NVIDIA Meeting — 1-Pager Briefing

The single page you read 10 minutes before walking in. Full picture in 2026-06-20-nvidia-pre-meeting-knowledge-map.

The picture in 60 seconds

NVIDIA is running its reverse supply chain on email and spreadsheets at $5T market cap. The post-SFDC workflow — RMAs, ASNs, repair handoffs to Wistron/Foxconn, customer status — is unmeasured, uncoordinated, and breaking under hyperscaler volume. The chip is rarely the failure — connector pins, heat sinks, network switches are. NVIDIA doesn’t repair GPUs; CMs do, under NVIDIA’s playbook, in Hong Kong / Taiwan / Mexico, with a Dallas line going live July 2026. ~60 of 100 returns get repaired; the other 40 come from new inventory. ~90% of all repairs are remanufacturing — known-issue ECO batches NVIDIA fixes free. Greg + Lonny have an EY-workshop-defined dashboard initiative, an unnamed exec sponsor, and an RFP (per our internal debrief — not their language) with Palantir, Accenture, or build-internal as finalists. They asked us to build it.

Who’s in the room

• Lonny Orona — Customer Service Ops lead. 8 weeks in as of mid-May; ex-Meta 8 years (in-fence fiber → global DC capex → $20M efficiency program). The customer-side perspective. Says “we come to the conversation, not armed.”
• Greg Dalcello — Senior Director, Service Team. Ex-Cisco 22 years (lived the 2007–09 ERP transformation), 2.5 yr at MyNextPower (solar). Operations-transformation specialist. Senior to Lonny by ~1 week. Says “we don’t know the score of the game.”
• Possibly via Brian Feller — VP Global Planning, Logistics & Services. Round Rock, TX. Likely the exec sponsor above Greg + Lonny but unconfirmed. Strong Dell pedigree. May not show in person.

The 5 facts you absolutely walk in with

1. Scale: Meta 100K → 1M GPUs over 5 yrs. Lonny: “struggling to get hundreds back; thousands will break us.” Math checks: 9% annualized failure rate × 1M = 90K failures/yr (~7.5K/month). [Meta Llama 3 paper]
2. Stack: SFDC + SAP + Baxter Planning + Expeditors (replaced Omni). SAP has a $2M+ planning deal in flight. WMS + TMS rolling out now, including into 3PL partners. EDI being turned on for ASN / receipt acknowledgment.
3. Warranty reserve $2.81B FY26 (NOT $8.22B). Accrual rate 0.46% → 0.92% → 1.15% in two years. Claims paid $957M (+337% YoY). Attribution: Compute & Networking. [NVIDIA FY26 10-K]
4. Failure mix: ~60/100 repair-vs-replace ratio. ~90% remanufacturing (ECO/recall pattern) — all free to customer. The chip is rarely the problem; bent connector pins, network switches, heat sinks are. NVIDIA does NOT repair its own GPUs — Wistron + Foxconn do.
5. Telemetry walled off: customers strip chips before return. NVIDIA’s first signal of product perf is ticket volume. Mission Control + Run:ai is the forward fleet layer; nothing covers the reverse layer. Hyperscalers wrote OCP RAS v1.7 (Oct 2025) — they are setting the serviceability spec.

The 5 things you don’t know to listen for

1. Who actually owns budget on this engagement. Brian Feller? The (publicly open!) VP Global Service Operations req that names Baxter and SAP? Shoquist directly? Ask: “To make sure we scope to the right approval path — who would own sign-off on a Phase 1 engagement of this size?”
2. The second EY workshop initiative. They’ve shared one; not the other. After NDA, get the full pair so we know what we’re sitting alongside.
3. What Palantir and Accenture are actually proposing. Bliss came from Palantir — anticipate the Foundry pitch and structurally undercut it (per Joe Malchow: “two solutions people + arm’s-length independent data repository + milestone economics”).
4. Whether the “RFP” framing is real. Greg + Lonny did not use the word RFP in the meeting. They said “target rich environment” and asked for a bullet doc. We reframed it as an RFP internally. Pressure-test whether this is a structured RFP or an executive-sponsored direct procurement.
5. Dallas — production vs. repair. Wistron Fort Worth is two buildings, not one. The Mobility Way 767K sq ft building (described publicly only as “renovations”) is the likely repair line. Ask: “Which building, and is it the same SOW as production?”

The 3 things that would change the picture if they say them

• “Our sponsor is [X].” — One name is the highest-information sentence Thursday can produce. If they name someone we haven’t found in public sources (i.e., not Feller, not Shoquist, not the open VP req), that person is who matters and the entire approach updates.
• “The decision is in [N] weeks.” — Anything tighter than 4 weeks means accelerate Phase 1 prototype and skip discovery. Anything longer than 8 weeks means we have room to embed and shape the spec.
• Anything about telemetry-sharing being possible. Lonny softened in the night-before chat (“not that the data doesn’t exist; it’s just inaccessible”). If a customer has agreed to share triage data — even one — that’s the wedge into Phase 2. Listen for: “well, we’ve got an arrangement with [customer]…”

What we want from the meeting

• Concrete: NDAs (NVIDIA-issued, two individuals — straightforward); the second EY initiative; the procurement vehicle; the timeline.
• Relational: the unnamed exec sponsor; warm intro path to Brian Feller; visibility into Alex Zhu’s reverse-supply-chain transformation (so Phase 1 doesn’t overlap or conflict).
• Signal: how committed Greg + Lonny actually are to the outside-the-box framing they used in the meeting (“we really need to think out of the box here”). If they walk us through their pre-baked answer instead, we are in a different game.

What we walk in offering

Bullet doc + clickable shiny-object prototype that demonstrates we understand the latent problem in their own quotes: “we don’t even know the score of the game” → here’s the score. The economic logic is straightforward: NVIDIA’s reverse-flow data layer is the substrate for both the dashboard NVIDIA needs now and the financialization layer (warranty risk transfer, parametric coverage) that sits on top later. Phase 1 ships the dashboard. Phase 2 is the data + underwriting story we don’t bundle into the first conversation.

One last thing — surface, don’t reframe

If government / defense comes up, flag it for the founders. Don’t reframe around it. Per compliance_wedge_killed.

---

Anchor sources: 2026-05-12-lonny-orona; 2026-05-27-nvidia-reverse-logistics-supply-chain-discussion; 2026-06-17-logistics-catch-up-wgreg-lonny-nvidia; 2026-06-17-lonny-greg-debrief; 2026-06-17-lonny-chat. Full evidence + research: 2026-06-20-nvidia-pre-meeting-knowledge-map.