conf 3/10 partial
enthusiasm:24.0 · trend:8 · quantifies:5 · impact:0 · under_radar:5 · credibility:0 · business_impact:8 · disruption:0 · commitment:-4 · confirmation:3
Enthusiasm latest 8 / prev 3 (rising)
The latest call presents AI as a concrete build-out optimization tool through Palantir, with management tying it to real-time data, project management, lead-time reduction and $300 million of identified cost savings. The previous call only links AI indirectly to future demand through AI data centers and hyperscalers, without a company-specific AI operating initiative. Credibility is stronger in Q1 FY2026 because management named the platform, use case and quantified savings, though the savings are described as identified or declared rather than fully realized.
Grounded on actual base — revenue $0.4B · net income $0.1B · net margin 17.3% · diluted EPS 3.9
These are next-fiscal-year annual uplift estimates, not next-quarter numbers.
Aggregate next-FY est. rev uplift: 0.0% · next-FY EPS uplift: % · vs analysts: unclear · priced in: medium · confidence: 3/10
| Claim | Figure | Arithmetic | Next-FY Rev % | Next-FY EPS % |
|---|---|---|---|---|
| ~$300M identified potential cost savings (since late Jan) cost | approximately $300 million | $300M is a PROGRAM/LIFETIME potential capital & manufacturing-cost reduction (AI used to cut centrifuge/HALEU build cost and lead times), not a recurring annual opex line. Mechanical one-time after-tax @21% = 300M*0.79 = $237M; /$91.2M consensus adjusted NI = 260% (vs 305% off GAAP $77.8M) — both breach the ~150% guardrail, an artifact of a thin earnings base ($448.7M rev, <$1B) and a non-recurring/capital item. Topline ~0 (only 'lead times', no volume add). No defensible next-FY P&L phasing disclosed, so annual EPS flow is indeterminate. | 0 | |
| $300M cost savings already declared ('so far') cost | $300 million | Same $300M figure as the 'identified' claim — the declared/realized subset of the SAME pool, not an additional $300M. Excluded from the aggregate to avoid double-counting. Same non-recurring/capital character. | 0 | |
| $5M Q1 prepayment for Palantir agreement other | $5 million | A Q1 prepayment/INVESTMENT outflow to adopt Palantir AI tooling, not a saving or revenue. Magnitude = $5M/$448.7M rev = 1.1%. If expensed, after-tax drag = 5M*0.79 = $3.95M / $91.2M consensus adjusted NI = -4.3% EPS (a near-term cost, the enabler of the savings program, not a benefit). | 0 | -4.3 |
Assumptions: Tax rate 21%. Incremental margin N/A (no AI revenue claim). EPS sized against the consensus adjusted basis: FY25 consensus NI $91.19M / EPS $4.48 (GAAP NI $77.8M would inflate the %, so not used per EARNINGS BASIS rule). $300M treated as a multi-year capital cost & lead-time reduction (centrifuge/HALEU manufacturing), not a recurring P&L opex line — no disclosed next-FY phasing, so its annual earnings flow is not estimable. The two $300M lines are one figure, counted once. LEU-for-AI-datacenter / hyperscaler-offtake comments are forward and unquantified.
Top line: Effectively zero quantifiable topline. Every quantified AI claim is operational/cost-side; the only revenue angle (LEU demand from AI data centers / hyperscaler prepayment-offtake talks) is forward-looking and unanchored, with no figure to size. AI here improves Centrus's own cost/lead-time profile, not its sales — it ranks as an adopter, not a supplier of AI capacity.
Bottom line: The headline $300M 'potential cost savings' dwarfs the $448.7M revenue base and the ~$91M earnings base — mechanically it implies a ~260% one-time after-tax EPS swing, which the thin-base guardrail flags as an artifact, not a transformational annual uplift. The figure is a program/capital cost-and-lead-time reduction spread over the multi-year manufacturing build, with no disclosed conversion to next-FY net income, so a meaningful annual EPS% cannot be derived (est_eps_uplift_pct null). The only clean near-term P&L item is the $5M Palantir prepayment, a ~-4.3% drag — AI is currently a small cost, with the savings still 'potential'.
Consensus already expects earnings to FALL, not rise: FY25->FY26 EPS $4.48->$2.84 (-37%), NI $91.2M->$61.8M (-32%). The $300M is real and management-declared but is a non-recurring capital/manufacturing-program saving (lead-time and build-cost reduction), so it does not map onto the consensus earnings line — it neither clearly reverses the expected decline nor produces an estimable above-consensus annual EPS. Net of the $5M Palantir cost, near-term EPS effect is mildly negative. Revenue base <$1B and the only sizeable figure yields a >150% mechanical % — both flag low reliability for any earnings translation.
partial
Adopted Y's adjusted-base, null-EPS treatment of the non-recurring multi-year $300M program; conservative verdicts on conflicts.
| Field | Opus 4.8 | GPT-5.5 |
|---|---|---|
| Rev uplift % | 0 | 0 |
| EPS uplift % | – | 299.5501 |
| Priced in | medium | low |
| vs analysts | unclear | ahead |
| Confidence | 3 | 4 |
| Top line | Effectively zero quantifiable topline. Every quantified AI claim is operational/cost-side; the only revenue angle (LEU demand from AI data centers / hyperscaler prepayment-offtake talks) is forward-looking and unanchored, with no figure to size. AI here improves Centrus's own cost/lead-time profile, not its sales — it ranks as an adopter, not a supplier of AI capacity. | No quantified adopter revenue uplift. The $300M claims are cost savings, and the supplier-side hyperscaler/offtake demand statements have no disclosed revenue amount, so supplier revenue uplift is null. |
| Bottom line | The headline $300M 'potential cost savings' dwarfs the $448.7M revenue base and the ~$91M earnings base — mechanically it implies a 260% one-time after-tax EPS swing, which the guardrail flags as a thin-base artifact, not a transformational annual uplift. The figure is a program/capital cost-and-lead-time reduction spread over the multi-year manufacturing build, with no disclosed conversion to next-FY net income, so a meaningful annual EPS% cannot be derived (est_eps_uplift_pct null). The only clean near-term P&L item is the $5M Palantir prepayment, a ~-4.3% drag — i.e. AI is currently a small cost, with the savings still 'potential'. | One non-duplicated $300M savings pool equals 66.9% of current revenue and $237.0M after tax, or 304.6% of current net income. Netting the $5M Palantir prepayment as an expensed cost reduces this by 5.1 points to 299.6%, but this EPS percentage is too large to treat as a normal annual run-rate without phasing disclosure. |
| Reasoning | Consensus already expects earnings to FALL, not rise: FY25->FY26 EPS $4.48->$2.84 (-37%), NI $91.2M->$61.8M (-32%). The $300M is real and management-declared but is a non-recurring capital/manufacturing-program saving (lead-time and build-cost reduction), so it does not map onto the consensus earnings line — it neither clearly reverses the expected decline nor produces an estimable above-consensus annual EPS. Net of the $5M Palantir cost, near-term EPS effect is mildly negative. Revenue base <$1B and the only sizeable figure yields a >150% mechanical % — both flag low reliability for any earnings translation. | Consensus 2026 revenue is $461.2M, only +$12.5M or +2.8% vs $448.7M current revenue, while consensus 2026 EPS is $2.84011, down 27.2% from current EPS of $3.90. A one-year $300M pretax savings realization would add $237.0M after tax before Palantir cost, vastly above the consensus net income trajectory of $61.8M for 2026. The math points ahead of consensus, but confidence is limited because LEU has a sub-$1B revenue base and management did not disclose savings timing or P&L recognition. |
Rows highlighted where the two models disagreed.
—/100 (no quantified promises) no-quantified-promises 6 calls reviewed
Across all six calls, AI references are overwhelmingly macro demand context (AI/data-center electricity needs as a tailwind for nuclear fuel), and management explicitly excluded AI/hyperscaler growth from its business case. The only company-level AI adoption is the Q1 FY2026 Palantir partnership (~$300M potential cost savings 'identified'), but it is freshly announced and undated, so no genuine prior quantified AI promise has matured for evaluation.
Est. revisions flat · Fwd P/E 44.4 · EV/Sales 6.8x
AI claim maps to Product, Separative Work Units, Uranium
AI enthusiasm across 6 calls — trend ↗ rising
AI was absent until hyperscaler-demand mentions, then became concrete via Palantir AI tied to build-out optimization and cost savings.
7/10 qualitative impact material medium-term · mixed evidence
Where AI matters: manufacturing build-out cost and lead-time optimization
Centrus is using Palantir's AI platform for a concrete operational problem: integrating systems, managing hundreds of suppliers, reducing lead times, and identifying roughly $300 million of potential build-out savings. This is meaningful for HALEU/enrichment expansion economics, but it is project/capital-efficiency driven rather than a recurring AI-enabled revenue model transformation.
Caveats: $300 million savings are identified/potential and not yet proven in realized earnings or cash flow; Benefits may be multi-year capital-project savings rather than recurring margin expansion; Execution risk remains high given classified/unclassified systems, supplier complexity, and nuclear regulatory constraints; AI data-center demand is an external tailwind, not evidence of Centrus monetizing AI directly
AI DISRUPTION / CANNIBALIZATION RISK tailwind · 1/10
AI does not commoditize uranium enrichment, nuclear fuel contracting, licensing, or centrifuge manufacturing; the core model is governed by physical capacity, regulation, geopolitics, and supply scarcity. Any AI effect is additive through operational efficiency and possible nuclear demand from data centers, not cannibalization of what Centrus sells.
option liquidity: fair
proxy inputs — dollar-ADV $167M · beta 1.436 · px $199.13
source: proxy (no options chain on FMP)
FMP /stable/ exposes no options-chain endpoint on this key, so ATM IV, bid-ask spread and open interest are unavailable. Liquidity below is a PROXY from dollar-ADV, beta and price level (a stand-in for option depth), not measured option-market data.