Scott Hara
Jan 21, 2026 - Oct 23, 2026
11
Decisions
0
Active Todos
13
Patterns
Categories
Decisions (11)
Peter personally drives NVIDIA to commit, to Google, that it will support CIQ-built RLC on NVIDIA hardware
At the 10/22 GDC roadmap meeting Google disclosed NVIDIA had put in writing that Google must reproduce NVIDIA-hardware bugs on RHEL before NVIDIA will look at them. Peter took it on personally: called Scott Hara (NVIDIA) that evening - NVIDIA engineering to align, a three-way NDA, and a statement to Google that NVIDIA will work with both CIQ and Google; emailed Tissa (Google) and shaped Google question to include joint support given CIQ custom kernel work for Google; forwarded to Scott asking for a positive written response; asked Greg to be ready to make the case to NVIDIA. In the same meeting he told Google CIQ can support their roadmap for the next year line item by line item, in response to Google assembling its own senior kernel team.
Vera Rubin to Google - commit to a bounded duration (at most 3 weeks after node access), not a date, and invite Google to supply its own hardware
Replying to Tissa request for Vera Rubin timelines, Peter wrote that NVIDIA has committed lab time but cannot confirm when; once CIQ has a node it needs at most 3 weeks to validate Rocky, ensure full functionality and performance, and fix issues; CIQ will update as soon as it has visibility; and if Google has a node it could provide, CIQ would happily do the 3 weeks on that instead.
Vera Rubin access is drop-everything, the goal is proving it works for Google rather than benchmarks, and kernel pre-work starts now
NVIDIA is working to give CIQ a short window (likely about 5 days, one node) on Vera Rubin hardware, with no advance notice of when it opens. Peter set three things: when the window opens, whoever is involved drops all other work; the goal is getting Rocky/RLC working so CIQ can tell Google it works, not benchmarking; and Sultan starts pre-work now (patch identification, test branch) so the window is productive.
Peter Computex condition: product must be production-ready, not a POC
For the proposed early-June Computex Fuzzball-on-DGX-Spark announcement, Peter set one engineering-side condition: the product must be production-ready (not just a POC). Bjorn separately set the GTM-cadence gates (max 6-week lag between announcement and delivery; sufficient PR-runway for Lindsay and Cathay). Peter held the engineering line cleanly and let Bjorn hold the product-marketing line.
NVIDIA Partnership - Resource Commitment for Grace Vera Patch Support
Committed to assessing headcount needs for NVIDIA Grace Vera patch support — both for the first 6 months and then ongoing. Forwarded NVIDIA patch list to Nathan for SWAG assessment. Nathan estimated 3-6 months for RLC, faster for CLK 6.18. Communicated requirements to Scott Hara: hardware access, test suites, functional and performance targets.
Expanded NVIDIA licensing ask to include Fabric Manager + NVISM
Expanded the NVIDIA licensing negotiation scope beyond DOCA-OFED to include Fabric Manager, NVISM, and other critical InfiniBand components not currently in the CUDA bundle. Sent email to Scott Hara while Bjorn was drafting the DOCA-OFED amendment language.
Committed to aggressive NVIDIA GPU Operator self-certification timeline for GTC
Led GPU Operator Self Certification meeting with NVIDIA. Committed CIQ to building pre-compiled GPU driver containers for Rocky Linux (mirroring Ubuntu model) and pursuing self-certification targeting preliminary completion by end of next week to support a GTC announcement.
Set NVIDIA positioning: benchmarks are AMD-only, offer to collaborate on tuning
Defined strategic positioning for NVIDIA benchmark situation: frame existing benchmarks as AMD-only, and offer to work with NVIDIA to tune Rocky Linux for their platform. This turns a potential embarrassment into a partnership opportunity.
Mobilized team for Saudi meeting prep and escalated NVIDIA DOCA blocker
Peter personally intervened to prepare team for critical Saudi Arabia partner meeting on RLC-AI. Posted in #product-rlc-ai asking about CUDA/DOCA availability, discovered NVIDIA written approval for DOCA OFED still pending. Emailed Scott Hara (NVIDIA) directly to advance the approval. Tagged Nathan, Justin, Jeff Uphoff, and Damen Knight demanding they answer Max's detailed technical questions within 24 hours. Set hard deadline: '24 hours from now.' Bjorn committed to calling Scott to reaffirm DOCA modification rights.
Pursue NVIDIA self-certification path for NVAIE integration
CIQ will pursue the self-certification path for NVIDIA NVAIE integration, targeting a GTC announcement. Will adopt NVIDIA preferred embedded license model, integrating NVAIE license cost (~$4,500/GPU) into CIQ product and providing L1/L2 support.
NVIDIA partnership scope agreed - CIQ Rocky Linux for NVIDIA AI at GTC
Agreed with Scott Hara on NVIDIA partnership deliverable: CIQ will deliver a free CIQ Rocky Linux for NVIDIA AI (with AI patches + NVIDIA drivers) for GTC showcase in ~8 weeks. No requirement to upstream to community Rocky short-term. Support cadence: 1-2h kickoff Q&A, then 3 weekly 1h sessions, then ad hoc. CIQ maintains distro with minimal NVIDIA intervention; NVIDIA supplies patches, system access, and optimization guidance.
Related Patterns (13)
Executive Sponsorship for Strategic Partnerships
Strategic cross-company initiatives and major client partnerships require executive-level accountability to move at the right pace and ensure proper prioritization.
Protect Engineering Capacity
When external demands threaten to overload engineering capacity, protect capacity by either requiring the demand to come with additional resources, or forcing hard prioritization choices upstream.
Small Circle for Sensitive Operations
When executing sensitive strategic operations, keep the circle of informed people as small as possible to prevent leaks that could accelerate hostile action or undermine the initiative.
Reclassify to Route - Decide Who Owns It, Then Give a Criterion Not a Verdict
When a question arrives at Peter framed in one domain, he often does not answer it in that framing. He reclassifies it into the domain that actually governs the outcome, which relocates ownership to whoever owns that lane, and then hands down a decision criterion rather than a verdict. The reclassification IS the decision - it determines who decides. He does this rather than adjudicating on the merits himself, even when he was explicitly cc-ed and even when a direct report challenges him on the substance.
Purpose Is the Decision Procedure
When someone asks how a process, board, tool or artifact should handle their specific case, Peter refuses to answer on the askers terms. He re-derives the answer from what the artifact is for, states that purpose as the only decision procedure, and lets the answer fall out. Anything found to be serving a second purpose is deleted rather than debated. Tools and views are explicitly subordinate to definitions. He will accept that the askers real problem is simply not solved by this artifact rather than widen the artifact to cover it.
Route Non-Differentiating FTE Classes to Partners
When CIQ would otherwise need to staff a function whose work does not differentiate the company — compliance bureaucracy, audit/cert paperwork, ongoing regulatory attestation, etc. — Peter routes the load to a partner who already owns adjacent capability rather than adding the FTE class internally. Org-shape decision dressed as a partnership decision.
Demonstrate the Standard, Then Collect It
When a new leadership expectation has to land across multiple orgs, Peter does not write a spec. He runs a live exemplar in public with the strongest performer first, says openly that the ordering was deliberate, keeps the form of the deliverable open so the ask stays about the thinking rather than the artifact, lets the delta be self-evident to the people who will have to close it, and then collects the same from everyone else on a named rotation. He prefers showing imperfect work now over polished work later.
Decide on the Precedent, Not the Case
When a request arrives framed as a one-off, Peter does not evaluate the one-off. He evaluates the rule that granting it would write, and answers that instead. Peter generalized this himself on 2026-08-11: it is always about understanding precedent - whether it is a comp change or anything that structurally affects the company. It is therefore NOT a compensation pattern; the domain of the presenting request is incidental. The tell that this pattern is live is that the stated reason for the answer refers to future cases rather than to the merits of the present one.
Switch Instrument, Not Intensity
Applies when a message, rule or ask has already failed through the channel Peter is currently using - repeated advocacy that has not landed, coaching that produced defensiveness rather than improvement, exhortation that did not change behaviour, or a rule that keeps losing case by case. His response is never to restate the same thing more forcefully. He changes the instrument and usually names the switch out loud. Mined from 7 independent observations between 2026-01-06 and 2026-08-21.
Time Over Cost
Peter treats elapsed time as the scarce quantity and money as the abundant one, and will spend cash to buy schedule, clarity or optionality. Notable because he named it himself as a transferable principle on 2026-05-21 and coached the lens explicitly to a direct report, which is rare - most of his patterns are inferred rather than stated. It generalises past procurement: he will pay to convert an ambiguous dispute into a binary one, remove cost as a variable entirely when a priority has executive visibility, and reason about a commitment in terms of lead time rather than budget. Mined from 5 independent observations between 2026-02-01 and 2026-09-11.
Promise Only What Is Inside CIQ Control
Applies when someone asks CIQ to commit to an outcome whose delivery depends partly on something CIQ does not own - a lead time, a third-party publisher, a customer-provided environment, an unsigned contract. Peter does not accept or refuse the request as posed. He isolates the specific component that sits outside CIQ delivery chain, refuses that component by name, and substitutes an adjacent commitment CIQ can actually keep - then often adds a harder problem worth solving in its place, so the counterparty gets more engagement rather than less.
Measure, Commit Publicly, Then Execute
A three-beat escalation arc Peter runs when he needs a commitment from a peer or the CEO that private pressure has not produced. Beat one: measure the thing and get a number. Beat two: state the commitment in a venue where the people who matter are watching. Beat three: execute against it. The 2026-04-15 observation names the sequence explicitly and notes he does not skip steps, because each beat builds the legitimacy of the next. The venue choice is treated as a decision in its own right, separate from what the message says. Mined from 8 independent observations between 2025-12-29 and 2026-08-28. NOTE: scope this to the venue-and-sequence mechanism specifically - forcing function is a phrase that appears loosely across many unrelated decisions and should not be taken as a match on its own.
Install the Criterion Before the Argument
Applies to a contested question that has not been argued yet - a meeting about to start, a test about to run, an escalation just received, a proposal about to be written. Peter moves first to fix the criterion, the boundary or the posture, before anyone argues the substance. The consistent effect, named explicitly in the 2026-07-29 observation, is that the other party cannot win by being more persuasive on the merits, because the merits are no longer the axis. Distinct from Reclassify to Route in mechanism: that pattern reassigns who owns a question, this one controls WHEN the criterion lands relative to the debate. The two frequently co-occur. Mined from 7 independent observations between 2026-01-20 and 2026-08-31.