% | 347.15 | 354.65 | |
| AMD | · unclear | PRICED | +7.16% | +1.03% | 522.30 | 546.54 | |
| AVGO | · unclear | PRICED | +2.04% | -0.12% | 380.86 | 390.09 | |
| AMZN | · unclear | FRESH | -1.00% | 0.00% | 246.50 | 250.30 | |
| peer NVDA | ▼ bearish | FRESH | +0.93% | -0.33% | 204.23 | 207.50 | |
| MSFT | · unclear | FRESH | -0.91% | +0.20% | 395.85 | 401.03 | |
| META | · unclear | FRESH | +0.35% | +0.18% | 640.50 | 657.09 |
🧭 Decisions
| strategy | ticker | action | side | rationale |
|---|---|---|---|---|
| S1 | GOOGL | ● enter | ▲ long | verdict FRESH, direction bullish -> buy GOOGL (product: Google is developing a new custom AI server chip codenamed Frozen V2 that reportedly offers significantly higher efficiency by hard-coding parts of the Gemini AI system directly into the circuitry.) |
| S2 | - | ○ skip | — | no counterparty in universe |
| S3 | NVDA | ● enter | ▼ short | verdict FRESH, direction bearish -> sell NVDA (product: Google is developing a new custom AI server chip codenamed Frozen V2 that reportedly offers significantly higher efficiency by hard-coding parts of the Gemini AI system directly into the circuitry.) |
| S4 | GOOGL | ○ skip | — | overnight drift: buy GOOGL at close, exit next close (product: Google is developing a new custom AI server chip codenamed Frozen V2 that reportedly offers significantly higher efficiency by hard-coding parts of the Gemini AI system directly into the circuitry.) [skipped at close: signal conflict — wanted buy but open S4 sell GOOGL exiting 2026-07-22] |
📈 Trades
| strategy | ticker | side | qty | entry | exit | P&L $ | status |
|---|---|---|---|---|---|---|---|
| S1 | GOOGL | ▲ long | 2 | 349.71 | 347.29 | -4.84 | closed |
| S3 | NVDA | ▼ short | 4 | 205.29 | 207.40 | -8.44 | closed |
Retrospectives
| ticker | mfe | mae | reflection |
|---|---|---|---|
| GOOGL S1 | +0.01% | -0.70% | Entry vs. outcome: The entry treated a speculative product rumor (custom chip "Frozen V2") as a FRESH bullish catalyst, but the market clearly didn't care — the trade never got more than +0.01% in the green and bled steadily to a -0.69% loss. Unverified chip-development leaks are slow-burn strategic news, not tradeable near-term catalysts, especially for a mega-cap where a single component rumor moves nothing. Improvement: Tighten the FRESH verdict to require a confirmable, dated event (official announcement, earnings, regulatory action) rather than "reportedly developing" language — or add a filter rejecting rumor-stage product news for stocks above ~$500B market cap, where such items are noise. Alternatively, require a minimum expected-impact score so a fresh-but-immaterial story doesn't trigger an entry. |
| NVDA S3 | +0.14% | -1.03% | Comparison: The entry treated a Google custom-chip headline as a fresh bearish catalyst for NVDA, but the market shrugged — the trade never got more than -1.03% in your favor and closed at a small loss. That's the expected outcome: "Google building its own AI chips" is a years-old narrative (TPUs), so this was an incremental iteration story mis-scored as FRESH, and the second-order competitive-threat logic was too indirect to move NVDA on the day. Improvement: Add a rule that news about a competitor's product only earns FRESH if it names a concrete customer loss, order shift, or revenue impact for the traded ticker; otherwise downgrade to STALE/no-trade. Alternatively, for indirect-catalyst shorts, require price confirmation (e.g., stock down on the news before entry) — here NVDA was flat-to-up, which should have vetoed the entry. |
📖 Glossary
- S1 primary — trades the alert's primary ticker in the direction the article itself indicates; exits at the same day's close.
- S2 counterparty — trades the inferred counterparty ticker (weakest signal, LLM-inferred sympathy); exits at the same day's close.
- S3 peer — trades an inferred peer ticker (also LLM-inferred sympathy); exits at the same day's close.
- S4 overnight — holds the primary ticker from today's close to the NEXT day's close.
- Verdicts are decided top-to-bottom, first match wins; REPRICING is also the fallback when nothing else matches (2026-08-08 #220).
- UNKNOWN — a needed price input (move since close or move in the last 30m) is missing.
- PRICED — |move since previous close| ≥ 2.0%: already reflected in price.
- REPRICING — |move in the last 30m| ≥ 0.5%: the tape is actively moving on it. Also covers a quieter move — |move since previous close| in [1.0, 2.0) with a quiet last-30m tape: still developing, not yet priced.
- FRESH — |move since previous close| < 1.0%: a genuinely new move.
- materiality — how market-moving the story was judged, 0-100 — alerts require materiality ≥ 60.