Argentina and Paraguay’s
AI Infrastructure Bet:
From Energy to Compute

Market analysis framework

The market signal has changed

The useful question is no longer whether Argentina and Paraguay want AI data centers. Both clearly do. The market question is now which announced capacity is acquiring the characteristics of an investable project: a site, a power route, a legal structure, financing, procurement responsibility, cooling and fiber design, an operator and a customer.

That distinction matters because AI infrastructure can absorb enormous capital while still remaining optional for years. A political announcement may create visibility. A dedicated tariff or incentive can improve economics, but only while the rule remains in force. Site control, grid studies, equipment orders, permits, financing and binding offtake move the project into a different category again.

Paraguay now has two AI infrastructure tracks

Paraguay’s first track is institutional. The agreement with Taiwan is no longer only a broad technology memorandum. MITIC has described a concrete first stage: 10 MW of compute, a 50/50 model and a phase-one investment estimate of US$300 million to US$500 million. A dedicated binational entity is planned, and the first stage is expected to use the site of Paraguay’s future state data center.

This creates a different type of demand from a private hyperscale project. The government-led model can generate procurement around data-center construction, electrical systems, cooling, networking, cybersecurity, high-performance computing, operations and public-sector AI capacity. The decisive next steps are the legal structure, budget and financing, technical architecture, procurement route and project calendar.

The second track is HIVE. The company has already demonstrated that Paraguay can host large hydro-powered, power-intensive digital infrastructure. It completed a 300 MW Paraguay buildout across Yguazú and Valenzuela, primarily for Bitcoin-mining operations, and is now adding another 100 MW at Yguazú. That makes HIVE relevant to the AI discussion because the power and site infrastructure already exist, while the workload transition remains separate.

But the August disclosure changes how the project should be read. HIVE says the extra 100 MW remains unallocated. Its August disclosure targets energization in Q4 2026 and leaves the capacity available for hashrate services, AI or HPC depending on returns. A separate June roadmap places Tier-III AI/HPC construction after energization and targets ready-for-service in H2 2027. Paraguay therefore has a credible powered-site option for AI, but the full 100 MW is not committed AI capacity.

The power-policy layer is less settled than the earlier version of this analysis suggested. ANDE described special provisions for convergent industries in May, including data centers and AI. On June 9, however, the executive revoked the four decrees associated with that special framework. Public statements since then continue to support digital infrastructure investment, but a supplier, developer or lender should verify the current project-specific tariff and connection regime rather than rely on the earlier special-decree treatment.

The August 13 proposal to create a Ministry of Mines and Energy and a separate electricity regulator adds another layer. It may ultimately strengthen predictability, but for now it means the institutional framework itself is under redesign.

Argentina is assembling the conditions before FID

Argentina’s potential project remains much larger. OpenAI and Sur Energy announced Stargate Argentina in October 2025, and OpenAI’s public position still describes a letter of intent to explore a large-scale data-center project. Sur Energy would lead energy and infrastructure development and form a consortium with a cloud-infrastructure partner, while OpenAI has described itself as a potential offtaker.

By September, the national pipeline had broadened beyond Stargate. Reuters reported multiple Patagonia proposals on September 7. Pampa Energía is evaluating a facility of up to 500 MW near its Loma de la Lata power plant in Neuquén, while projects in Chubut, Buenos Aires and other locations range from roughly 30 MW to 3,000 MW. The same reporting stresses that many projects are still early-stage and depend on financing, formal agreements and additional infrastructure, including fiber.

Since then, the project has gained more local detail. Infobae reported in June that site work had focused on Neuquén’s Confluencia department, including Senillosa and Arroyito, and that energy-supply agreements had been reached with Central Puerto and Genneia. Those details make the proposal more tangible, but they remain reported project-development signals rather than a public FID.

Peter Thiel and Palantir: context, not project evidence

Peter Thiel’s growing Argentina exposure adds a technology and capital-market layer to the country’s AI infrastructure story. On April 23, President Javier Milei received Thiel at the Casa Rosada together with representatives of Thiel Capital and Founders Fund. Public information from the Argentine government confirms the meeting but does not establish a Palantir contract, a data-center mandate or a direct Palantir role in Stargate Argentina.

The signal became more concrete in August from a different direction. A U.S. filing reported by Reuters showed that Thiel Macro acquired roughly a 1 percent stake in Vista Energy, one of the largest producers in Vaca Muerta, in a position valued at about US$76 million. That creates a direct capital connection between Thiel and Argentina’s principal energy basin.

For the data-center story, the distinction matters. OpenAI and Sur Energy represent a defined potential infrastructure pathway. Thiel’s Vista investment represents a concrete energy and capital exposure. Palantir remains a strategic technology and data-system signal rather than a confirmed infrastructure participant. The combination is relevant because Argentina’s AI proposition increasingly depends on the intersection of compute, energy and long-duration capital, but current public evidence does not support treating Palantir as part of the Stargate project.

The policy environment is also moving. Argentina’s existing RIGI framework was extended in 2026, and Decree 105/2026 explicitly includes artificial intelligence within the technology sector. A separate “Super RIGI” for investments above US$1 billion passed the Chamber of Deputies in June with proposed long-duration fiscal, customs and foreign-exchange benefits. But the proposal has not advanced to enactment: on September 2 the governing coalition failed to secure enough Senate committee signatures for a report, and the expected September 10 Senate session was subsequently dropped. As of September 28, current legislative tracking still shows Super RIGI pending Senate approval.

Financing capacity is beginning to enter the public discussion as well. On August 3, CAF said it is evaluating ways to finance data centers and other digital-infrastructure projects in Argentina, including the possibility of syndicated financing with other multilateral and international institutions. CAF did not identify Stargate as a specific financing mandate, so this should be read as a broader market signal rather than project financing already secured.

Two different execution models

The next bottleneck is bankable compute demand

Power is only one side of the business model. AI infrastructure needs customers willing to commit to expensive compute capacity for long periods. The distinction between “available megawatts” and “bankable megawatts” is becoming central.

Paraguay has a sovereign-demand route through the Taiwan partnership, but the long-term customer and workload model for a much larger regional AI market remains to be proven. HIVE has demonstrated that BUZZ HPC can win large AI cloud contracts internationally, including new long-term enterprise agreements in Canada, but its August disclosure does not commit the additional Yguazú power to those workloads.

Argentina has a potentially powerful demand signal because OpenAI has positioned itself as a possible offtaker. But “potential offtaker” is different from a disclosed long-term capacity contract. For lenders, developers and equipment suppliers, the commercial threshold will be the point at which demand becomes contractual enough to support financing and procurement.

What companies should watch next

Three business questions that require deeper research

Where Published Information Stops

Public sources now reveal far more than they did earlier in 2026: Paraguay’s first-phase scale and institutional model, a physical site, HIVE’s power expansion and timing, the reversal of the special convergent-industry decrees, and a broader Argentina project pipeline extending beyond Stargate. They still do not provide a complete package-level procurement calendar or a durable project-specific power contract for the Paraguayan AI projects.

Key gaps include approved vendor lists, EPC responsibilities, detailed grid-connection studies, power pricing and indexation, final cooling architecture, fiber routes and redundancy, GPU or server procurement, construction package timing, binding customer commitments, financing terms and the exact boundary between public infrastructure and private investment.

Those gaps are where direct project validation, supplier mapping, buyer research and market interviews become necessary.

AI infrastructure and data-center research

Econosur can extend the public analysis with targeted market and project research for companies evaluating Argentina, Paraguay or the wider South American digital-infrastructure market.

Strategic conclusion

The strongest 2026 signal is not that one country has “won” the AI-infrastructure race. It is that the two markets now fail at different gates.

Paraguay has proven that large renewable-powered digital infrastructure can be built and has defined a sovereign-AI concept with Taiwan. But the June repeal of the convergent-industry decrees and the August confirmation that the 10 MW project remains in feasibility mean that physical tangibility is ahead of regulatory and commercial closure.

Argentina has the opposite profile. Its incentive architecture is becoming more supportive and the potential Stargate scale is much larger, but the flagship project remains an LOI without public FID. Here, policy optionality is ahead of project commitment.