Snowflake SNOW

Reference table

1. What the business is, and two segment blocks that actually reconcile

Snowflake operates a consumption-priced cloud data platform. Chief executive Sridhar Ramaswamy; approximately 9,060 employees; Menlo Park, California; NYSE-listed since September 2020. Fiscal year ends 31 January; fiscal 2026 ended 2026-01-31.

Both blocks sum to reported revenue in every year shown, and the United States home market is present at 77.9%. In a batch where Medtronic's geography block overstated US revenue by $53 billion, S&P Global's product block lost an entire $2.3 billion segment, and Howmet's lost two, Snowflake's disclosure reconciles cleanly and that is worth recording explicitly.

Two structural facts follow. Product revenue is 95.5% of the total and grew 29.2% — this is a consumption business, not a seat-licence business, so revenue tracks how much customers actually run rather than how many people they employ, which is the specific reason the AI argument matters more here than for a per-seat vendor. And the international mix is rising only slowly: EMEA and Asia-Pacific were 20.2% of fiscal 2024 revenue and 22.1% of fiscal 2026, so growth is broadly proportional rather than being carried by geographic expansion.

2. Growth and the margin trajectory — the strong half of the case

Revenue compounded 51.3% a year from fiscal 2021 to fiscal 2026 — from $592 million to $4.68 billion. Growth decelerated from triple digits to roughly 29% and then STABILISED there for two years, which at this scale is a good outcome rather than a warning.

And the quarterly series shows re-acceleration:

Two things in that table are genuinely impressive. First, growth ACCELERATED from 25.7% to 33.5% across five quarters at a $5 billion run rate — the reverse of what almost every software company at this size does. Second, the GAAP operating margin improved from minus 42.9% to minus 23.4% — 1,950 basis points in a year — while that acceleration happened. Growth and margin improving together is the hardest combination in software.

Third, and this is the counterweight: stock-based compensation was 34.1% of revenue in fiscal 2026. It has fallen from 41.7% in fiscal 2023, which is progress — but $1,599.5 million of stock compensation against $1,120.3 million of free cash flow means the company paid its people more in shares than it generated in cash, and the share count rose from 337.5 million to 346.6 million despite $873.5 million of buybacks. Every non-GAAP figure in Section 4 excludes that cost.

3. Balance sheet, cash flow and the enterprise-value rebuild

From bal_a at 31 January 2026:

The net-cash position, corrected. bal_a.netDebt reads minus $87.0 million — total debt of $2,741.1M less cash and equivalents of $2,828.2M only. It omits $1,201.5 million of short-term investments and $755.0 million of long-term investments. The correct figure is net cash of approximately $2,043.6 million, or $2,455.3 million if lease obligations are excluded from debt as they properly should be for a net-cash calculation. This is the PDD/ISRG omission pattern named in the data contract.

The vendor's enterprise value implies something different again. km_ttm.enterpriseValueTTM of $110,479,862,000 against a market capitalisation of $109,792,482,000 implies net debt of $687.4 million — presumably from the April 2026 quarter, and unverifiable here. We use the corrected annual figure: enterprise value of approximately $107.75 billion, giving EV/trailing sales of 21.4x against the vendor's 21.95x. The correction is small in relative terms because the equity value swamps everything, but it is reported.

Note the equity trajectory, which is the dilution story in balance-sheet form. Total equity has fallen from $5,190.6 million (FY2024) to $1,924.1 million (FY2026) — a 63% decline — while the accumulated deficit grew from $4,075.6M to $9,494.2M. priceToBookRatioTTM of 56.4x and returnOnInvestedCapitalTTM of minus 27.2% are the arithmetic consequences and neither is a useful lens; both are recorded because a screen would surface them.

Cash flow:

Free cash flow has compounded 31.1% a year since fiscal 2023 at a consistent 24-28% margin, which is genuinely good. And in every single year, stock-based compensation has EXCEEDED free cash flow — by $365M, $389M, $566M and $479M respectively. The share count has risen in every year despite $3.4 billion of cumulative buybacks, and stood at approximately 346.6 million at the current market capitalisation. That is the arithmetic definition of growth funded by dilution, and it is why the 164x forward non-GAAP multiple in Section 4 must be read with care.

4. Valuation — priced in or room?

At $316.77 (market cap $109.79B, ~346.6M shares, corrected enterprise value ~$107.75B):

Read the P/E row across. 164x on the year ending January 2027. 116x on 2028. 83x on 2029. And 53x on the year ending January 2031 — five years out, on THREE analysts. There is no year in the entire estimate block at which this stock trades below 50 times non-GAAP earnings.

And the non-GAAP earnings exclude stock compensation of 34.1% of revenue. On the fiscal 2027 consensus revenue of $6,098.4M, stock compensation at even 30% of revenue would be roughly $1.83 billion, against consensus net income of $736.2 million. The cost excluded from the denominator is roughly two and a half times the profit in it.

Estimate coverage is excellent in the near years and collapses after fiscal 2028: 36 analysts on fiscal 2027 and 2028 revenue and 27 on both EPS lines — the best-covered pair in this batch — falling to 25 and 7 for fiscal 2029, 12 and 4 for 2030 and 11 and 3 for 2031. The fiscal 2029, 2030 and 2031 EPS rows are used only for the multiple-progression illustration above and for no conclusion.

est.ebitdaAvg and est.ebitAvg are fixed-ratio fabrications AND internally impossible — REJECTED in full. ebitdaAvg is exactly 3.63% of revenueAvg in every year from fiscal 2027 to fiscal 2031; ebitAvg is exactly minus 3.50%. Worse, ebitAvg is NEGATIVE in every single forward year while netIncomeAvg is strongly POSITIVE — fiscal 2027 shows EBIT of minus $213.5 million against net income of plus $736.2 million. A company cannot earn $736 million after interest and tax on an operating loss of $214 million. And the fiscal 2024 row compounds it: epsAvg of PLUS $0.799 against netIncomeAvg of MINUS $1,017.5 million — the sign is inconsistent within a single row. Every non-EPS field in this estimate block is unusable.

Peer context. The vendor peer set contains genuine adjacencies — Datadog and Cloudflare are the closest in business model (consumption-priced infrastructure software) — alongside Workday, Autodesk, Cadence, Synopsys, MicroStrategy, CoreWeave, Motorola Solutions and Corning. No peer multiples are supplied in the file, so no peer-multiple comparison is drawn. The single most relevant comparator, Databricks, is PRIVATE and therefore has no observable multiple at all — a material gap in any valuation of this name, and one this dive states rather than works around. Note also that ServiceNow (NOW) is in this batch and appears alongside Snowflake in three separate knowledge-base claims.

4a. What today's price assumes (the inversion)

At $316.77 — 164.0x FY2027 consensus non-GAAP EPS and 21.8x trailing sales — the price embeds:

• Revenue reaches $6.10 billion in the year to January 2027, $7.67 billion in 2028 and $9.63 billion in 2029 — +30.2%, +25.7%, +25.6% (consensus; 36, 36 and 25 analysts on revenue). Revenue more than doubling in three years from a $4.68 billion base.
• The GAAP operating margin continues improving by roughly 1,000-2,000 basis points a year toward breakeven. (Our number, derived from the consensus non-GAAP path.) It went from −42.9% to −23.4% in the last year. This is the most fragile assumption in the price, because the improvement so far has come substantially from scale against a roughly fixed cost base, and the remaining 2,340 basis points to breakeven must come from somewhere harder.
• Stock-based compensation keeps falling as a share of revenue and eventually stops diluting. (Filing-confirmed at 34.1%, down from 41.7% in fiscal 2023.) At the current rate the share count still rises. For the 164x multiple to mean anything to a shareholder, that has to stop.
• The market keeps paying more than 50x non-GAAP earnings for the next five years. (Our number.) There is no year in the estimate block below 53x. At 60x fiscal 2029E the stock is $228; at 100x it is $380.
• Snowflake, not its private competitor, captures the AI-driven expansion in queryable enterprise data. (Our inference from the knowledge base.) A named-speaker claim argues the opposite.

4b. The return bridge (why the multiple moves)

Expected return over the next twelve months decomposes as: EPS growth (+41.1%, from FY2027E $1.932 to FY2028E $2.726) + multiple drift (COMPRESSION, from 164.0x to roughly 110x on the respective forward year, −33%) + shareholder yield (NEGATIVE, because the share count rises) ≈ −5% to −10%.

A distinction the format guardrails require and which matters enormously at this multiple. At a constant price, 164x fiscal 2027 becomes 116x fiscal 2028 automatically as earnings compound. That is mechanical rolldown and is NOT re-rating, and we do not count it. What we DO assume is that the market pays roughly 110x on the then-forward year against 164x today — a genuine de-rating of about a third — because a multiple at that level has historically not been sustained through a growth deceleration from 30% to 25%.

The shareholder-yield leg is NEGATIVE and that is unusual enough to isolate. In every year of this file the share count has risen despite buybacks, because stock compensation exceeds free cash flow. A holder is diluted roughly 2-3% a year while waiting.

If the multiple simply HELD at 164x on FY2028E the price would be $447 (+41.1%); at 90x it would be $245 (−22.6%). That 64-point spread on a multiple assumption, against a base case of −14.8%, is the entire argument, and it is not resolvable by analysing the company.

4c. Variant perception (where we differ, what would surprise)

• We agree with the street on level, and the street disagrees with the market — that is the finding. Consensus is $285.20, 10.0% BELOW the current price, with a median of $289 also below. Our base is $270. When 52 analysts, 42 of whom rate the stock a buy, collectively target a price 10% under where it trades, the market has moved past its own analysts. We have no variant perception on value; we have agreement with the crowd that the price has overshot.
• We differ from the market on what the "earnings" in the multiple are. Stock-based compensation of $1,599.5 million is 34.1% of revenue and exceeds free cash flow of $1,120.3 million. The consensus non-GAAP EPS of $1.932 excludes it entirely. A 164x multiple on a number that omits a cost larger than the company's entire cash generation is not the same thing as a 164x multiple, and the share count rising from 337.5 million to 346.6 million despite $873.5 million of buybacks is the proof that the cost is real.
• We think the knowledge-base lane on this name is substantially contaminated, and we say so rather than using it. Of 30 claims, 8 come from a chip vendor whose products the claims themselves describe as accelerating Snowflake's stack, and 9 more — a five-claim cluster and a four-claim cluster, each from a single date — discuss Snowflake Intelligence, partnership terms and internal roadmap in a register no outside analyst uses. We assess those nine as probable company voice and half-weight them with the suspicion disclosed. That leaves roughly nine independent claims and they are split.
• The sharpest independent claim is bearish and specific, and it deserves to be the counterweight. Yan Sua (business_breakdowns, 2023-11-07): "AI requires unstructured-data analytics, Databricks' strongest muscle and not Snowflake's world; Snowflake diverges into commoditized structured infrastructure, and even Snowflake customers will have to adopt Databricks." That claim is nearly three years old and Snowflake's revenue has since roughly doubled — so it has not been right yet. It has also not been refuted, because the competitor is private and there is no public number to test it against.
• Positive surprise that would force a re-rate: revenue growth above 33% again on 2026-08-26; the GAAP operating margin above minus 18%; stock compensation below 30% of revenue; the share count falling; or any disclosure that materially quantifies the AI-driven consumption uplift.
• Negative surprise that would break the thesis: revenue growth below 27%; the margin improvement stalling; stock compensation rising back above 36%; the private competitor filing to list, which would give the market a direct comparison; further insider selling after the roughly $23 million already executed; or simply a sixth consecutive beat that fails to hold the price.

Synthos fair values

All three anchors are multiples of the fiscal 2028 (year ending January 2028) consensus non-GAAP EPS distribution — mean $2.726, low $2.432, high $3.596 (27 analysts) — the best-covered forward EPS line in the file. Cross-checked against enterprise-value-to-sales on the same year.

• Bear ~$158 — 65x the fiscal 2028 consensus LOW of $2.432. Cross-check: 6.9x FY2028E sales; 11% below the street's low target of $177 and 30% above the 52-week low of $121.11. The scenario: revenue growth decelerates below 25%; the margin improvement stalls short of breakeven; stock compensation stays above a third of revenue and dilution continues; the private competitor is shown to be winning the unstructured-data workload; and a 25%-growing, loss-making platform earns a mid-single-digit sales multiple. −50.1%.
• Base ~$270 — 99x the fiscal 2028 consensus MEAN of $2.726. Cross-check: 11.9x FY2028E enterprise-value-to-sales; 72.4x FY2029E. Sensitivity, stated openly: 85x gives $232 and 115x gives $313 — essentially the current price. The whole answer lives in an 85-115x band, and the market is at the top of it. The scenario: revenue and margin land on consensus, the AI-driven data expansion is real but shared with the competitor, stock compensation grinds down toward 30%, and the multiple de-rates by roughly a third as growth passes from 30% to 25%. −14.8%.
• Bull ~$465 — 129x the fiscal 2028 consensus HIGH of $3.596. Cross-check: 20.6x FY2028E sales; 46.8% above the current price, 36.8% above the street's high target of $340, and roughly 15% above the six-year peak. The scenario: the document-and-unstructured-data unlock is as large as the conviction-85 claim suggests and Snowflake captures its share; revenue growth reaccelerates above 33% and stays there; GAAP breakeven arrives inside two years; stock compensation normalises and the share count starts falling; and the market pays a full platform multiple. +46.8%.

Base is 14.8% BELOW spot; asymmetry roughly 0.93:1 — more downside than upside on our own numbers, on a stock whose consensus target is also below the price. That is a Watch by arithmetic and it is not a close call, notwithstanding that the business itself is one of the two or three best in this batch.

5. Knowledge base — 30 claims, 17 of them conflicted or suspected company voice

Raw entity hits: 30 — the second-largest lane in this batch. Excluded as conflicted partner voice: 8. Half-weighted as SUSPECTED COMPANY VOICE: 9. Genuinely independent: approximately 9, and split.

The two exclusions, handled first

Eight of the thirty claims come from the jensen_huang channel, and the claims themselves establish the conflict: they describe the enterprise data platforms as "NVIDIA-accelerated", name cuDF and cuVS — NVIDIA's own data-processing libraries — and argue that "GPU-accelerating structured (cuDF) and unstructured/vector (cuVS) data processing is a huge emerging market", citing a customer example 5x faster at 83% lower cost. A chip vendor's chief executive arguing that data platforms accelerated by his chips are the future is not an independent voice on those platforms. All eight are bullish at convictions of 75 to 88 and all make the same argument. Excluded from the conviction pool, half-weighted, disclosed — the same treatment applied to the same channel in the ServiceNow dive in this batch, where it supplied 29% of a 42-claim lane.

A further NINE claims are strongly suspected company voice, and this is the more consequential finding. A five-claim no_priors cluster dated 2025-11-06 discusses "Snowflake Intelligence" by name, the terms of the Microsoft and Azure partnership and "pursuit of similar GCP arrangement", the SAP data-sharing arrangement, and asserts that a data-first platform "has durable right to be as large as a CSP". A four-claim cognitive_revolution cluster dated 2026-01-14 discusses document-extraction product capability, retrieval-augmented-generation tuning and open semantic interchange naming Snowflake, Tableau and Omni. Both discuss internal roadmap and partnership terms in a register no outside analyst uses, and both are single-date clusters — the signature of one interview. We assess them as probable company voice, half-weight them, and disclose the suspicion rather than either ignoring it or asserting it as fact. This is the LMT case — where two suspected-management claims flipped a lane from 5-3-1 to 3-3-1 — occurring at four and a half times the scale.

CONCENTRATION SENSITIVITY TEST, and it is decisive. Removing the eight conflicted-partner claims and half-weighting the nine suspected-company claims takes a lane that reads roughly 26 bullish, 2 bearish, 2 neutral at face value down to approximately nine independent claims splitting about 5 bullish, 2 bearish, 2 neutral. The apparent overwhelming bullishness of this lane is, on inspection, a chip vendor and probably the company itself.

The independent claims

Bearish, and the sharpest in the lane:

> 2023-11-07 · BEARISH · conviction 55 · entity: SNOW · speaker: Yan Sua · channel: business_breakdowns

> "AI requires unstructured-data analytics, Databricks' strongest muscle and not Snowflake's world; Snowflake diverges into commoditized structured infrastructure, and even Snowflake customers will have to adopt Databricks."

> 2026-03-27 · bearish · conviction 65 · entities: Snowflake, ServiceNow, Atlassian, Workday · channel: all_in

> "SaaS names are being re-rated hard: market-cap/free-cash-flow payback periods have roughly halved as market prices in AI fragility of their cash flows."

Bullish:

> 2026-02-07 · bullish · conviction 82 · entities: Databricks, Snowflake, ClickHouse · channel: all_in

> "Data/transformation platforms are true AI beneficiaries — reaccelerating growth (Databricks >60% at scale) because agentic AI depends on their data layers."

> 2026-06-08 · bullish · conviction 75 · entities: SNOW, MDB, ORCL · speaker: Nikesh Arora · channel: all_in · MANAGEMENT VOICE — HALF WEIGHT

> "Infrastructure software (Databricks, Snowflake, MongoDB, Oracle) is undervalued — enterprises will need 10x the data stored in 3 years, so core storage/data infra is needed."

> Disclosure: the speaker is the chief executive of Palo Alto Networks. He is not a Snowflake competitor, but he is a software chief executive talking his sector, and the house rule half-weights management voice regardless.

> 2024-01-25 · bullish · conviction 70 · entities: MSFT, AMZN, SNOW · channel: bill_gurley

> "Forecast that Microsoft, Amazon and Snowflake all accelerate their growth rates this year — accelerating growth at that scale is 'unheard of.'"

> This proved correct: Snowflake's growth went from 25.7% to 33.5% across the last five quarters. A dated, falsifiable forecast that came true is weighted accordingly.

> 2023-07-25 · bullish · conviction 66 · entities: Snowflake, Databricks · channel: chinatalk

> "Both abstract cloud data complexity — Snowflake for analytics/past insights (shorter-term), Databricks for data science/prediction (longer-term); both shepherd ML deployment."

> Read against the Yan Sua claim, this is the same structural observation with a neutral rather than a negative valence: the two platforms serve different halves of the problem, and AI sits closer to the prediction half.

And the one that matters most for the exponential score, from the suspected-company cluster and therefore half-weighted:

> 2026-01-14 · bullish · conviction 85 · entity: Snowflake · channel: cognitive_revolution · SUSPECTED COMPANY VOICE — HALF WEIGHT

> "80-90% of enterprise data sat trapped in PDFs/documents; AI now makes it extractable and queryable for the first time — a massive unlock."

If that claim is right it is the single most important fact about this company — a consumption-priced platform whose addressable data volume multiplies rather than whose share grows. It is also the claim we trust least, because we assess its source as probably the company.

Text-sweep discards include several matched only on Databricks — a private competitor — and one on an Iranian-diaspora economic claim that names Databricks among companies founded by émigrés, plus a 2022 arthur_hayes claim using Snowflake's volatility as an analogy for crypto. None is a view on the equity.

Conclusion on the lane

Conviction: MIXED-CONFLICTED. Breadth 8 channels. Claim count 30 raw, roughly 9 independent. At face value this lane reads as one of the most bullish in the batch; after excluding a commercially interested chip vendor and half-weighting two probable company interviews, it reads as split with a specific, named, unrefuted bear thesis about competitive positioning in exactly the workload that is growing fastest. The lane raises our understanding of the argument and lowers our confidence in the consensus, which is the opposite of what a 26-bullish-claim lane appears to do.

6. Data integrity — what we rejected and why

Six findings, and the file is cleaner than most in this batch.

1. est.ebitAvg is NEGATIVE in every forward year while est.netIncomeAvg is strongly POSITIVE — REJECTED. Fiscal 2027: EBIT minus $213.5 million against net income plus $736.2 million. The same pattern holds for every year to fiscal 2031. A company cannot earn $736 million after interest and tax on a $214 million operating loss. And the fiscal 2024 row compounds it: epsAvg of PLUS $0.799 against netIncomeAvg of MINUS $1,017.5 million — the sign is inconsistent within one row. Every non-EPS field in the estimate block is unusable.

2. est.ebitdaAvg and est.ebitAvg are fixed-ratio fabrications — REJECTED. ebitdaAvg is exactly 3.63% of revenueAvg and ebitAvg exactly minus 3.50% in every year fiscal 2027 to fiscal 2031. Snowflake's actual fiscal 2026 EBITDA margin was minus 23.1%.

3. bal_a.netDebt omits $1,956.5 million of investments — corrected. The vendor reports net debt of minus $87.0 million (total debt $2,741.1M less cash and equivalents $2,828.2M only), ignoring $1,201.5 million of short-term and $755.0 million of long-term investments. Corrected net cash is approximately $2,043.6 million, or $2,455.3 million excluding lease obligations from debt. This is the PDD/ISRG pattern. km_ttm.enterpriseValueTTM of $110,479.9M implies net debt of $687.4M, presumably from the April 2026 quarter and unverifiable here; we use the corrected annual figure and report enterprise value at approximately $107.75 billion, giving EV/sales of 21.4x against the vendor's 21.95x.

4. est EPS coverage collapses after fiscal 2028 and the illustrative multiples reflect it. 36 analysts on fiscal 2027 and 2028 revenue and 27 on both EPS lines — the best-covered pair in this batch — falling to 7 on fiscal 2029 EPS, 4 on 2030 and 3 on 2031. The 53.1x fiscal 2031 multiple quoted in Section 4 rests on three analysts and is used to illustrate the shape of the multiple progression, not as a valuation input. Note also the fiscal 2029 EPS dispersion: low $1.911, high $5.415 — a 2.83:1 range, the second-widest in this batch after ServiceNow's.

5. priceToBookRatioTTM of 56.4x and returnOnInvestedCapitalTTM of −27.2% are arithmetically correct and analytically useless — reported, not used. Book equity has fallen from $5,190.6M to $1,924.1M in two years while the accumulated deficit grew to $9,494.2M. Neither figure tells a reader anything about the business; both are recorded because a screen would surface them.

6. tech.pct_from_hi is 0.00% while max_dd_from_peak is −21.2% — correctly different, not a defect. The stock is at its highest close of the past 252 days, but the six-year peak (from late 2021) is 21.2% higher. Both figures are right and their divergence is informative: this is a stock making a 52-week high inside a five-year drawdown. quote.yearHigh of $321.31 is the intraday figure against tech.hi52 of $316.77, the closing figure; we use tech throughout.

Not defects, correctly reported and independently confirmed — and this file is notably clean: seg_prod and seg_geo BOTH sum to reported revenue exactly in every year shown, with the 77.9% United States home market present — in a batch where three separate names had segment blocks missing entire divisions or overstating a region by $53 billion, Snowflake's disclosure reconciles without adjustment; inc_q April-quarter revenue of $1,390.951M matches the earn_cal actual exactly; dividendPerShareTTM of zero is correct; stockBasedCompensationToRevenueTTM of 32.24% reconciles to the cash-flow statement; profile.currency and inc_a[0].reportedCurrency are both USD — no currency mixing; and earn_cal uses a consistent non-GAAP basis across all five reported rows, so the five-beat series is like-for-like.

8-K scan for corporate actions — completed, and it found a compensation event rather than a transaction. The three retained 8-Ks are 2026-05-27 (Item 2.02, first-quarter results), 2026-06-30 (Item 5.07, annual meeting voting results) and 2026-07-16 (Item 5.02), the last disclosing that on 15 July 2026 the Compensation Committee granted a performance-based restricted stock unit award to chief executive Sridhar Ramaswamy, structured "over an extended performance period and with stock price milestones that require material long-term stockholder value creation in order to be earned", citing "a highly competitive environment for AI talent". No merger, acquisition, divestiture or financing event appears. Verified clean on transactions — but a further large equity grant at a company already spending 34.1% of revenue on stock compensation is worth recording rather than passing over.

Non-equity tripwire — checked and passed. SNOW is Class A common stock, NYSE-listed. The $316.77 price is not par-like; beta is 1.353, the highest in this batch; there is no dividend at all; volume was 5.47M shares (~$1.73B of turnover); and the 52-week band of $121.11 to $316.77 is a 162% range. This is common equity.

A note on the superseded file. A dive for this ticker and date already existed on disk and has been overwritten. It carried a verdict: field containing several sentences of rationale rather than one of the five permitted values — the exact front-matter breach the contract warns hard-fails or mis-tiers the renderer. This version uses verdict: Watch with all rationale in the body.

7. Technicals

• Price $316.77, up 3.00%, and it is the highest close in the six-year series over the past year: tech.hi52 equals tech.last and pct_from_hi is 0.00%. max_dd_from_peak is −21.2%, so the stock is making a 52-week high while still 21% below its 2021 peak.
• +161.6% above the 52-week low of $121.11. Position within the annual range: 100th percentile.
• Moving averages, both far below and both rising: +25.0% above a 50-day of $253.40; +50.0% above a 200-day of $211.23.
• RSI 76.7 — the highest reading in this batch and clearly overbought. MACD +15.21 — the highest absolute reading here by a factor of nearly four.
• Relative performance, and it is extraordinary: 3-month +119.7% vs SPY +7.6% and QQQ +7.7% — a 112-point outperformance in one quarter, more than four times the next-largest three-month move in this batch (Fortinet's +88.6%); 6-month +66.1% vs SPY +11.1%; 12-month +54.5% vs SPY +24.3%.
• Sentiment, and it is inverted: 42 buy, 9 hold, 1 sell across 52 analysts — overwhelmingly positive. Consensus target $285.20 — 10.0% BELOW spot. Median $289 — 8.8% below. Only the high target of $340 is above the price. Low $177 (−44.1%).

What the technical picture does to the entry. A stock at the 100th percentile of its annual range, 25% above its 50-day average, with RSI at 76.7, after a 119.7% three-month move, trading above the average of 52 analyst targets, with insiders selling and results in 22 days. Every one of those facts individually argues for waiting; together they are close to definitive. The first support on the chart is the 50-day average, 20.0% below at $253.40; the second is the 200-day, 33.3% below at $211.23. This is not a judgement on the business — it is an observation that the security has run a very long way in a very short time and the people who follow it most closely think it has run past them.

8. Insiders — roughly $23 million sold into the run, days before this dive

Two insiders executed roughly $23.0 million of DISCRETIONARY open-market sales between 29 and 31 July 2026, four to six days before this dive. Christian Kleinerman, executive vice president of product management, sold 29,986 shares at $300.00 across three transactions ($9.0 million). Benoit Dageville, a director and one of Snowflake's co-founders, sold 50,000 shares at $280.00 ($14.0 million) and gifted a further 16,668.

The reading, stated carefully. These are S-Sale transactions — decisions to sell — not the F-InKind tax withholding that dominates most insider blocks in this batch. The file does not indicate whether they were executed under Rule 10b5-1 plans, and executives at this level routinely sell for diversification; a co-founder director monetising a portion of a founding stake after a 119.7% three-month move is entirely normal behaviour. What makes it worth recording rather than dismissing is the coincidence of price and timing: $280 and $300 are 11.6% and 5.3% below where the stock closed on the day of this dive, and both sales landed in the same three-day window at the top of the largest move in the batch.

Separately, Frank Slootman — a director and the former chief executive — exercised 589,685 options at $8.88 on 20-21 July. The records available show the exercises (M-Exempt) but no corresponding sale, so no inference about disposal is available. At a $316.77 share price those options carry an intrinsic value of roughly $181 million.

The file contains zero open-market purchases. Net reading: a mild negative, sized modestly, and specifically noted because it is the only batch name where discretionary insider selling coincided with an all-time-range high in the week before the dive.

9. Verdict, kill-criteria and flip conditions

Watch.

What is genuinely excellent, and it should be said clearly: revenue compounding 51.3% a year over five years to $4.68 billion and still accelerating — 33.5% in the April 2026 quarter at a $5 billion run rate, which almost no software company achieves at this size; a GAAP operating margin improving 1,950 basis points in a year, from minus 42.9% to minus 23.4%; five consecutive earnings beats of 13.0%, 31.1%, 12.3%, 17.4% and 22.1% — the largest and most consistent in this batch; free cash flow of $1,120.3 million compounding 31.1% a year at a 24% margin; roughly $2.04 billion of net cash; a 67.2% gross margin; 42 buy ratings; and segment and geographic disclosure that reconciles exactly in every year, which in this batch is genuinely unusual.

What we are declining to pay for: 21.8x TRAILING sales, 164.0x the fiscal 2027 consensus, 116.2x fiscal 2028, and 53.1x even fiscal 2031 — five years out; a 1.07% free-cash-flow yield, the lowest here; stock-based compensation of $1,599.5 million at 34.1% of revenue, exceeding the entire free cash flow, with the share count RISING despite $873.5 million of buybacks — meaning the non-GAAP earnings the multiple is applied to are not earnings a shareholder receives; a GAAP loss of $3.95 per share and an accumulated deficit of $9,494.2 million; a stock at the 100th percentile of its annual range with RSI at 76.7 after a 119.7% three-month move; roughly $23 million of discretionary insider selling at $280-$300 in the week before this dive; a consensus target 10.0% BELOW the price and a median 8.8% below; a base fair value 14.8% below spot; and a knowledge-base lane whose apparent overwhelming bullishness is, on inspection, a commercially interested chip vendor and two probable company interviews.

The distinction that matters. This is one of the two or three best businesses in this batch and it is priced as such and then some. Our objection is not to Snowflake; it is that a security which has more than doubled in a quarter, trades above the average target of all 52 analysts who cover it, carries a 1.07% cash yield on earnings that exclude a third of revenue, and has just seen two insiders sell $23 million into the move, has no margin for anything going wrong — and reports in 22 days. When our number, the crowd's number and the median target all sit below the price, waiting costs nothing and buys a great deal.

Pre-registered KILL criteria — what would take this to Avoid:

• Revenue growth falling below 27% in the 2026-08-26 print or thereafter, which at 116x fiscal 2028 earnings would be repriced violently.
• The GAAP operating margin stalling above minus 25%, ending the improvement trajectory the entire earnings path requires.
• Stock-based compensation rising back above 36% of revenue, or the share count accelerating upward.
• Any evidence that the private competitor is winning the unstructured and vector workload — the specific claim a named speaker made in 2023 and which nothing in this file refutes.
• A sixth consecutive beat that fails to hold the price, which after a 119.7% three-month move would confirm the news is fully discounted.
• A break below the 50-day average at $253.40 on volume.

Pre-registered FLIP conditions — what would take this to Buy — Tactical:

• Price below approximately $200 (73x FY2028E, 8.5x FY2028E sales) with revenue growth still above 28%. At $200 the base case is +35% against a bear case only 21% lower, and the arithmetic becomes attractive rather than adverse. That is 36.9% below today's close and below the 200-day average.
• Revenue growth above 33% again on 2026-08-26 combined with a GAAP operating margin above minus 18%, which would put GAAP breakeven inside two years and change the earnings basis of the whole valuation.
• Stock-based compensation falling below 30% of revenue with a FALLING share count — the single change that would make the non-GAAP multiple meaningful to a shareholder.
• A material disclosure quantifying AI-driven consumption uplift, which would let the market value the data-volume expansion the knowledge base describes rather than infer it.
• Sell-side targets moving decisively above the price, resolving the current inversion in the direction of the fundamentals rather than the price.

Where SNOW fits in the Synthos Framework Portfolio. The data-infrastructure sleeve, at 0% today with a 2% target on a fill near $200 or on GAAP margin clearing minus 18%. Batch overlap note, and it is material: ServiceNow (NOW) is in this batch and appears alongside Snowflake in three separate knowledge-base claims — including a bearish claim at conviction 80 naming CRM, NOW, SNOW and ADBE together, and an all_in claim at conviction 65 grouping Snowflake with ServiceNow, Atlassian and Workday as SaaS names being 're-rated hard'. These are two expressions of one debate — whether AI expands or destroys enterprise-software value — and holding both is concentration dressed as diversification. Of the two, ServiceNow has the stronger contracted backlog ($29.0 billion of remaining performance obligations, 98% renewal) and a 45.0x forward multiple; Snowflake has the faster growth (33.5% against 24.0%) and a 164x forward multiple. Logged as a tracked Synthos call (Watch) as of 2026-08-04 at $316.77.

Single biggest risk: the multiple, and the fact that the earnings it is applied to exclude a third of revenue. At $316.77 the market pays 164 times the fiscal 2027 consensus non-GAAP earnings per share of $1.932. That $1.932 excludes stock-based compensation, which was $1,599.5 million in fiscal 2026 — 34.1% of revenue, more than the $1,120.3 million of free cash flow the company generated, and enough that the share count rose from 337.5 million to 346.6 million despite $873.5 million of buybacks. On the fiscal 2027 consensus revenue of $6.1 billion, stock compensation at even 30% of revenue would be roughly $1.83 billion against consensus net income of $736 million — the excluded cost is two and a half times the reported profit. A shareholder buying at 164x is therefore buying a multiple of a number that overstates what accrues to them, on a security at the 100th percentile of its annual range, above every analyst target, after a 119.7% three-month move, days after two insiders sold $23 million into it, and 22 days before a print. The business may well be worth owning. At this price, on this measure, the arithmetic says wait — and our own base of $270, the consensus $285.20 and the median $289 all agree.

Provenance & disclosures

• Traceability: 30 raw knowledge-base entity claims name Snowflake — the second-largest lane in this batch after ServiceNow's 45. EIGHT (27%) come from the jensen_huang channel — the chief executive of NVIDIA — and the claims themselves establish the conflict, describing the platforms as "NVIDIA-accelerated" and naming NVIDIA's cuDF and cuVS data-processing libraries as the mechanism. All eight are bullish (conviction 75-88), all make the same argument, and all are EXCLUDED from the conviction pool and half-weighted. A further NINE are STRONGLY SUSPECTED COMPANY VOICE and are half-weighted with the suspicion disclosed: a five-claim no_priors cluster dated 2025-11-06 discussing "Snowflake Intelligence", the Microsoft/Azure partnership terms and the pursuit of a similar Google Cloud arrangement, the SAP data-sharing arrangement, and asserting a "durable right to be as large as a CSP"; and a four-claim cognitive_revolution cluster dated 2026-01-14 on document extraction, retrieval-augmented-generation tuning and open semantic interchange. Both are single-date clusters discussing internal roadmap and partnership terms in a register no outside analyst uses. This is the LMT suspected-management-voice case at four and a half times the scale, and it is flagged rather than resolved. A CONCENTRATION SENSITIVITY TEST was run and is decisive: at face value the lane reads roughly 26 bullish, 2 bearish, 2 neutral; after the exclusions it is approximately nine independent claims splitting about 5 bullish, 2 bearish, 2 neutral. The independent set includes the sharpest bear — Yan Sua (business_breakdowns, 2023-11-07, named speaker, bearish 55): "AI requires unstructured-data analytics, Databricks' strongest muscle and not Snowflake's world; Snowflake diverges into commoditized structured infrastructure" — plus an all_in claim (2026-03-27, bearish 65) on SaaS free-cash-flow payback periods halving; and on the bullish side an all_in claim (2026-02-07, conviction 82) that data platforms are "true AI beneficiaries", a bill_gurley forecast (2024-01-25, conviction 70) of accelerating growth which proved correct, a chinatalk structural claim (2023-07-25, conviction 66), and Nikesh Arora (all_in, 2026-06-08, conviction 75) — management voice of another software company, half-weighted — that infrastructure software is undervalued because enterprises "will need 10x the data stored in 3 years". Text-sweep discards include claims matched only on the private competitor Databricks, an Iranian-diaspora economic claim, and a 2022 claim using Snowflake's volatility as a crypto analogy. AI model names appearing in one quoted claim's entity list are redacted per house rules. All quotes are verbatim from the stored claim text apart from that redaction.
• Data as-of: inc_q and the earnings calendar run to the quarter ended 2026-04-30 (first quarter of fiscal 2027); annual statements fiscal 2026 ended 2026-01-31 from inc_a/bal_a/cf_a, corroborated by the 10-K filed 2026-03-20 · estimates 2026-08-04 · prices 2026-08-04, quote timestamp 1785873602 = 2026-08-04T20:00:02Z ($316.77, +3.00%; 50-DMA $253.40; 200-DMA $211.23; RSI 76.7; MACD +15.21; beta 1.353) · knowledge-base claims 2026-08-04 · insider records to 2026-07-31. Fiscal note: Snowflake's fiscal year ends 31 January. "Fiscal 2026" means the year ended 2026-01-31, and the first quarter of fiscal 2027 ended 2026-04-30. All figures come from the Synthos vendor data file for SNOW or from the SEC filings in the SNOW archive; no figure comes from memory, recall or external retrieval.
• Filing archive contents: 10-K filed 2026-03-20 (fiscal 2026); 10-Q filed 2025-12-05 (October 2025 quarter); 10-Q filed 2026-05-29 (April 2026 quarter); 8-K filed 2026-05-27 (Item 2.02, first-quarter fiscal 2027 results — cover document with Exhibit 99.1 furnished but not extracted); 8-K filed 2026-06-30 (Item 5.07, annual meeting voting results); 8-K filed 2026-07-16 (Item 5.02 — a performance-based restricted stock unit award to chief executive Sridhar Ramaswamy with share-price milestones over an extended performance period, granted 15 July 2026 and citing "a highly competitive environment for AI talent").
• Where the filings contradicted or corrected the vendor (detailed in Section 6): bal_a.netDebt of minus $87.0M corrected to net cash of approximately $2,043.6M after establishing that the vendor nets total debt against cash and equivalents only, omitting $1,201.5M of short-term and $755.0M of long-term investments — the PDD/ISRG pattern — with enterprise value restated from the vendor's $110.48B to approximately $107.75B and EV/sales from 21.95x to 21.4x; est.ebitAvg rejected for being NEGATIVE in every forward year while est.netIncomeAvg is strongly POSITIVE (fiscal 2027: minus $213.5M against plus $736.2M), with the fiscal 2024 row additionally showing epsAvg of PLUS $0.799 against netIncomeAvg of MINUS $1,017.5M — a sign inconsistency within one row; est.ebitdaAvg and est.ebitAvg rejected as fixed-ratio fabrications at exactly 3.63% and minus 3.50% of revenue in every forward year against an actual fiscal 2026 EBITDA margin of minus 23.1%; priceToBookRatioTTM of 56.4x and returnOnInvestedCapitalTTM of minus 27.2% reported but not used; quote.yearHigh rejected in favour of tech.hi52. Where vendor and filing AGREED — and this file is notably clean: seg_prod and seg_geo BOTH sum to reported revenue EXACTLY in every year shown, with the 77.9% United States home market present — in a batch where Medtronic's geography block overstated US revenue by $53 billion, S&P Global's product block omitted an entire $2.3 billion segment and Howmet's omitted two, Snowflake's disclosure reconciles without any adjustment; inc_q April-quarter revenue of $1,390.951M matches the earn_cal actual exactly; stockBasedCompensationToRevenueTTM of 32.24% reconciles to the cash-flow statement; dividendPerShareTTM of zero is correct. The 8-K scan found no merger, acquisition, divestiture or financing event — verified clean on transactions, with the 15 July 2026 chief-executive equity award recorded as a compensation rather than a corporate event. A superseded dive for this ticker and date was overwritten; it carried a verdict: field containing multiple sentences of rationale rather than one of the five permitted values, which is the exact front-matter breach the contract warns against, and that is recorded in Section 6.
• Basis note, and it is the most important caveat on this name: analyst estimates and earn_cal actuals are non-GAAP; reported diluted EPS is GAAP, and the two are not merely different — they have opposite signs. Fiscal 2026 GAAP diluted EPS was MINUS $3.95; the fiscal 2027 consensus non-GAAP EPS is PLUS $1.932. The difference is principally stock-based compensation of $1,599.5 million, which is 34.1% of revenue, EXCEEDS the entire $1,120.3 million of free cash flow, and did not prevent the share count rising from 337.5 million to 346.6 million despite $873.5 million of buybacks. Every forward multiple in this dive — including the headline 164.0x — is computed on a number that excludes a cost roughly two and a half times the reported profit. The earn_cal non-GAAP basis is consistent across all five reported rows, so the beat series (+13.0%, +31.1%, +12.3%, +17.4%, +22.1%) is like-for-like and is the strongest in this batch.
• Estimate coverage: 36 analysts on fiscal 2027 and fiscal 2028 revenue and 27 on both EPS lines — the best-covered pair in this batch — falling to 25 and 7 for fiscal 2029, 12 and 4 for fiscal 2030 and 11 and 3 for fiscal 2031. The fiscal 2029, 2030 and 2031 EPS rows are used only to illustrate the multiple progression (83.4x, 63.7x, 53.1x) and for no conclusion. The fiscal 2029 EPS dispersion — low $1.911, mean $3.796, high $5.415, a 2.83:1 range — is the second-widest in this batch and is the estimate community's own measure of how unresolved the AI-and-data-platform question is. The fair-value anchors rest on the fiscal 2028 row (27 analysts).
• Peer note: the vendor peer set contains genuine adjacencies — Datadog and Cloudflare are the closest in business model (consumption-priced infrastructure software) — alongside Workday, Autodesk, Cadence, Synopsys, MicroStrategy, CoreWeave, Motorola Solutions and Corning. No peer multiples are supplied in the file, so no peer-multiple comparison is drawn. The single most relevant comparator, Databricks, is PRIVATE and therefore has no observable multiple at all — a material gap in any valuation of this name, stated rather than worked around.
• Fair-value caveat: the $158 / $270 / $465 anchors are multiples of the fiscal 2028 (year ending January 2028) consensus non-GAAP EPS distribution — 65x the low of $2.432, 99x the mean of $2.726, and 129x the high of $3.596 — cross-checked against enterprise-value-to-sales on the same year (6.9x, 11.9x and 20.6x). Stated arithmetic, not a discounted cash flow. The base is sensitivity-disclosed: 85x gives $232 and 115x gives $313, which is essentially the current price — the market is at the top of the band. Section 4b distinguishes explicitly between mechanical forward-multiple rolldown (164x becoming 116x at a constant price as earnings compound, which is NOT re-rating) and the genuine de-rating the base case assumes (the market paying roughly 110x on the then-forward year against 164x today). All anchors are struck on non-GAAP earnings that exclude stock compensation exceeding free cash flow, and that limitation is disclosed rather than adjusted for, because no consensus estimate on a GAAP basis exists in this file.
• Timing: first-quarter fiscal 2027 results were released 2026-05-27 and the 10-Q filed 2026-05-29, both more than two months before this dive. The next print is 2026-08-26, 22 days away, with consensus revenue of $1,482.2M and non-GAAP EPS of $0.4467. The verdict is set with that proximity as an explicit input, and the dive states plainly that after a 119.7% three-month move into the print, with the stock above every analyst target and insiders selling into it, a sixth consecutive beat is substantially priced. 2026-08-04 was a strong day for the stock (+3.00%) that took it to the highest close in the file; no company-specific news for that date appears here.
• Accessibility note: no information in this dive is conveyed by colour. All emphasis is carried by bold text, table structure and explicit labelling.
• Not investment advice. Independent research, educational and informational only, never personalised. No recommendation to buy, sell or hold any security is made to any person.
• Version: 2026-08-04-full.