Post-NVIDIA Outreach Push — Target Sheet
What this is
Twelve companies across three tracks to activate during the procurement/legal waiting period that follows Thursday’s NVIDIA meeting. The job of the push is not pipeline development. It is to answer one question: is the reverse-flow problem NVIDIA-specific or generalizable across hardware OEMs? That answer determines whether TBD is a venture or a single-customer services engagement.
Decision Bliss needs to make to activate: which 3-5 of these to lead with this week, who carries each conversation, and what framing (see open questions §). Templates come after prioritization, not before.
Prioritization recommendation — lead with these three
- AMD (Track A). The cleanest test of generalization. Same chip-then-board-then-hyperscaler stack, same ODM partners (Quanta, Foxconn), same hyperscaler customers. If AMD’s reverse-flow ops look like NVIDIA’s, the venture is real. If AMD has solved this internally, the venture is harder and NVIDIA may be an outlier. Tae Kim flagged as a warm path via Josh per Josh, Apr 30 — exercise that.
- Meta GPU infra reliability (Track A’). The single highest-leverage customer-side call we can make. Lonny was there for 8 years; he knows who runs reliability today. If Meta confirms the “let it fail” pattern Lonny described from the customer side (Lonny, May 12), we have buyer-side validation of the Phase 2 hedge/warranty product, not just NVIDIA-side seller pain.
- Aon (Track C). Preston Wilson at Guy Carpenter (Marsh) has been the most generous external advisor on the financialization side across three sessions. He offered intros in May 7 that we haven’t taken yet. Aon is a competing broker, but Preston’s offer to point us at his firm’s parametric specialist is the closer path; both should be in scope. Light up the parametric/warranty-transfer conversation in parallel with the dashboard work so Phase 2 has a buyer-side counterpart by the time NVIDIA closes Phase 1.
Calibration — what assumes generalization vs. safe anchors
| Target | Assumes reverse-flow generalizes? |
|---|---|
| AMD, Broadcom, Marvell | Yes — directly tests it. If they’re fine, thesis weakens. |
| Dell, HPE, Supermicro | Partial — they live the customer-warranty side already; generalization here is more about whether dashboards aren’t already built |
| Meta, CoreWeave, Microsoft | Partial — tests the buyer-side hedge demand independent of who supplies |
| Aon, Munich Re, Swiss Re | Safe anchor — the financialization wedge stands on its own; these calls light up Phase 2 regardless of how generalization shakes out |
| Hursh’s debt network | Safe anchor — Spencer Powers and Ronit Jain both independently named debt financiers as the first hedging adopter (Spencer, May 22; Ronit, May 22) |
Track A — Other hardware OEMs (validate the category)
| Company | Why relevant | Persona | Named individual | Warm/Cold | Warm-path | First-touch | Hook |
|---|---|---|---|---|---|---|---|
| AMD | Same MI300/MI350 stack, same ODM partners, same hyperscaler customers as NVIDIA — the cleanest direct test of whether the email-and-spreadsheets reverse-flow problem is industry-wide or NVIDIA-specific. AMD is also growing into volumes where Lonny’s “hundreds → thousands” scaling break should be hitting them now. | VP / Sr Director, Customer Service Operations or Supply Chain | Research needed — start with LinkedIn search for “Director Customer Service” or “VP Service Operations” at AMD, Austin. Tae Kim (AMD, supply chain) flagged as warm-path entry per Josh, Apr 30. | Warm (via Tae Kim → Josh) | Josh → Tae Kim → AMD Service Ops. Cite 2026-04-30-josh-bliss-dustin. | Warm intro via Josh, with a 2-line “what we’re learning” share. | ”We’re talking to NVIDIA about how they manage reverse flow at hyperscaler volume — failure rate, repair cycle, customer comms. Curious how AMD’s setup differs as MI350 ships ramp.” |
| Broadcom | Networking switches were the failure point Greg flagged at NVIDIA — “bent connector pins, network switches, heat sinks” not GPUs. Broadcom’s Tomahawk/Jericho switches sit in the same racks; if reverse logistics for network gear in DC racks looks like GPU returns, the category broadens beyond accelerators. Recommended by Michelle for parallel diligence per Michelle, May 22. | VP, Service / Customer Operations (networking BU) | Research needed — start with LinkedIn search for “VP Service” or “Customer Operations” at Broadcom networking division (Palo Alto / San Jose). | Cold (Michelle declined direct intro until scope narrows) | Michelle offered to consider intros once we’re more specific. 2026-05-22-michelledustinbliss. | Cold email via IR with “students researching reverse logistics at scale” framing — or wait until Michelle path opens. | ”Three of the failure modes NVIDIA’s service team names — connector pins, switches, heat sinks — sit upstream of GPUs. We’re trying to understand how networking OEMs structure the return-and-repair flow for hyperscaler-scale switch deployments.” |
| Marvell | Custom-silicon path (Trainium, MTIA, Google TPU partner) means Marvell sees the inside of hyperscaler reliability programs that NVIDIA doesn’t. Different vantage point on the same fleet. Also recommended by Michelle for parallel diligence. | Director, Customer Support / Reliability Engineering | Research needed — LinkedIn search “Customer Support” or “Reliability” at Marvell, Santa Clara. | Cold | None currently. Possible second hop via Michelle if AMD/Broadcom convert. | Cold email via IR or LinkedIn outreach. | ”Custom-silicon programs at AWS, Meta, Google touch your reliability flow in a way the merchant-GPU flow doesn’t. We’re trying to map the differences.” |
| Dell | Brian Feller (NVIDIA’s likely exec sponsor) is Dell pedigree per 2026-06-20-nvidia-meeting-1pager-briefing. Dell ProSupport runs one of the largest server-warranty operations on the planet — they may have already solved the dashboard problem internally. Whether they have or haven’t is load-bearing for the thesis. | Director, ProSupport Operations or Global Services Strategy | Research needed — LinkedIn search “ProSupport” or “Global Services” at Dell Round Rock. Cross-reference Brian Feller’s Dell tenure for shared colleagues. | Cold (potential warm via Feller if NVIDIA meeting opens that door) | Speculative: NVIDIA Thursday → Feller → Dell alums. 2026-06-20-nvidia-meeting-1pager-briefing. | Cold email via IR or LinkedIn; wait on warm path until after Thursday. | ”Dell ProSupport handles server warranty at scale most others can’t match. We’re researching how OEM warranty ops look at hyperscaler GPU volumes — what’s portable and what isn’t.” |
| HPE | GreenLake + Cray inheritance puts HPE in the AI server warranty business with a managed-services model NVIDIA does not have. Pure structural comparison: does the managed-services wrapper solve the reverse-flow data problem, or just hide it? | Director, GreenLake Operations or HPE Services | Research needed — LinkedIn search “GreenLake Services” or “HPC Customer Support” at HPE Spring (Houston). | Cold | None identified. | Cold email via IR. | ”HPE wraps GreenLake hardware in a managed-services warranty model that NVIDIA’s direct-to-hyperscaler flow doesn’t have. Curious whether that wrapper actually simplifies reverse logistics or just relocates the complexity.” |
Track A named-individual hit rate: 1 of 5 (Tae Kim → AMD). Research needed on the other four before activating. Recommend deferring Broadcom / Marvell / Dell / HPE outreach until LinkedIn pulls produce a real human name to address — cold emails to “VP of X” don’t clear the Konrad S.H.I.T. test.
Track A’ — Customer-side (the people who ship broken chips back)
| Company | Why relevant | Persona | Named individual | Warm/Cold | Warm-path | First-touch | Hook |
|---|---|---|---|---|---|---|---|
| Meta | Lonny was there 8 years and named Meta’s “let it fail rather than authorize downtime” pattern as a recurring planning headache for NVIDIA (2026-05-12-lonny-orona). Meta is at 100K GPUs heading to 1M — exactly the scaling profile Lonny says will break the current system. Best single buyer-side validation we can run. | Director / VP, GPU Infrastructure Reliability or Capacity Engineering | Research needed — Lonny’s old Meta network. Names to ask Lonny for once NDA is signed: anyone currently running GPU reliability or capacity at Meta DCs. Lonny mentioned breaking up “monolithic network infrastructure group” — his old org. | Warm (via Lonny, post-NDA) | Lonny → former Meta colleague in GPU reliability. Cite 2026-05-12-lonny-orona. | Ask Lonny for one named intro at the Thursday meeting or in the follow-up. | ”Lonny mentioned the ‘let it fail’ pattern from the supplier side. We’re trying to understand how the customer side weighs proactive replacement against business-unit downtime cost — and whether anything has changed as the 100K-to-1M trajectory lands.” |
| CoreWeave | The largest pure-play GPU neocloud, debt-funded against GPU collateral, the canonical Spencer/Ronit buyer of compute hedging instruments. Failure rate at CoreWeave is a P&L line item, not an operational annoyance. They also ship broken chips back to NVIDIA — both halves of the wedge run through them. | VP Operations or Director, Hardware Reliability | Peter Salanki (CTO and Co-founder, CoreWeave — public via CoreWeave leadership page). Reliability ops lead unknown — research needed. | Cold | Possible hop via Hursh Agrawal’s neocloud-debt network per 2026-06-20-phone-call-with-hursh-agrawal; possible hop via Joseph at Compute Exchange (Spencer Powers’ intro offer, 2026-05-22-45-mins-dustin-ross-and-spencer-powers). | Warm intro request to Hursh or Joseph; reserve cold email as fallback. | ”We’re hearing from NVIDIA that broken-chip return cycles are getting worse as hyperscaler volumes climb. Want to test whether CoreWeave sees the same break point and how it shows up in your P&L vs. ops queues.” |
| Crusoe | Smaller and newer than CoreWeave — different question. If Crusoe doesn’t yet feel the reverse-flow pain, the thesis is volume-gated; if they already do, it’s structural. Useful contrast data point. | Director, Site Reliability or Data Center Operations | Research needed — LinkedIn search “Site Reliability” or “Data Center Operations” at Crusoe, Denver/SF. | Cold | None currently. Possible hop via Hursh’s debt network. | Cold email or LinkedIn; defer until Hursh hop tested for CoreWeave. | ”You’re at the volume where a single bad rack hurts. Trying to understand how a smaller-scale GPU operator sees the broken-chip return cycle vs. how a CoreWeave or Meta would.” |
| Microsoft Azure (Azure HPC / GPU Fleet) | Hyperscaler with the largest NVIDIA install base outside Meta. They wrote OCP RAS v1.7 (Oct 2025) — the serviceability spec NVIDIA is now reacting to. If anyone has already built the customer-side dashboard NVIDIA wants, it’s Azure. | Principal PM / Director, Azure GPU Fleet Reliability | Research needed — LinkedIn search “GPU Fleet” or “AI Infrastructure Reliability” at Microsoft Redmond. OCP RAS spec authors may surface in public spec PDF. | Cold | None warm currently. Possible distant hop via Bliss’s Palantir/MSFT alumni — flag for Bliss. | Cold email or LinkedIn after spec-author research; or wait for Bliss network check. | ”OCP RAS 1.7 reads like a customer-side serviceability spec written for a supplier that hasn’t caught up. Curious how the Azure fleet team is operationalizing it on the customer side.” |
Track A’ named-individual hit rate: 1 of 4 (Salanki at CoreWeave, and that’s the CTO, not the reliability lead). Lonny is the unlock for Meta. Without warm Meta intro, this track is weaker than Track A. Microsoft and Crusoe are research-gated.
Track C — Financial counterparts (Phase 2 buyer/partner side)
| Company | Why relevant | Persona | Named individual | Warm/Cold | Warm-path | First-touch | Hook |
|---|---|---|---|---|---|---|---|
| Aon | Aon is one of the two big brokers (vs. Marsh, where Preston sits). Going to Aon in parallel with Marsh is the normal diligence motion for any insurance-product company. Aon’s Reinsurance Solutions arm has built parametric specialty placements that Guy Carpenter has not. Preston has been generous on Marsh side — Aon is the second probe. | Director, Parametric Solutions or Property Specialty | Research needed — Aon’s parametric leadership is named publicly; start with Aon Reinsurance Solutions parametric page and identify the parametric lead. | Cold | Preston (Marsh) is a soft warm-path equivalent inside the other big broker; not a direct Aon intro. | Cold email via Aon’s parametric solutions team, or LinkedIn to the parametric lead once identified. | ”We’re building a measurement-and-modeling layer for chip-fleet reliability that would feed a parametric warranty or BI product. Talking to brokers across the industry to understand what would clear your four-pillar test for a novel man-made-risk line.” |
| Munich Re | Munich Re reinsured Hithium’s 15-year battery storage performance warranty (2023) with the TWAICE data partnership — the closest existing structural analog to a GPU warranty risk-transfer deal, per insurance-market-overview-2026-06-15 §2. They’ve already done the “novel man-made warranty + sensor data partner” trade. | Underwriter, Specialty Lines (Tech / Warranty) — Munich Re Group Specialty or HSB (Munich Re subsidiary, U.S. equipment warranty) | Research needed for HSB and Munich Re specialty leadership. Hithium-TWAICE deal team may be findable via Munich Re press, 2023-10. | Cold | None directly warm. Possible hop via Preston / Marsh (broker would normally intro a placement-stage carrier). | Cold email referencing the Hithium structural parallel — or warm via Preston once we’re ready to bring a structure to him. | ”The Hithium warranty placement is the structural parallel for a GPU warranty risk-transfer trade. We’re early but want to understand what would have to be true on the data and sensor side for an equivalent placement in semiconductors.” |
| Swiss Re | Swiss Re overtook Munich Re as #1 reinsurer by 2024 GPW. Their sigma reports are the industry benchmark and their specialty teams shape novel-risk uptake. Talking to both Munich and Swiss is the right diligence move; betting on one means the other won’t write. | Specialty Underwriter, Tech / Engineering Lines | Research needed — Swiss Re Corporate Solutions has named specialty leaders; Pina Albo (formerly Hamilton Insurance Group, ex-Munich Re Americas) historically a relevant relationship. Verify current roles. | Cold | None directly warm. Same potential broker-hop path as Munich Re. | Cold email via specialty / engineering lines team. | ”We’re mapping the universe of reinsurers who could write a GPU warranty parametric. Munich Re’s Hithium deal is the closest analog. Want to understand how Swiss Re’s specialty desk would price the same structure.” |
| Goldman Sachs (Infrastructure debt — neocloud financing) | Hursh’s network. Spencer Powers and Ronit Jain both named “providers of debt capital to neoclouds” as the first adopter of compute hedging — collateral value goes up if asset price stability goes up. Goldman, Brookfield, Blackstone Infrastructure all underwrite neocloud debt. Goldman is the canonical first call. | VP / MD, Infrastructure & Asset Finance (data center / AI infra debt) | Research needed — names via Hursh. Possibly via [Goldman Sachs Asset & Wealth Management infrastructure team page]. | Warm (via Hursh) | Hursh → Goldman infra debt contact. 2026-06-20-phone-call-with-hursh-agrawal. | Ask Hursh for one name on a 20-min call next week. | ”Spencer Powers at DRW and Ronit Jain at Pluto both predicted debt financiers would adopt compute hedging before AI CFOs do — because their collateral is the GPU. Trying to test that with someone actually underwriting the loan.” |
Track C named-individual hit rate: 1 of 4 (Hursh’s contact, but Hursh has to name them). Aon, Munich Re, Swiss Re all need a research pass to land on a specific person before activating. Preston’s offered Marsh intros remain unspent — easier path than cold Aon/Munich/Swiss.
What would invalidate this list
- NVIDIA Thursday goes south. If Greg/Lonny walk us through a pre-baked answer rather than the “out of the box” framing they used previously, the dashboard wedge weakens and the financialization tracks have to carry the venture alone. Track C becomes load-bearing; Tracks A and A’ become validation work, not wedge-development.
- AMD says “we already built this internally.” If AMD’s reverse-flow ops are clean, NVIDIA is an organizational outlier and the generalization thesis cracks. Track A’ (customer-side hedge demand) becomes the more important question to answer fast — the wedge moves from “dashboard for OEMs” to “warranty product for hyperscalers.”
- Meta says the “let it fail” pattern Lonny named no longer exists. If Meta has solved business-unit-vs-DC-team approval, the cascade Lonny described doesn’t generalize and the inventory-imbalance pain point is NVIDIA-specific. Phase 2 hedge product has weaker buyer-side urgency. Move Track C up, deprioritize Track A.
Open questions for Bliss
- Framing: “students researching” or “startup building”? The strategy session (2026-06-22-tbd-internal-nvidia-process-walk-thru-shiny-object-discussio) settled on “students” for cold IR outreach. But for warm intros via Hursh, Lonny, Preston, that’s the wrong frame — they know we’re a startup. Pick the line per track, not per company. Default: “students” for cold IR, “TBD founders researching reverse-flow generalization in the semi supply chain” for warm.
- Who owns Track C? Dustin’s been carrying the financialization conversations (Preston, Spencer, Ronit, Hursh). Bliss has been carrying NVIDIA. If Track C activates this week in parallel with NVIDIA follow-up, do we double-team or split? Spencer’s “Joseph at Compute Exchange” intro is sitting unactioned — that’s the cleanest single ask if Track C runs in parallel.
- Do we put any of these conversations on hold until NDA-clearance from NVIDIA? Some of the Meta and CoreWeave conversations risk implying we have insider NVIDIA data we don’t yet have permission to discuss. Recommend: all Track A and Track C calls run pre-NDA; Track A’ (especially Meta) waits until we know what we can and can’t reference.
- Conference path for Track A. Dustin registered for the chip/AI infrastructure conference (Jul 22-23, SF) per 2026-06-22-tbd-internal-nvidia-process-walk-thru-shiny-object-discussio. Several of these targets (AMD, Broadcom, Marvell, Supermicro) will have people there. Worth a pre-conference outreach wave specifically engineered to convert into in-person 20-minute meetings at the event.
Anchor sources: 2026-06-20-nvidia-meeting-1pager-briefing; 2026-06-22-tbd-internal-nvidia-process-walk-thru-shiny-object-discussio; insurance-market-overview-2026-06-15; 2026-06-20-phone-call-with-hursh-agrawal; 2026-05-12-lonny-orona; 2026-05-22-45-mins-dustin-ross-and-spencer-powers; 2026-05-22-30-min-meeting-between-ronit-jain-and-dustin-j-ross; Preston May 22; 2026-05-22-prestonbliss; 2026-04-30-josh-bliss-dustin; 2026-05-22-michelledustinbliss; 2026-05-22-jason-dustin-bliss-if-avail; 2026-05-22-mo-islam.