# SLB data and AI thesis 24 October 2026 · Sunya · Raj Mistry SUNYAENERGY RESEARCHCopy linkDownload textPrintCOMPANY BRIEFING / 24 OCTOBER 2026 The case forSLB’s AI business. The data matters. So do the rights to use it, the software around it, and whether customers pay more.Edited by Raj MistryOriginal-source review · about 8 minutesWhat holds upFollow the moneyWho controls the data?What would prove it?Sources 01 / THE TAKEAWAY There is a business here.The valuation claim needs more work. Our assessment: SLB has a credible route to earning more from AI because it can put technical models into software and operating workflows its customers already use. Its advantage could come from the combination of specialized data, engineering knowledge, governed access and distribution. That is a stronger basis for diligence than assuming every historical customer dataset belongs to SLB. The October 20 Robert Mullin interview with George Noble prompted this review. The host introduces an oil services company whose AI business might be worth more than the whole company. That is an investment thesis. We have not substantiated that valuation, or confirmed the same passage on Eisman’s podcast. Mullin interviewRobert Mullin interview: 3.5% of the S&P, 20% of the Cash · 2026-10-20Host introduction and linked interview; exact SLB interview timestamp not established. SUPPORTED BY DISCLOSURE Specialized models and a way to sell them SLB explains its training approach, application integration and subscription/usage mechanisms. These are management disclosures, rather than an independent test of competitive superiority. Menon · pp. 3–5, 11, 14–15Shashi Menon: The Race to Scale Digital Operations and AI · 2026-06-17PDF pp. 3–5: data access and control; p. 11: seismic training; pp. 14–15: monetization. NEEDS A NARROWER CLAIM A differentiated data position A multiclient seismic library, customer-controlled data and model intellectual property are different assets. Access to one does not establish ownership of all three. Terms · §8–10SLB Online Services Terms and Conditions · UnknownDefinitions: Customer Data and Interaction Data; sections 8(a)–(b), 9(a), 10(a). NOT ESTABLISHED AI profits justify a higher stock value The reviewed materials do not isolate AI earnings, customer acquisition costs or incremental cash returns. A stock valuation needs those bridges and the price paid today. 02 / HOW THE BUSINESS COULD WORK Start with the work an engineer already does. Imagine a geoscientist looking for usable well and seismic data, checking its quality and interpreting a reservoir. The opportunity is to make those steps faster and easier inside the technical tools already used to make expensive decisions. The sequence below is our reading of SLB’s architecture, not a claim that every customer has deployed the complete workflow. 01 Connect the dataLumi and the data bridge SLB describes connecting to customer systems without moving or duplicating their data. Governed access matters when the information is commercially sensitive. 02 Handle the technical formatsDomain foundation models The seismic example starts from a vision transformer, adds geoscience context, and trains on public and SLB multiclient datasets. Customers can fine-tune with their own data. 03 Put it into the workflowTela and existing applications General LLMs help with interaction and orchestration; technical models handle domain tasks. Engineering recommendations can stay under human decision control. 04 Charge for useful workSubscriptions and metered use SLB describes upgrades, Canvas subscription and consumption, tailored workflows and marketplace sharing. The materials describe monetization mechanisms, not an AI-only revenue breakout. Source: SLB’s June 17 presentation. The autonomous capabilities and technical leadership assertions are supplier claims; customer-specific limits still need checking. Menon · pp. 3–5, 11, 14–15Shashi Menon: The Race to Scale Digital Operations and AI · 2026-06-17PDF pp. 3–5: data access and control; p. 11: seismic training; pp. 14–15: monetization. Distribution is a meaningful part of the thesis. In June, the CFO said Digital served more than 1,500 customers, including 90 of the top 100 oil and gas producers. That is the broader Digital customer base, not a count of paid Tela users. CFO · pp. 4, 7–11Stephane Biguet: Proven Growth. Accretive Returns. · 2026-06-17PDF p. 4: customer base; pp. 7–10: commercial models; p. 11: 2030 ambitions. 03 / THE ECONOMICS Digital is already profitable.Digital is broader than AI. In FY2025, SLB reported $2.660 billion of Digital revenue and $928 million of adjusted EBITDA, a 34.9% margin. Those are supplier segment results, not customer savings or a return attributable to AI. FY2025 resultsSLB full-year 2025 results — Digital financial baseline · 2026-01-23Supplementary Information question 12, Digital FY2025 table. What customers pay Digital for Compare the same period in each year. USD millions, as reported.Third quarterFirst nine months 2025 2026 · bar scale starts at zeroPlatforms & Applications273299Digital Operations131163Digital Exploration8078Professional Services174153 As-reported growth includes acquisitions. Category totals may differ slightly from the reported total due to rounding. This is not AI-only revenue, organic growth or a ranking of product quality. Q3 resultsSLB third-quarter 2026 results — Digital economics and category definitions · 2026-10-23Digital results; Supplementary Information questions 10–13; Digital customer highlights.Read the exact chart data and accounting limitsQ3: July–September · USD millionsCategory20252026Platforms & Applications273299Digital Operations131163Digital Exploration8078Professional Services174153Reported total658693 Digital Operations revenue is also recognized in the relevant Core divisions and eliminated at group consolidation. Do not add Digital to Core segment revenues without the eliminations. ARR measures recurring subscriptions and maintenance, excludes variable usage and one-off licenses, and is not additive to revenue. Q3 resultsSLB third-quarter 2026 results — Digital economics and category definitions · 2026-10-23Digital results; Supplementary Information questions 10–13; Digital customer highlights.Download source-linked data · JSON Q3 2026 provides a useful check on the enthusiasm: Digital revenue was $693 million, up 5% year on year; adjusted EBITDA was $228 million. ARR was $1.04 billion, unchanged from June and up 13% year on year. Those figures support an established digital business, without establishing an AI-led earnings acceleration. Q3 resultsSLB third-quarter 2026 results — Digital economics and category definitions · 2026-10-23Digital results; Supplementary Information questions 10–13; Digital customer highlights. Keep the ambition separate from the result Digital adjusted EBITDA · USD billions. Both bars use a zero-to-$2bn scale; hatching marks the ambition.FY2025 · reported0.9282030 · management ambition1.8–2.0 The CFO also targeted about $2 billion of ARR by 2030 and 38–42% adjusted EBITDA margins. These are forward-looking ambitions, not booked results, a valuation or discounted cash flows. CFO · pp. 4, 7–11Stephane Biguet: Proven Growth. Accretive Returns. · 2026-06-17PDF p. 4: customer base; pp. 7–10: commercial models; p. 11: 2030 ambitions. Our reading: the next question is how much additional profit AI creates after compute, support, integration and customer discounts. Revenue that shifts from a traditional software license into a new subscription is not automatically new economic value. Nor does a new pricing model prove customers will accept it. 04 / THE DATA QUESTION Do not put every datasetin the same bucket. The public standard terms make the sweeping ownership argument hard to defend. They also leave room for specified product-improvement uses. Both points matter. Click each category to see what the reviewed sources establish.SLB multiclient libraryCustomer field dataAI prompts and outputsModels, software and know-how A concrete training-data example Menon identifies public and SLB multiclient seismic datasets as training inputs. The source does not establish the rights schedule, geographic coverage or reuse permissions for every dataset. A seismic library is not the same as every customer’s downhole history. Menon · pp. 3–5, 11, 14–15Shashi Menon: The Race to Scale Digital Operations and AI · 2026-06-17PDF pp. 3–5: data access and control; p. 11: seismic training; pp. 14–15: monetization. Access and ownership are different Section 8(a) disclaims ownership of Customer Data. Section 8(b) permits specified uses to provide, monitor and secure services and improve products, plus other agreed uses. Confidentiality and individual agreements still matter. These clauses neither prove universal pooling rights nor prove that reuse never occurs. Terms · §8–10SLB Online Services Terms and Conditions · UnknownDefinitions: Customer Data and Interaction Data; sections 8(a)–(b), 9(a), 10(a). Consent is an explicit boundary For Embedded Gen AI, prompts and outputs are Customer Data. Section B(iii) requires customer consent before using them for model training or further service development. Connected third-party Gen AI follows separate provider terms. Do not extend this specific restriction into a blanket rule for every data category. Gen AI terms · B(iii)SLB Online Services Service Specific Terms · UnknownGenerative AI Services: definitions; Connected Gen AI A(i), A(iv); Embedded Gen AI B(iii). A different kind of asset SLB and its licensors retain the Online Services and related intellectual property under section 9(a). Owning software or a model does not establish ownership of the source data it consumes. Terms · §8–10SLB Online Services Terms and Conditions · UnknownDefinitions: Customer Data and Interaction Data; sections 8(a)–(b), 9(a), 10(a). Terms publication and effective dates are unknown. We reviewed the public text on October 24. It is not a contract-by-contract audit. The relevant diligence is which assets SLB owns or licenses, which uses customers authorize, and how restrictions affect model development and sales. 05 / WHERE THE THESIS COULD DISAPPOINT What would make this less valuable? Better tools, limited pricing power AI may improve the product while customers expect the improvement within their existing fee. Ask for actual incremental spend and renewal behavior. Great demonstrations, uneven field performance Models need to work across unfamiliar geology, data quality and asset conditions. A supplier demonstration is not a held-out-basin evaluation or a production approval. A narrower pool of reusable data Customer consent, confidentiality and negotiated rights could limit cross-customer learning. Conversely, stronger governance could make the platform more attractive to operators. Economics hidden inside the segment Digital includes services, data licenses and established software. Segment growth can reflect acquisitions, mix and license timing as well as AI. A high multiple needs a view of durable post-elimination earnings. These are our proposed tests of the thesis, not documented deployment failures. We found no evidence in this review that the whole AI strategy has failed, or that the investment valuation has been proved. 06 / WHAT TO ASK NEXT Five answers would makethis case much easier to judge. Who pays specifically for AI? Separate paid AI seats, renewals and consumption from bundled trials and the overall Digital customer base. What survives after all the costs? Reconcile incremental revenue to cash contribution after inference, training, implementation, field support and any hardware or channel overlap. Which data can improve the common model? Show ownership, licenses, consent and derivative-model rights by dataset category and jurisdiction. Does the advantage travel? Test against alternative tools on unseen basins and poor-quality data. Publish failure rates, confidence thresholds and approval boundaries. What is already reflected in the stock? Use current valuation, Core earnings, debt, eliminations and cash flows. Do not compare a future divisional EBITDA ambition directly with today’s equity value. For operators, the practical implication is to test a valuable workflow and insist on control over data, access and approvals. For AI providers, the opportunity is to work with the domain tools and deployment constraints already in place. Those are implications of our review, not endorsements of a supplier or a claim that one model will win. Explore SLB’s company dossierRead the expanded Lumi profileSee the Azule drilling example SOURCE NOTES Read the evidence yourself. Dates below are source publication or event dates. The original June investor-day page links the prepared scripts. Selected financial tables were checked against the issuer release. No new adoption or customer-return count was created for this briefing. Open the investor-day source index (https://investorcenter.slb.com/news-events/slb-digital-investor-day) 2026-10-23 / Supplier earnings release SLB third-quarter 2026 results — Digital economics and category definitions (https://investorcenter.slb.com/news-releases/news-release-details/slb-announces-third-quarter-2026-results) Digital results; Supplementary Information questions 10–13; Digital customer highlights. $693 million decreased 1% sequentially Editor read the accessible original publisher text on 24 October 2026. Statements remain publisher-reported; no independent performance validation. 2026-01-23 / Supplier earnings release SLB full-year 2025 results — Digital financial baseline (https://investorcenter.slb.com/news-releases/news-release-details/slb-announces-fourth-quarter-and-full-year-2025-results) Supplementary Information question 12, Digital FY2025 table. Digital revenue increased 9% on a full-year basis Editor read the accessible original publisher text on 24 October 2026. Statements remain publisher-reported; no independent performance validation. 2026-06-17 / Investor presentation and prepared script Shashi Menon: The Race to Scale Digital Operations and AI (https://investorcenter.slb.com/static-files/cf773c3d-348f-479f-aeb0-81f5418844ca) PDF pp. 3–5: data access and control; p. 11: seismic training; pp. 14–15: monetization. we trained it with public and SLB multiclient seismic data sets Editor read the accessible original publisher text on 24 October 2026. Statements remain publisher-reported; no independent performance validation. 2026-06-17 / Investor presentation and prepared script Stephane Biguet: Proven Growth. Accretive Returns. (https://investorcenter.slb.com/static-files/0835f6dc-d2fb-4bc0-b85e-afecadcf36fe) PDF p. 4: customer base; pp. 7–10: commercial models; p. 11: 2030 ambitions. more than 1,500 customers, including 90 of the world’s top 100 oil and gas producers Editor read the accessible original publisher text on 24 October 2026. Statements remain publisher-reported; no independent performance validation. Unknown / Published online-service terms SLB Online Services Terms and Conditions (https://www.software.slb.com/slb-online-services-terms-and-conditions) Definitions: Customer Data and Interaction Data; sections 8(a)–(b), 9(a), 10(a). SLB claims no ownership rights in Customer Data. Editor read the accessible original publisher text on 24 October 2026. Statements remain publisher-reported; no independent performance validation. Effective/publication date unknown. Public standard terms do not establish the terms of every negotiated customer agreement. Unknown / Published online-service terms SLB Online Services Service Specific Terms (https://www.software.slb.com/slb-online-services-service-specific-terms) Generative AI Services: definitions; Connected Gen AI A(i), A(iv); Embedded Gen AI B(iii). SLB shall not use Customer Prompts or Output to train models or further develop the Online Services without Customer’s consent. Editor read the accessible original publisher text on 24 October 2026. Statements remain publisher-reported; no independent performance validation. Effective/publication date unknown. Public standard terms do not establish the terms of every negotiated customer agreement. 2026-10-20 / Investor opinion / interview Robert Mullin interview: 3.5% of the S&P, 20% of the Cash (https://georgenoble.substack.com/p/35-of-the-s-and-p-20-of-the-cash) Host introduction and linked interview; exact SLB interview timestamp not established. The oil service giant hiding a real AI business that could be worth more than the whole company Original host introduction reviewed; confirms the investment framing and interview identity. Full interview passage and exact timestamp not independently checked. An Eisman episode containing the same argument remains unconfirmed. 2026-06-17 / Investor presentation and prepared script Trygve Randen: Our Digital Advantage (https://investorcenter.slb.com/static-files/7aa731c6-2f38-47e4-9aa2-473a8d04ec6e) PDF pp. 2–3: governed platforms; pp. 5–6: hosting and trust; p. 7: operational footprint. more than ninety percent of global production is simulated or modelled using at least one of our digital solutions Editor read the accessible original publisher text on 24 October 2026. Statements remain publisher-reported; no independent performance validation.Back to Sunya Energy AIEdited by Raj Mistry Chart data: https://energyai.sunya.ai/assets/slb-data-ai.json