Phase A — Understand the business
Lens 1 · Company Overview
VAST Data sells software that stores and serves the data behind AI — the persistence, retrieval, and increasingly the compute layer that sits underneath GPU clusters. Founded January 2016, launched from stealth February 2019. HQ New York City with the engineering core in Israel; ~1,000 employees (2025).
The product has evolved from a storage system into what VAST now brands the VAST AI Operating System, a stack of composable services on one architecture:
- DataStore — exabyte-scale unstructured (file + object) storage.
- DataBase — a unified database / warehouse / data-lake with a native vector store.
- DataEngine — a serverless function-execution fabric ("AWS Lambda for data"), schedulable across CPU / GPU / DPU.
- InsightEngine — real-time RAG built with Nvidia: event-triggered chunk → embed → store the moment data lands.
- AgentEngine — production deployment/orchestration for long-running AI agents (state, lifecycle, audit).
- DataSpace (global namespace) + SyncEngine (ingest/catalog) + Polaris control plane (manage geo-distributed fleets across on-prem, neocloud, and public cloud).
Business model: B2B infrastructure software, subscription priced on raw capacity deployed; VAST is software-only and media-agnostic — customers (or OEMs) buy the commodity hardware. The pitch is consolidation: displace separate storage + warehouse + ETL + vector DB + serverless compute with one platform, at all-flash performance and near-HDD economics.
Customers / suppliers / competitors are mapped in Lenses 2–3 and 5. In one line: buyers are neoclouds, AI labs, sovereigns, and federal; the critical supplier/partner is Nvidia (simultaneously chip vendor, investor, and channel); rivals are Weka, DDN, Pure Storage, NetApp, Dell, and the hyperscalers' own end-to-end stacks.
Contract structure / concentration: large multi-year commitments — the CoreWeave deal alone is $1.17B (Nov 2025); three customers each represent >$100M in total commitments, and the top-100 new customers average ~$1.2M each. Recurring + committed, but lumpy and concentrated — the defining tension of the whole business.
Lens 2 · Supply Chain
VAST is a software company, so its "supply chain" is the hardware+silicon stack its software runs on and the buyers it sells through. Named, end to end:
Upstream (silicon / media):
- QLC NAND flash — VAST's foundational bet was running the cheapest, densest flash (QLC) at enterprise reliability via software, undercutting HDD economics. QLC NVMe SSD supply comes from the NAND oligopoly: Solidigm (SK Hynix), Micron, Kioxia, Samsung. This ties VAST's customer economics to the NAND price cycle — a 2025–26 memory upcycle raises the cost of the flash its story depends on.
- Nvidia BlueField DPUs (‑3, now ‑4) — VAST's OS now runs natively on the DPU, offloading storage/data services onto Nvidia's networking silicon. GPUDirect Storage gives GPUs a direct path to VAST data.
- Fabric: low-latency Ethernet / InfiniBand (NVMe-oF) — Nvidia (Spectrum-X / Quantum InfiniBand), Arista, Cisco.
The company (software) + hardware assembly:
- Hardware is built by server ODMs/OEMs: confirmed partners include Cisco (UCS), plus Supermicro/Dell-class OEMs and white-box ODMs. VAST ships software; the "several billion dollars" of servers/storage/networking behind a deal like CoreWeave's is bought from OEMs, not from VAST.
Downstream (buyers):
- Neoclouds: CoreWeave (largest), Lambda, Crusoe, Core42, Sharon AI, Mistral Compute.
- Frontier labs: xAI Colossus — 100,000+ H100 GPUs in Memphis, "well over an exabyte" of VAST flash, handling Grok training-token storage + checkpointing.
- Sovereign AI: G42 (UAE), Humain (Saudi).
- Federal: U.S. Air Force; DoD Joint Warfighting Cloud Capability (JWCC) vehicle added 2025.
Chokepoints / single-source dependencies:
- Nvidia is the chokepoint — supplier (DPU, GPUDirect, fabric), investor (Series E + F), and channel (NVIDIA Cloud Partner certification, DGX SuperPOD reference architecture). No other relationship is remotely as load-bearing.
- NAND supply / price — the QLC cost advantage is exposed to the memory cycle.
- Neocloud concentration — the biggest buyers (CoreWeave, xAI) are themselves Nvidia-financed and debt-financed; VAST inherits their capital-cycle risk.
Lens 3 · Competitive Advantages (moats)
The architecture is the moat — DASE (Disaggregated Shared-Everything). VAST separates stateless compute nodes (CNodes) from flash enclosures (DNodes) across an NVMe-oF fabric, so any server sees all data and metadata directly. This dissolves the classic storage trade-off: you scale capacity (add DNodes) and performance (add CNodes) independently, and the system scales linearly from terabytes to exabytes without the sharding/coordination tax of shared-nothing designs. Combined with similarity-based global data reduction and QLC-at-HDD-economics, DASE is a genuine architectural edge that shared-nothing incumbents cannot bolt on.
Durable moats:
- Architecture + IP — DASE is patented and hard to copy without a ground-up rewrite; incumbents (NetApp, Dell, Pure's older lines) carry legacy shared-nothing baggage.
- Switching costs / data gravity — once exabytes of training data, checkpoints, and a vector store live in VAST inside a 100k-GPU cluster, migration is effectively a re-platforming. This is real lock-in and it compounds with cluster size.
- The Nvidia flywheel — first NFS storage certified for DGX SuperPOD (2023), NCP-certified, DPU-native, co-engineered InsightEngine. Being inside Nvidia's reference stack is a distribution moat competitors must each fight for.
- Platform breadth — the DataBase + DataEngine + Agent/InsightEngine layers move VAST up the stack from "storage" to "AI data OS." If the AgentEngine/InsightEngine narrative lands, the moat widens from data gravity to workflow gravity.
Bargaining power: strong over suppliers (media-agnostic, plays QLC vendors off each other; hardware is commoditized under it). Weaker over its largest customers — a CoreWeave or xAI is a >$100M account with the leverage of scale and credible alternatives (Weka, or building in-house). Net: pricing power over the stack, but concentration erodes it at the top of the customer list.
Lens 4 · Segments
segments.csv is empty — VAST is private and does not break out revenue by product or geography. ``. No audited segmentation exists; do not fabricate one. What is directionally disclosed:
- By product line: storage (DataStore) remains the revenue base; the DataBase / DataEngine / InsightEngine / AgentEngine layers are the forward growth vector but are early — the "AI OS" is more roadmap-and-narrative than booked revenue today.
- By customer type: neoclouds dominate the bookings mix (CoreWeave, Lambda, Crusoe, Core42, xAI). Enterprise + sovereign + federal are the diversification story management is actively pushing (98+ named customers).
- By geography: U.S.-centric with a fast-growing Gulf sovereign pipeline (G42, Humain).
- Trend: accelerating — revenue tripling YoY — but the mix is the risk: heavy on a handful of capital-cycle-exposed neoclouds. Requirement met honestly: no segment number is asserted as fact; all figures are web-derived and unaudited.
Phase B — Measure performance
+private overlay: Lens 5 → Funding & valuation trajectory; Lens 7 → Cap table & secondary marks (+comps); Lens 8 → funding/product catalysts. "Traction & unit economics" added.
Lens 5 · Funding & Valuation Trajectory (private swap)
Full round history — all ``, unaudited:
| Round | Date | Amount | Post-$ valuation | Lead(s) |
|---|
| A | 2016 | $15M | — | 83North, Norwest |
| A1 | 2018 | $25M | — | Dell Technologies Capital, 83North, Norwest |
| B | 2019 | $40M | — | Greenfield Partners, 83North, Norwest |
| C | 2020 | $100M | $1.2B | Next47 |
| D | 2021 | $83M | $3.7B | Tiger Global |
| E | 2023-12 | $118M | $9.1B | Fidelity (w/ NEA, BOND, Drive, + Nvidia, Dell, Goldman, Tiger) |
| F | 2026-04-22 | ~$1.0B | $30B | Drive Capital + Access Industries (w/ Fidelity, NEA, Nvidia) |
Total raised pre-F ≈ $381M — strikingly little for the scale reached, consistent with the cash-generative claim. The Series F is the headline:
- ~$1B at $30B post-money — >3× the $9.1B Series E mark in ~28 months.
- >$500M is secondary (over half the round) — a liquidity tender for employees + early investors, explicitly framed as IPO on-ramp, and a tell that a listing is 12–18 months out, not imminent.
- Nvidia participated again; CapitalG (Alphabet) reported as a possible co-lead but not consistently confirmed — flag as unverified.
Disclosed financials at Series F (FY2025 close):
- >$4B cumulative bookings; >$500M committed ARR (CARR).
- Profitable and cash-flow positive — reportedly >$100M cash generated per quarter, positive operating margin, gross margin ~90%. Profitability at triple-digit growth is genuinely rare.
- Company-cited "Rule of X" score of 228% (growth + FCF margin).
Lens 6 · Founder & Management Narrative (calls swap → interviews/keynotes)
No earnings calls exist. The equivalent signal is Renen Hallak's public thesis, sharpened at VAST Forward 2026 (the company's first global customer event, Feb–Mar 2026):
- "The AI era is defined not by models but by the operating system" that manages the data, memory, identity, and coordination of billions of agents — collapsing orchestration + scheduling + security into one surface.
- "Thinking machines" — perpetual/recursive learning systems that need continuous, adaptive, safe infrastructure.
- Three pillars forming inside VAST: unstructured data (DataStore) → compute (DataEngine/AI engine) → a kernel-like control plane. The explicit bet: turn a decade of vertically-integrated data infrastructure into an "AI Operating System" and make the narrative real in 2026+.
Tone trend: consistent and escalating in ambition — from "fastest-growing storage company" (2019–23) to "the AI OS / infrastructure of intelligence" (2025–26). The rhetoric has outrun the booked revenue of the newer layers; execution on DataBase/Agent/InsightEngine is the thing to watch. Founder-led, technically credible, high-conviction — closer to a platform founder's arc than a storage-vendor CEO's.
Lens 7 · Cap Table, Secondary Marks & Comps (private swap)
Syndicate quality — high, with strong IPO-proximity tells:
- Crossover / mutual-fund anchors: Fidelity (Series E lead, Series F participant), Tiger Global, BOND (Mary Meeker) — the Fidelity anchor is a classic pre-IPO signal.
- Strategics: Nvidia (E + F — the most important name on the sheet), Dell, Goldman Sachs.
- Series F leads: Drive Capital + Access Industries (Len Blavatnik) — deep-pocketed, late-stage.
- Early VCs: 83North, Norwest, Next47, Greenfield.
- Secondary marks: the >$500M Series F secondary tender clears at the $30B mark — a live secondary print, not just a primary paper valuation.
Comps table — provenance-critical; multiples are arithmetic on inputs, n/a where I can't source cleanly:
| Company | Status | Valuation / Mkt cap | Revenue / ARR | Implied multiple |
|---|
| VAST Data | private | $30B | CARR >$500M; ARR ~$200M (Jan-25) → ~$600M proj. 2026 | ~60x CARR / ~50–150x ARR |
| Weka | private | $1.6B | ~$100M ARR (Jun-25) | ~16x ARR |
| DDN | private | ~$5B reported (Blackstone, 2024) | not disclosed | n/a |
| Pure Storage (PSTG) | public | ~$26B | ~$3.63B rev FY26, +~15% | ~7x P/S; P/E n/a |
| NetApp (NTAP) | public | ~$30B (range $28–33B) | ~$6.8B rev run-rate (Q2 FY26 $1.71B ×4) | ~4.4x P/S; ~+3% growth |
The crux: VAST is marked at ~50–150x ARR, versus its closest private peer Weka at ~16x, and public all-flash storage at ~4–7x sales. Even discounting for hypergrowth + profitability, this is an extreme premium that prices multi-year flawless execution and continued AI-capex expansion. This single row is the whole valuation debate.
Lens 8 · Catalysts (price-catalysts swap → funding/product events)
No stock, so the "what moves the mark" events are funding + product + customer milestones:
- Valuation step-ups: $1.2B (2020) → $3.7B (2021) → $9.1B (2023) → $30B (2026). Each re-rate tracked an AI-capex inflection, not a storage cycle.
- Customer landmarks: CoreWeave partnership (Sep 2023) → CoreWeave $1.17B (Nov 2025, became Customer #1, surpassing xAI) → xAI Colossus exabyte deployment → DoD JWCC (2025).
- Product/partnership: DGX SuperPOD certification (2023) → InsightEngine w/ Nvidia (2024) → DPU-native OS on BlueField-3/-4 + Polaris control plane (2026) → AgentEngine (2026).
- Governance/IPO prep: hired ex-Shopify CFO Amy Shapero (~late 2024) → Series F secondary tender (2026) → stated IPO-readiness by end-2026.
- Pattern: the mark reacts to Nvidia-ecosystem proximity and marquee neocloud wins — the same two variables that are also the largest concentration risks.
Traction & unit economics (private add)
- Revenue run-rate >$1B, "possibly much higher," triple-digit growth, margins ">50%" — one analyst read; company-disclosed gross margin ~90%.
- Conflicting ARR figures — surfaced, not reconciled: company-disclosed >$500M CARR (FY25 close) vs Sacra's ~$200M ARR (Jan-25) → ~$600M projected 2026 vs a loose "$2B total ARR incl. non-committed" claim that appears unverified/likely conflated. CARR (~$500M) is the most defensible anchor; the $2B figure should be treated skeptically.
- Storage-attach economics (the key unit-econ variable): VAST's co-founder pegs storage at 3–5% of neocloud spend vs IDC's 1.9% forecast for 2029. VAST's entire TAM math assumes attach ~2× IDC's base case — a bull assumption to stress-test.
Phase C — Judge people & books
Lens 9 · Management
- Renen Hallak (Founder/CEO) — the single most important asset. First engineer / VP of R&D at XtremIO, which EMC bought for $430M (2012) and which reached $1.2B revenue in its first year of sales — the fastest-growing enterprise-storage product in history. He has literally done this before: architected a category-defining all-flash system, then went and built a bigger one. Technical founder, high conviction, platform-scale ambition.
- Co-founders: Jeff Denworth (ex-CTERA; marketing/GTM), Shachar Fienblit (ex-Kaminario; storage veteran), Alon Horev (CTO). Michael Wing (ex-Dell EMC) president. A tight, storage-native founding team with EMC/XtremIO/Kaminario pedigree — deep domain credibility.
- CFO Amy Shapero — ex-Shopify CFO, hired ~late 2024, an unambiguous IPO-readiness hire.
- Skin in the game: founder + early-employee ownership is large; the Series F secondary lets them take some chips off the table (>$500M) without a forced exit — aligned, not cashing out.
- Capital allocation: disciplined — reached ~$30B on only ~$381M primary raised, staying cash-generative. The one aggressive move is the Red Stapler acqui-hire (Sep 2025) now under litigation (Lens 10) — a flag on M&A diligence.
- Archetype: repeat technical founder in build-the-platform mode. For this stage (pre-IPO hypergrowth) that is the right archetype — the risk is ambition (the "AI OS" narrative) outrunning execution.
Lens 10 · Forensic Red Flags + Regulatory
Accounting caveat: VAST is private and unaudited per public sources — no income statement, balance sheet, or cash-flow statement is available, so classic forensic ratio work (receivables vs revenue, SBC add-backs, cash-vs-earnings divergence) cannot be performed. That opacity is itself the flag: every profitability/ARR figure is management-asserted. Specific structural concerns:
- Bookings vs revenue: ">$4B cumulative bookings" and "CARR" are the disclosed metrics — bookings are not revenue, and multi-year committed contracts (CoreWeave $1.17B over ~5yrs) inflate the headline vs recognized/collected cash. Watch for the gap at S-1.
- Customer concentration: three customers >$100M each; CoreWeave Customer #1 (reportedly <25% of revenue, but the single largest) — a concentration and receivables-quality risk, especially as the top buyers are themselves debt/Nvidia-financed.
- Related-party / circular-financing optics: Nvidia is investor and supplier and channel; VAST's largest customers (CoreWeave, neoclouds) are also Nvidia-backed. No wrongdoing implied, but it is the same circular-AI-financing pattern regulators and short-sellers are scrutinizing across the neocloud complex — an S-1 disclosure item.
Regulatory findings (regulatory/regulatory-findings.md): 0 SEC findings — no CIK, private, no EDGAR filings to search (LR/AAER n/a). Non-SEC web search results:
- Active litigation — NetApp v. former executives (Red Stapler): NetApp filed 2025-11-17 (California Superior Court) alleging former exec Sigurd Stefánsson and co-founder Björn Hrafnsson misappropriated trade secrets while still employed/paid at NetApp, built Red Stapler (GitHub "redstapler-is" traced to 2025-06-16, days before Stefánsson's departure), incorporated it 2025-07-03, and sold it to VAST Data on 2025-09-09 (~10 weeks post-incorporation). NetApp obtained a restraining order limiting Stefánsson's VAST role; status is contested/evolving (reports of partial dismissal and possible settlement). VAST is not the direct defendant, but the target of its acqui-hire is — an M&A-diligence and reputational flag heading into an IPO.
- No other material FTC/DOJ/FDA/enforcement findings surfaced via web search as of 2026-07-10.
Net: no accounting scandal, but unauditable-by-outsiders + one live IP suit tied to an acquisition + circular-financing optics — the three things a forensic analyst flags for the S-1.
Phase D — Project & stress-test
+private overlay: Lens 11 → IPO-readiness & path-to-tradeable.
Lens 11 · IPO-Readiness & Path-to-Tradeable (private swap)
No EPS projection is possible (private, no share/financial data) — the be-early question is "when does this become tradeable, and at what setup?"
Readiness assessment (scale 1–5, per private-watch schema): 4 — pre-IPO / secondary-active.
- Milestones already cleared for an S-1: profitable + cash-generative; >$500M CARR / >$4B bookings; IPO-grade CFO (Shapero); crossover investors (Fidelity) on the cap table; a live secondary market clearing at $30B.
- Stated window: CEO says IPO-ready by end-2026, then "maybe next year, maybe the year after"; The Information: H2 2026 or later. The secondary-heavy Series F implies ~12–18 months out — call it 2027 base case, with a late-2026 tape-permitting option.
- What still must happen: audited financials + revenue-recognition cleanup (bookings→GAAP revenue), demonstrate the neocloud concentration is diversifying, and clear/settle the NetApp overhang. Macro/AI-capex sentiment at listing is the swing factor.
- Path to tradeable for a MenFem be-early position: (a) pre-IPO secondary at ~$30B (accredited/fund access only), or (b) the IPO itself (2026–27), or (c) the read-through basket — own the listed beneficiaries (CoreWeave, Nvidia, and public storage comps Pure/NetApp as sentiment proxies) until VAST is tradeable.
Do NOT log a Brier EPS forecast (no EPS; unattended --watchlist skips forecast.ts create). The one scoreable binary worth tracking conversationally: "VAST Data files an S-1 (or completes a direct-listing/IPO) before 2027-12-31" — base-case ~55–60%. Open item: add a vast-data entry to research/private-watch.json (beat: datacenters, stage: pre-ipo, ipo_readiness: 4, lead_investors: "Drive Capital, Access Industries, Nvidia, Fidelity, NEA", catalyst: "$30B Series F Apr-2026; IPO-ready end-2026 per CEO; secondary-heavy tender", dossier: this file) — left for the master session to write back (out of this wave's boundary).
Lens 12 · Bull vs Bear
Bull case. VAST is becoming the default data-and-memory layer of the AI build-out. DASE is a real architectural moat that legacy vendors can't retrofit; data gravity + checkpoint/vector lock-in inside 100k-GPU clusters compounds with cluster size. It is Nvidia's blessed storage/data partner (DPU-native, NCP-certified, co-engineered RAG) at the exact moment inference/agentic workloads make the data layer the bottleneck. The company is doing something almost no hypergrowth infra name does — printing cash (~$100M/qtr) at triple-digit growth, ~90% gross margin. The "AI OS" (DataBase + DataEngine + Agent/InsightEngine) is a credible path to expand from storage TAM into the far larger data-platform + compute-orchestration TAM, widening the moat from data gravity to workflow gravity. TAM tailwind is enormous — AI-powered storage ~$30–36B (2025–26) → ~$118B by 2030 at ~26% CAGR, and that's before the up-stack expansion.
Bear case (permanent-impairment risks).
- The multiple is the risk. ~$30B at ~50–150x ARR vs Weka ~16x vs public storage ~4–7x. Any growth stumble or AI-capex pause re-rates this violently; a public-market IPO will be marked against Pure/NetApp comps, not private-round comps.
- Storage-attach reversion. VAST's TAM assumes 3–5% of neocloud spend goes to storage vs IDC's 1.9% for 2029. If attach normalizes toward IDC, the growth math compresses by ~half.
- Concentration + circularity. CoreWeave/xAI/neoclouds dominate bookings and are themselves capital-cycle-exposed and Nvidia-financed. A neocloud funding winter hits VAST's biggest customers first.
Pre-mortem (18 months out, thesis broke): AI-capex growth decelerated; a marquee neocloud (or xAI) renegotiated/slipped a large commitment; the IPO priced down to public-storage multiples (~10–15x forward ARR at best), the $30B mark looked like the peak, and the newer "AI OS" layers hadn't converted to material revenue — leaving VAST re-rated as a very good, very fast storage company rather than the operating system of AI.
Contrarian view (what the market refuses to see): bulls extrapolate the storage line; the real optionality — and the real risk — is the compute/agent layer. If AgentEngine/InsightEngine become where enterprises actually run agentic AI, VAST is radically undervalued at $30B. If they stall as marketing while hyperscalers ship end-to-end stacks, the storage business alone cannot justify a growth-software multiple. The bet is not on storage; it's on whether "AI OS" is real.
Lens 13 · Devil's Advocate (short-seller)
Dismantling the bull case:
- Revenue is concentrated in a handful of Nvidia-financed neoclouds whose own economics are unproven; if CoreWeave/xAI capex growth slows, VAST's bookings curve bends first, and the "committed" ARR tests how enforceable multi-year neocloud commitments really are in a downturn.
- The moat may be narrower than bulls think. Weka already beats VAST on specific GPU-throughput/KV-cache benchmarks; Pure's FlashBlade//EXA (10+ TB/s) and DDN target the same certifications; hyperscalers (AWS/Azure/GCP) and Dell's AI Factory ship end-to-end stacks that make third-party storage a line item, not a platform. Software-defined storage has always commoditized.
- The "AI OS" is largely narrative. DataBase/DataEngine/Agent/InsightEngine revenue is early; the durable business today is still fast QLC storage. Paying a ~50–150x-ARR mark for a roadmap is the whole short.
- Governance flag: the Red Stapler acqui-hire is under a NetApp trade-secret suit with a restraining order — aggressive talent M&A that could recur, and a live liability into an IPO.
- Circular financing: Nvidia as investor + supplier + channel + backer-of-your-customers is exactly the pattern that unwinds ugly if the AI-capex cycle turns.
- If growth disappoints 20–30%: at these multiples the mark could halve or worse on the IPO tape; the $30B secondary could prove to be the last, best price.
- Single scenario that permanently impairs: a broad neocloud/AI-capex retrenchment (2027-style) that simultaneously slows the largest customers, resets storage-attach toward IDC's 1.9%, and forces an IPO into a hostile tape — turning $30B into the high-water mark.
Lens 14 · Management Questions (ordered by information value)
- What is GAAP revenue (not bookings/CARR) for the last two fiscal years, and what is the bookings-to-revenue conversion curve? (Cuts straight to the metric the IPO will be judged on.)
- What share of ARR is the top 1 / top 3 / top 10 customers, and how has concentration trended over the last 8 quarters?
- What is your assumed storage-attach rate as a % of customer AI spend, and how do you reconcile 3–5% with IDC's ~1.9% (2029)?
- Of the "AI OS" layers (DataBase, DataEngine, InsightEngine, AgentEngine), what is real recurring revenue today vs storage, and what's the attach rate?
- How enforceable are the multi-year neocloud commitments (e.g., CoreWeave's $1.17B) in a customer downturn — take-or-pay, or consumption-based?
- What is net revenue retention, and what does it look like excluding the top 3 accounts?
- What is your exposure to the NAND price cycle, and how does a memory upcycle affect customer TCO vs HDD/hyperscaler alternatives?
- Walk through the Nvidia relationship as investor + supplier + channel — where are the conflicts, and what happens to you if Nvidia ships or blesses a competing data layer?
- What is the status and maximum liability of the NetApp/Red Stapler litigation, and what diligence changes has it prompted?
- Against hyperscaler end-to-end AI stacks and Dell AI Factory, what is the durable reason a large enterprise chooses VAST over the incumbent it already runs?
- What is your cash-flow conversion (FCF/revenue) and how much is flattered by upfront multi-year prepayments?
- What is the realistic IPO window and the comp set you expect to be valued against — private AI infra, or public Pure/NetApp?
- How defensible is DASE's IP against Weka's throughput edge and Pure/DDN's certified AI lines — where do you lose bake-offs today?
- What is headcount + burn trajectory as you build the compute/agent layers, and does profitability hold through that investment?
- What is your sovereign-AI (G42/Humain) revenue and geopolitical/export exposure, and how concentrated is it?