Phase Space AI

Financial Model Notes

Dell Technologies [DELL]

Dell Technologies [DELL] — Financial Model Notes

Scope limitation, stated first. Tier-2 memo. No .xlsx workbook. Every figure was computed in reproducible Python against cached SEC XBRL facts and cached Alpaca bars; this file is the audit trail. No balance check, because there is no forecast balance sheet — tie-out is to the filed statements only.

Scripts (committed under work/): fetch_facts.py, facts.py, dump.py, bs.py, filings.py, pr.py, mentions.py, mult.py, anchor.py, calc.py, calc2.py.

---

1. Tie-out to the filed statements

Q1 FY2027 10-Q (0001571996-26-000030, filed 2026-06-09) and FY2026 10-K (0001571996-26-000008, filed 2026-03-16). All $m.

Cross-checks required by the brief: - Q1 FY2027: 3,438 ÷ 656 = $5.241 vs filed diluted EPS $5.24. Exact. - FY2026: 5,936 ÷ 684 = $8.678 vs filed $8.68. Exact.

---

2. Derived figures, each with its formula

---

3. The AI-server gross margin solve — full working

Two equations, two unknowns. Stated assumption: non-AI gross margin unchanged year-on-year. x = AI-optimized server gross margin, y = non-AI gross margin.

Q1 FY2026 revenue split:  AI 1,882 ;  non-AI 23,378 − 1,882 = 21,496 ;  gross profit 4,937
Q1 FY2027 revenue split:  AI 16,132 ; non-AI 43,842 − 16,132 = 27,710 ; gross profit 7,782

(1)   1,882x  + 21,496y = 4,937
(2)  16,132x  + 27,710y = 7,782

From (1):  y = (4,937 − 1,882x) / 21,496
Sub into (2):  16,132x + (27,710/21,496)(4,937 − 1,882x) = 7,782
               27,710/21,496 = 1.28908
               16,132x + 1.28908(4,937 − 1,882x) = 7,782
               16,132x + 6,364.2 − 2,425.0x       = 7,782
               13,707x                            = 1,417.8
               x = 0.10344   ->  AI gross margin  = 10.34%
               y = (4,937 − 194.7) / 21,496 = 0.22061  ->  non-AI = 22.06%

Verification against the reported figure:

16,132 × 0.10344 + 27,710 × 0.22061 = 1,668.7 + 6,113.3 = 7,782.0   == reported gross profit  ✓

Independent validation via the mix effect:

AI share of revenue, Q1 FY2026 = 1,882 / 23,378  = 8.05%
AI share of revenue, Q1 FY2027 = 16,132 / 43,842 = 36.80%

blended GM at  8.05% mix = 0.0805 × 10.344% + 0.9195 × 22.061% = 21.12%
blended GM at 36.80% mix = 0.3680 × 10.344% + 0.6320 × 22.061% = 17.75%
mix effect                                                      = −337 basis points

Dell reported:                                                    −330 basis points

The solve reproduces Dell's own disclosed gross-margin decline to within 7 basis points, from two independent equations. That is the validation, and it is why the 10.3% figure is used in the memo despite being derived rather than disclosed.

Forward sensitivity: d(blended GM)/d(AI mix) = y − x = 22.061% − 10.344% = 11.72pp, i.e. ~118 basis points of blended gross margin per 10 percentage points of AI mix.

Limitations of the solve, stated: 1. It rests on one assumption — that non-AI gross margin was unchanged year-on-year. If non-AI margin actually rose (Dell says CSG gross margin rose on "disciplined pricing"), the true AI margin is lower than 10.3%; if non-AI fell, the true AI margin is higher. The direction of error is therefore probably toward a lower AI margin, i.e. my figure is conservative against my own conclusion. 2. It is a two-period solve on one quarter pair. A third quarter would over-determine the system and allow a residual test; that is a next-print task. 3. It attributes the whole blended decline to product mix. The filing shows product gross margin fell only 30bp to 13.8%, so part of the blended decline is the services-to-product shift (services revenue fell 1% while product rose 117%) rather than AI economics specifically. Both facts are consistent with a structurally ~10% AI margin, and both are reported in DELL_Research.md §4.

---

4. WACC build

beta          = cov(DELL, SPY) / var(SPY) over 252 sessions                 = 1.95
Ke            = 4.2% + 1.95 × 5.5%                                          = 14.9%
w_e           = 254,493 / (254,493 + 31,161)                                = 89.1%
WACC          = 0.891 × 14.9% + 0.109 × 5.5% × (1 − 0.20)                   = 13.8%
realised vol  = stdev(daily returns) × sqrt(252)                            = 69.7%

Stated assumptions: risk-free 4.2%, ERP 5.5%, pre-tax cost of debt 5.5%, forward tax 20% (against a Q1 FY2027 effective rate of 12.9% that included $0.2bn of discrete SBC benefits — 20% is the normalised, conservative choice).

---

5. Own-multiple history — construction

1,505 daily sessions, 2020-07-27 → 2026-07-28 (Alpaca IEX, split-adjusted). TTM metric as known at each date, lagged to the filing date; fourth quarters inferred as (annual − three tagged quarters). Share count and net debt held at today's verified values.

Same known bias as elsewhere (today's net debt held constant), but for Dell net debt is only 7.1% of EV, so the distortion is materially smaller than for Oracle, and the P/E series — immune to it — agrees at the 96th percentile. The EV/EBIT series begins 2021-03-26 (1,337 sessions) because four quarters of as-known operating income are needed first.

The three-year window is used for the 12-month target anchor and the six-year window is reported alongside, because the six-year window spans pre-AI Dell (a 4–10x P/E business) and is a different company. That choice is the single largest judgement in the target and is stated as such in DELL_Valuation.md §5.

TTM EPS note: the P/E series sums four quarterly diluted EPS figures (12.56 for the current TTM). Summing quarterly EPS does not exactly equal an annual EPS when the share count changes — Dell's count fell 6.8% over five quarters — so the series carries a small upward bias in EPS and therefore a small downward bias in P/E. Disclosed; it does not change a 96th-percentile reading.

---

6. Growth-matched anchor — construction, and the sector override

Source: reports/scan_v3/_scan.json, 3,980 names with status == "OK".

The override. sectors.py line 51 maps (3570, 3579) -> "TECH"; Dell's SIC is 3571. The comparator set is therefore built on economics, not SIC:

The comparison of the last two rows is the quantified defect: the TECH set's median operating margin is 19.6% against the correct set's 6.4% — 3.1x — while the EV/EBIT medians differ by only 10% (30.3x vs 27.4x). The mis-mapping is a margin defect, not a multiple defect.

Market-cap floor of $5bn set explicitly to avoid the logged sub-$100m-shell defect; realised minimum $5.3bn. The set's growth range brackets the required exit-year growth, which is the condition valuation.md imposes.

Base exit multiple 19.9x = the anchor set's p25, which independently equals Dell's own three-year EV/EBIT p75 of 19.5x. Using p25 rather than the set median of 27.4x is the separately argued reason valuation.md requires for a base below an anchor, and the argument is given in DELL_Valuation.md §3.

---

7. What is NOT in this analysis

• No three-statement forecast, therefore no balance check. Deliberate Tier-2 scope.
• No disclosed AI-server gross margin. The 10.3% is derived (§3) with its assumption, its verification and its three limitations stated. Dell does not publish it.
• No AI backlog balance. Dell publishes an orders flow, never a stock. No coverage ratio can be computed and none is estimated.
• No AI-server customer concentration. Not disclosed, and unlike Oracle there is no management statement from which a bound can be derived. None is asserted.
• No ROIC or ROE. Book equity is negative (−$1,404m). Both are undefined or sign-flipped; scored INDETERMINATE, not FAIL, per calibration item D1.
• No consensus estimates. Alpha Vantage quota treated as exhausted; no call spent. Company guidance is used instead and is labelled as guidance everywhere it appears.
• No borrow rate. Not obtainable from any free source here; no rate is invented.
• No correlation matrix against the book. Flagged in DELL_Trade_Construction.md §7, not fabricated.
• No DFS credit-loss or delinquency detail. The financing-receivables allowance and ageing were not extracted; only the $77m trade-receivable allowance is used.
• No purchase-obligation figure. The 10-Q notes "higher purchase obligations" qualitatively; the balance was not extracted and no number is asserted.
• No transcript corpus. Mention-frequency uses 8-K Item 2.02 exhibits, one source across ten quarters. No prepared-remarks versus Q&A split is available, and no finding is read as carrying prepared-remarks weight.