Three structural bets in five business days
The Súper RIGI is Argentina's enhanced variant of its existing large-investment framework. Projects above the USD 1 billion threshold receive 30 years of fiscal stability, preferential foreign-exchange conditions, and exemptions from export duties. The regime explicitly covers artificial intelligence, data centers, semiconductors, lithium batteries, biotechnology, and the hydrogen industry — treating AI infrastructure as a category equivalent to oil wells and lithium mines, not as software. OpenAI's announced USD 25 billion Stargate Argentina project, a planned 500 MW computing campus in Patagonia, is already structured within this framework. The legislative signal is unambiguous: Argentina is offering AI infrastructure developers the same long-term certainty that resource-producing states have historically extended to extractive industry.
The Second Pax Silica Summit, convened in Washington in the week of June 23, brought Argentina, Chile, Costa Rica, Panama, Kazakhstan, and the European Union into the US-led initiative alongside existing members. The pact's architects have named the real chokepoint explicitly: it sits not at the chip layer but at the mineral layer. Argentina's lithium brine reserves are estimated at 20 million tonnes; Chile's at 9.3 million tonnes. The high-altitude salt flats of the Andean Lithium Triangle — straddling Argentina, Bolivia, and Chile — hold the largest known concentration of lithium brine deposits on Earth. The AI stack begins there. The US State Department's pilot of an AI supply chain credentialing platform in Panama is designed to create a verified provenance chain for AI-enabling minerals flowing through certified trade corridors, making origin and custody legible to customs authorities in the US and EU.
Brazil made a structurally comparable move through a different channel. The EU-Brazil Digital Partnership, formalized in June 2026, makes Brazil the first Global South country to align its AI governance architecture formally with the European Union's risk-based framework. Regulatory sandboxes, algorithmic transparency criteria, and cross-border data flow rules will now develop under a shared coordination mechanism. For Brazilian enterprises operating in both jurisdictions, this is a meaningful compliance simplification. At the geopolitical level it signals that the Global South is choosing governance alignment over regulatory arbitrage — a posture with long-run implications for where AI workloads can legally be processed and what data they can touch.
The enterprise counterpoint arrived in the same 72-hour window. KPMG's Q2 2026 Global AI Pulse, released June 24, surveyed 204 C-suite and senior leaders in organizations with annual revenue above USD 1 billion. Two-thirds have monitoring dashboards. Sixty-one percent have approval processes. But only 26% report full, real-time visibility into what their AI systems cost to operate. Only 36% have implemented token-level usage controls. Gartner's parallel projection is clarifying: AI agent software spending will reach USD 206.5 billion in 2026, a 139% single-year increase, while over 40% of agentic AI projects remain at risk of cancellation by 2027 precisely because operators cannot measure returns they cannot see.
So what: nations are instrumenting the substrate on a 30-year horizon; enterprise operators have not yet instrumented the model on a 30-day one. The gap between those two timescales is not a communications problem. It is an operating-model problem.
The Three-Layer Infrastructure Audit
The error in most enterprise AI conversations about geopolitical and supply-chain risk is one of category. The substrate beneath your AI stack has three distinct layers, each governed by different actors on different timescales, and most operational audits examine only the one that is closest to the model — and closest to the budget line.
Cost, reliability, and performance visibility. This layer includes token-level economics, agent orchestration costs, latency SLAs, and the decision-to-outcome measurement chain. KPMG's 26% real-time visibility figure is the diagnostic. If you cannot see this layer in real time, no governance framework above it will protect you from runaway cost or undetected quality degradation. The 36% token-level control rate is the leading indicator of future cost overruns in a world where AI agent software spending is growing 139% year on year.
Fiscal, regulatory, and sovereignty commitments made by governments. The Súper RIGI, the Brazil-EU Digital Partnership, and Pax Silica all operate here. These are 30-year signals, not quarterly adjustments. Enterprise operators who treat national policy as a background variable rather than an architecture input will find that compute availability, data residency requirements, and regulatory compliance burdens change beneath their production systems on timescales they did not model. Jurisdiction is not a compliance checkbox — it is a design parameter.
The physical origin of AI compute capacity. Lithium for battery-backed infrastructure, silicon for semiconductors, rare earth elements for power magnetics: the material substrate of AI compute is geographically concentrated and geopolitically contested. Pax Silica makes this explicit by treating the Andean lithium triangle as a supply-chain security asset on par with semiconductor fabs. The pilot AI supply chain credentialing platform in Panama creates verified provenance for AI-enabling minerals — a layer of traceability that will eventually surface in enterprise procurement, sustainability reporting, and vendor due diligence.
So what: the AI stack is not a cloud stack. It is a geopolitical stack. Auditing it requires looking at all three layers — not only the one you can bill to a cost center. KPIs before APIs, but also jurisdictions before APIs.
Three LATAM operating patterns
CABA Tier-1 bank — substrate instrumentation before vendor negotiation. A large retail bank in Buenos Aires is using the Súper RIGI passage as a catalyst to restructure its AI compute agreements. An internal AI operations audit, triggered by the KPMG Q2 findings shared through its Big Four advisor, identified that 58% of LLM inference costs were running uninstrumented: no per-transaction cost attribution, no latency logging, no override tracking against API gateways. Before approaching any Súper RIGI-eligible data center partner, the bank is establishing a cost-per-decision baseline across four high-value workflows — credit pre-screening, fraud detection, document processing, and customer-service deflection. Target: instrumentation ratio 100%, cost-per-decision delta ≥35% within 90 days, substitution latency for any single provider below 30 days. KPIs before APIs.
São Paulo industrial logistics — regulatory equivalence as a trade corridor. A third-party logistics operator serving EU-destined export flows is repositioning its AI governance stack in direct response to the EU-Brazil Digital Partnership. The partnership's shared regulatory sandbox and algorithmic transparency criteria allow LGPD-compliant document processing to be treated as functionally equivalent to EU AI Act Article 22 requirements for automated contractual decisions. The operator is consolidating AI document-processing workflows — currently split across three vendors with inconsistent auditability standards — into a single pipeline governed by a unified compliance report. Target: cross-border compliance overhead -42%, vendor concentration below 60%, decision auditability 100%, MTTR from audit trigger to resolution under 48 hours.
Argentine grain exporter — Pax Silica as a trade credentialing layer. A 14-port Argentine grain export network is piloting the US State Department's Pax Silica AI supply chain credentialing platform, first deployed in Panama. The platform verifies AI-augmented shipping manifests and cargo provenance data against a certified supply-chain ledger, enabling US and EU customs authorities to process certified shipments on an accelerated review track. The exporter's AI-augmented forecasting system currently covers 62% of outbound routes; coverage is being expanded to 85% specifically to qualify for Pax Silica credentialing. Target: customs clearance time -68% on credentialed routes, OTIF improvement +8.4 percentage points, cargo documentation error rate below 0.5%.
Audit all three layers — not just the model
The operating-model discipline this week's events demand is not geopolitical intelligence. It is infrastructure hygiene applied across a stack that now runs from lithium brine deposits in Patagonia to LLM inference costs on a cloud bill. Most enterprises have audited Layer 1 incompletely (the KPMG data proves it), have not mapped Layer 2 at all, and have never considered Layer 3 as an operational variable. The sequence matters.
So what: from pilot to policy means designing AI systems that survive not just the next model release but the next regulatory cycle, the next infrastructure investment regime, and the next supply-chain credentialing mandate. All three moved this week. They will move again.
Governance
Map all AI workloads against the three infrastructure layers — not just the model and vendor, but the national regulatory jurisdiction and the supply-chain provenance of compute capacity. For operations under Ley 25.326 (Argentina) or LGPD (Brazil), confirm that data residency and decision-auditability obligations remain compatible with the substrates you depend on. For Pax Silica-adjacent supply chains, initiate the credentialing conversation with trade corridor partners before customs authorities implement automated verification requirements. Include AI-infrastructure vendor agreements in your legal review of Súper RIGI investment proposals.
KPIs
Enterprise layer: real-time cost-visibility ratio ≥95% of production workloads; token-level cost attribution 100%; substitution latency below 30 days for any single provider. National layer: regulatory-equivalence coverage ≥80% for cross-border AI-augmented data flows; data-residency compliance ratio 100%. Mineral layer: provider concentration below 60% per substrate; sovereign-substrate coverage ≥30% on regulated workloads; supply-chain credentialing coverage for Pax Silica-eligible trade flows.
12-month roadmap
0–90 days (Layer 1): instrument the enterprise layer — establish cost-per-decision baselines, tag all AI workloads by risk tier and data-residency jurisdiction, and achieve token-level cost attribution across every production system. 90–180 days (Layer 2): map the national layer — review AI infrastructure vendor agreements against Súper RIGI, LGPD-EU equivalence, and Pax Silica timelines; identify substitutability risks; insert data-residency and exit clauses into renewals. 180–360 days (Layer 3): connect the mineral layer — begin Pax Silica credentialing where applicable; integrate a second substrate provider for regulated workloads; establish quarterly board-level reporting across all three infrastructure layers.
The substrate question is an operating-model question.
For most of the past three years, the enterprise AI conversation has been organized around model capability and agent deployment velocity. The Súper RIGI, Pax Silica, and the EU-Brazil Digital Partnership are reminders that the stack those agents run on is not neutral infrastructure — it is the product of decisions made at the national and geopolitical level, on timescales that most enterprise operating models do not track.
KPMG's 26% real-time cost visibility figure is the symptom. The disease is a conceptual frame that treats AI infrastructure as a software problem — flexible, modular, swappable at will — rather than as the infrastructure problem that three sovereign bets in five days confirm it to be. Thirty-year fiscal stability frameworks and supply-chain credentialing alliances do not get built around software. They get built around infrastructure.
The Socradata view is operational: the Three-Layer Infrastructure Audit is not a geopolitical exercise. It is the same discipline that sensible procurement functions applied to cloud concentration risk in 2018 and semiconductor supply chains in 2021. The lesson from those episodes was consistent: operators who mapped their dependency stack early had options. Operators who discovered the dependency after a disruption did not. The mineral, national, and enterprise layers of the AI stack are converging on a moment of similar legibility. This week's events are the signal.
Request an Operational Diagnostic
Socradata runs Three-Layer Infrastructure Audits for LATAM operators in financial services, logistics, agribusiness, and public-sector AI. The output is a workload-by-layer dependency map, a cost-visibility scorecard, a regulatory equivalence assessment, and a 90-day next-step plan aligned to the Súper RIGI, Pax Silica, and LGPD/EU AI Act governance timelines.