Autoresearch: As of 19 Sep 2026, have enterprises shifted real workloads to open-weight models on hardware they own? Did NVDA’s enterprise/ACIE mix broaden on that mechanism? Did Palantir disclose Nemotron/sovereign deployments in dollars? Did Anthropic/OpenAI enterprise API stop compounding?

Generated by /autoresearch on 2026-09-19. Synthesized across 3 rounds from issuer SEC/IR primaries plus qualified secondary revenue prints. See Provenance. Treat as raw material. Context: vault/projects/stock-market Research-only. No buy/sell/size.

Summary

Step 4 remains unconfirmed. No retrieved issuer primary shows enterprises moving material production workloads from Anthropic/OpenAI APIs onto owned-weight open models. NVIDIA’s Q2 FY2027 ACIE line was $40.313 billion (+25% QoQ, +138% YoY) and “sovereign AI” grew 35% sequentially — but ACIE includes neoclouds and hyperscalers buying cloud capacity, and the company recast a name from ACIE into Hyperscale. Palantir Q2 US government revenue was $809 million (+90% YoY); the June Nemotron engine named no agencies and no contract value. Secondary WSJ-recap prints still have Anthropic/OpenAI enterprise API compounding (qualify: not issuer 10-Q). The Step 2 “API keeps compounding” falsifier is leaning against a trust-break exodus, not for it.

Findings

NVDA ACIE grew — it is not an on-prem open-weight census

NVIDIA CFO commentary for Q2 FY2027 (quarter ended ~27 Jul 2026): Data Center $89,023 million; Hyperscale $48,710 million (55%); ACIE $40,313 million (45%), +25% sequential and +138% YoY (NVDA Q2 FY27 CFO commentary, SEC). “ACIE revenue increased 138% from a year ago and 25% sequentially driven by end-demand from AI natives, enterprises, and sovereign customers, as well as hyperscalers utilizing AI clouds.” The same exhibit: “we reclassified a company from AI Clouds, Industrial, & Enterprise (ACIE) to Hyperscale … and recast the prior period.” Sovereign AI is spoken as growing “35% sequentially and more than tripled year over year” on the call reprint (NVDA Q2 FY27 transcript reprint). That is buyer-class broadening inside Data Center, not a measured shift of enterprise tokens from Claude/GPT onto owned Nemotron weights.

Nemotron is listed among models NVIDIA develops on its own leased capacity — a product, not a customer-adoption KPI (same CFO commentary).

Palantir: sovereign pitch is real; Nemotron dollars are not

Palantir Q2 2026 Ex. 99.1: US government revenue $809 million, +90% YoY and +18% QoQ; US commercial $764 million, +149% YoY; total $1.935 billion (PLTR Q2 2026 Ex. 99.1). Karp: “Demand for AI sovereignty has now been unleashed… Their competitive advantage should never become the training data for future models.” The June 29 NVIDIA blog / Palantir IR pair announce an engine to run customized Nemotron in air-gapped environments and retain weights (NVIDIA blog 29 Jun 2026; Palantir IR news page). No agency names, no TCV, no “Nemotron ACV” line. Step 3’s concrete partnership exists as a product; Step 3’s 12-month revenue falsifier is not yet resolvable from filings.

Frontier-lab API: secondary prints still show compounding, not an exodus

WSJ-recap secondaries (not issuer financials) put Anthropic Q2 2026 revenue around $11.5–11.6 billion and OpenAI around $6.7 billion, with Anthropic described as enterprise/API/Claude Code-heavy (Value Add Pulse; XenoSpectrum WSJ recap). Friar is quoted that enterprise has overtaken consumer at OpenAI. These are not 10-Qs. They are directionally the opposite of “enterprises refuse and leave.” Menlo-style API-share surveys in tertiary recaps are not used as load-bearing numbers here.

Existing-chain tension stands: enterprise-token-budgeting still reports whale API spend growing. This pass did not retrieve a Bedrock/segment print that shows an on-prem substitution.

Contradictions and open questions

• ACIE 45% of Data Center looks like “enterprise broadening” until the footnote (neoclouds + hyperscaler cloud buy + one-name recast) is applied.
• Karp’s sovereignty language is first-party and interested. US gov +90% can be AIP/Gotham/Foundry without any open-weight migration.
• If WSJ Anthropic $11.6B is even roughly right, Step 2’s trust-break is not visible in lab revenue. Keep partial.
• Cost-gap (16.4×) remains Chamath-only; not re-litigated.

Provenance

Rounds run: 3 of 3. X sources: attempted; spend-cap 403. Not used. Grokipedia: not used for 2026 claims.

Web sources:

• NVDA Q2 FY27 CFO commentary — official / SEC — DC/Hyperscale/ACIE table; recast; Nemotron as internal model.
• NVDA Q2 FY27 transcript reprint — reprint — ACIE + sovereign sequential color.
• PLTR Q2 2026 Ex. 99.1 — official / SEC — $809M US gov; Karp sovereignty quote.
• NVIDIA Nemotron/Palantir blog — issuer — product, no dollars.
• Palantir IR Nemotron launch — issuer page (thin HTML).
• Value Add Pulse / XenoSpectrum — secondary WSJ recaps; not used as confirmed ARR.

Generated: 2026-09-19