# Pearl, NVIDIA & Attestable: The Public Record — Pearlified Canonical page: https://pearlified.com/face > Independent publication, not affiliated with Pearl Research Labs. Source quotations, team statements, and Pearlified interpretations are distinct. Historical odds are editorial guesses, not confirmed relationships. The historical archive is primarily through September 17, 2026, with selected research and corrections reviewed September 21–22 a compute-payments research update reviewed September 23, and community-directory and inference-provider context reviewed September 24. Protocol and proof-of-inference L2 statements reviewed September 25. Individual observation dates and limitations apply. In this file 5 sections [PEARLIFIED / THE OPEN FILE](https://pearlified.com/)PUBLIC RECORD / 01 THE PUBLIC FACE / IDENTITY STUDY # One company. Two readings. An inference platform. A compute-backed network. Follow how Pearl describes itself—and what the public record can actually establish. [Read the public record ↘](https://pearlified.com/face#exhibits-face) 01 / INFERENCE02 / NETWORKONE SUBJECT / TWO PERSPECTIVES Open the investigation brief THE INVESTIGATION / READ THE CASE BEFORE THE EXHIBITS ## What does Pearl’s public positioning actually establish? 01 / STRONGEST RECORDED CLUE The archived Pearl announcement describes an accelerated inference platform and NVIDIA Inception membership. The quoted reply reinforces that AI-facing description. [Listen to the hardware context ↗](https://pearlified.com/reading#clip-nvidia-asic-for-matmuls) 02 / COMPETING INTERPRETATIONS Our interpretation is that AI infrastructure is central to Pearl’s public identity. An alternative is simpler: a company can describe one part of a broader product without concealing the rest. 03 / THE LIMIT OF THE EVIDENCE Program membership and event participation do not establish a specific commercial agreement or explain an admission decision. 04 / WHAT WOULD SETTLE IT A named joint project, scoped commercial announcement, or attributable explanation would settle more than branding can. Historical September briefing; new research sections below identify their sources and observation dates. [Inspect the original exhibits ↓](https://pearlified.com/face#exhibits-face) THE EXHIBITS / SOURCES, QUOTES & EDITORIAL READS @prlnet✓ verifiedJul 15 Pearl Research Labs is officially a member of the @Nvidia Inception program! We are building an accelerated inference platform to run the world's leading open-weight models — powered by Hopper and Blackwell GPUs. THE PUBLISHED CRITERIA ## Who qualifies? NVIDIA’s current Inception eligibility excludes cryptocurrency-associated companies and cloud providers. ? THE ARCHIVED ANNOUNCEMENT ## Pearl says it is in. The membership announcement describes an accelerated inference platform built for Hopper and Blackwell GPUs. An unresolved policy question. NVIDIA’s current Inception criteria exclude cryptocurrency-associated companies and cloud providers. Pearl’s archived announcement describes membership. The admission rationale and the effective date of the policy wording have not been established. [Current eligibility ↗](https://www.nvidia.com/en-us/startups/) Our interpretation: the AI positioning may explain the fit. The archive does not establish NVIDIA’s admission rationale. Pearl presents itself here as "an accelerated inference platform." The token is downstream. Someone said this to their face: Evan Van Ness@evan_van_nessJul 15 Crypto barred from Nvidia. 👀 Pearl Labs is functionally an AI inference company. @prlnet✓ verifiedJul 15 Indeed 💪 “Pearl Labs is functionally an AI inference company.” “Indeed.” — Evan Van Ness, crypto commentator, noting NVIDIA Inception bars crypto startups · answered by @prlnet, Pearl's own verified account · July 15 2026 The reply supports an AI-facing description. It does not explain NVIDIA’s admission decision. Our interpretation is that inference is central to Pearl’s positioning; calling the exchange a confirmed workaround would go beyond the record. 🔒 And it's not just a badge — they're in the room The membership isn't a logo on a landing page. Pearl's own channels put CEO Omri Weinstein on a speaker slide at a closed NVIDIA event, discussing "the intelligence economy" — seated beside Prof. Freddy Gabbay, a Hebrew University chip-design professor who spent years as VP of Chip Design at Mellanox — the networking-silicon company NVIDIA acquired. So the NVIDIA tie isn't marketing: it's face-time inside closed NVIDIA rooms, next to the exact silicon people who'd matter for a mining protocol that lives and dies on GPU throughput. The "accelerated inference platform" pitch is being made in precisely the right rooms — and it lines up cleanly with the Inception badge, the Blackwell-kernel roadmap, and the founder's own ex-NVIDIA Research history already on this page. SOURCE: PUBLIC POSTS ON X — PEARL'S OWN VERIFIED ACCOUNT (@PRLNET) AND A PUBLIC REPLY THREAD, JULY 15 2026. NVIDIA INCEPTION CRYPTO EXCLUSION PER NVIDIA'S PUBLISHED ELIGIBILITY CRITERIA (CHECKED SEPTEMBER 21, 2026). CLOSED-EVENT APPEARANCE PER PEARL'S OWN PUBLIC POST; GABBAY / MELLANOX-NVIDIA PER PUBLIC ACADEMIC + CORPORATE RECORD. QUOTED AND DESCRIBED, NOT REHOSTED. ## The new pillar: verifiable compute for AI safety SEP 2026 The archive records a September 14 Attestable collaboration announcement. Its proposed combination of workload proofs and useful-work accounting adds a new question: what can an observer actually verify? “Pearl's proof of work together with attestable's ZK system is, to the best of our knowledge, the only way — with cryptographic guarantees — to prove that a given entity is actually running the computation that they say they're running.” — Idan Sugarman, co-founder · Sept 15 AMA. Framed around AI slowdowns and “compute treaties between countries.” The ambition is bigger than a benchmark. Proving a specified computation differs from accounting for all of an organization’s compute, and neither alone establishes safe model behavior. The AMA’s large-ZK-network description is a team statement without a defined comparative metric here. Future transaction privacy remains a separate proposal. ## The circuit Follow the project through research, infrastructure and crypto events. These are historical appearance records, not evidence of commercial agreements. THE APPEARANCE INDEX Follow the rooms. Open the record. 7 appearance records · Select a title to inspect it. 01Pearl summitPearl's own summitJul 2026+ Invite-only, multi-speaker. The keynote tape this whole case file rests on — and where the CEO described his own audience. APPEARANCE RECORD / A COMMERCIAL RELATIONSHIP REQUIRES SEPARATE EVIDENCE 02HardwareClosed NVIDIA eventJul 2026+ Omri Weinstein on stage with Prof. Freddy Gabbay (ex-VP Chip Design, Mellanox — acquired by NVIDIA), discussing “the intelligence economy.” Not a logo on a landing page: face-time inside a closed NVIDIA room. APPEARANCE RECORD / A COMMERCIAL RELATIONSHIP REQUIRES SEPARATE EVIDENCE 03ResearchMLSys 20262026+ Hawkeye presented — Published in the MLSys proceedings. Co-authored by Stanford's Dan Boneh and co-founder Ilan Komargodski. The paper is a public research artifact; it does not identify commercial counterparties. APPEARANCE RECORD / A COMMERCIAL RELATIONSHIP REQUIRES SEPARATE EVIDENCE 04AI industryRAISE AI Summit — ParisAttended+ Listed in Pearl's own roadmap post as an event the team participated in. APPEARANCE RECORD / A COMMERCIAL RELATIONSHIP REQUIRES SEPARATE EVIDENCE 05HardwareAMD Advancing AI — San FranciscoJul 2026+ Flagged as “attending… this week” in the 21 July roadmap post. Note the vendor: AMD, not only NVIDIA — the mining pitch is hardware-agnostic by design. APPEARANCE RECORD / A COMMERCIAL RELATIONSHIP REQUIRES SEPARATE EVIDENCE 06HardwareNVIDIA GTCNamed+ Cited among the events the team attends. Combined with the Inception badge and the closed-event appearance, that's three separate NVIDIA touchpoints. APPEARANCE RECORD / A COMMERCIAL RELATIONSHIP REQUIRES SEPARATE EVIDENCE 07ResearchESA 2026 — keynoteAug 31+ Omri gave an invited keynote at the European Symposium on Algorithms (ALGO 2026, L'Aquila): “Proofs of Useful Work from Arbitrary Matrix Multiplication” — the only consensus/PoW talk among the invited keynotes. The academic credibility isn't a costume; it's a podium. APPEARANCE RECORD / A COMMERCIAL RELATIONSHIP REQUIRES SEPARATE EVIDENCE SOURCE: PEARL'S OWN PUBLIC POSTS AND DISCORD ANNOUNCEMENTS, JULY 2026. ATTENDANCE AS STATED BY THE TEAM; WE HAVEN'T INDEPENDENTLY VERIFIED EACH APPEARANCE. NEW CONNECTION / AN OLDER ARTIFACT ## Before the partnership, there was a paper. Research edition / 22 September 2026 The trail starts before the announcement. In May 2025, Yogev Bar-On, Ilan Komargodski and Omri Weinstein appeared together as authors of research about mining incentives. Attestable now lists Bar-On as its CEO. That gives the later collaboration a documented research history. 27 MAY 2025 / AUTHORSHIP RECORD ### Bar-On × Komargodski × Weinstein A shared paper about external rewards and mining incentives. [Open the original artifact ↗](https://arxiv.org/abs/2505.21685) This establishes scholarly collaboration, not the date, terms or deployment of a commercial agreement. The interesting next question is how that shared research translates into a working system. [Attestable’s leadership ↗](https://attestable.com/company) · [Follow the research connections ↗](https://pearlified.com/team#research-lineage) TECHNICAL FILE / REPORTED RESULTS ## What can the proof actually promise? Research edition / 22 September 2026 An execution receipt has a precise job. Read the fine print with the headline. Attestable describes proving that committed model weights, an input and execution parameters produced a specified output. Its undated technical article reports alpha measurements; this is company evidence, not our independent reproduction. Reference hardware One H100 Reported verification 157–648 ms Current context limit 16K tokens Matrix multiplication Integer-quantized Fast verification is not fast proof generation. The article does not establish that the joint Pearl integration is deployed, or that a model behaves safely. Observed September 22; publication date not established. [Read the benchmark and limitations ↗](https://attestable.com/blog/proving-llms-scale) THE QUESTION WE’RE LEAVING OPEN ## How does the research become a working collaboration? Shared authorship and an announcement open the case. The integration itself is the next piece to look for. FOLLOW THIS CLUE [Follow the authors ↗](https://pearlified.com/team#research-lineage) See the documented research relationships behind the partnership story. BACK TO THE SOURCE [Hear the trust-layer pitch ↗](https://pearlified.com/reading#clip-ama-attestable-trust-layer) September 15 AMA · 15:03. Revisit the team’s description in context. Explore the whole file