Argentina has been announcing AI infrastructure for a year. The most credible development so far is not a hyperscaler pledge. It is an incumbent energy company registering a shell with AR$30 million in capital.
What actually happened
Pampa Energía, chaired by Marcelo Mindlin, has created Pampa AI S.A.U. According to dolarhoy's report on the Boletín Oficial entry, the company was constituted on 26 October with initial capital of $30 million pesos. Pampa Energía is its only shareholder. Its stated purpose is to build and operate data centers, sell computing capacity and provide AI services. The same report stresses that incorporation does not mean an investment has been confirmed or that any operator has signed.
The physical case is stronger than the corporate one. At Argentina Week in Paris, Mindlin said Pampa already had "casi 800 megavatios disponibles" for next year, and he invited hyperscalers to build in Vaca Muerta. That is a company claim, not an independently audited figure, and no hyperscaler has accepted the invitation.
The case for caution
The strongest argument against celebrating this is also the simplest one: announcements are cheap. OpenAI and Sur Energy's "Stargate Argentina" was unveiled in October 2025, and a year later it shows no construction and no enrollment in the RIGI incentive regime. A skeptic would say Pampa AI is the same pattern in a smaller wrapper. Large tax incentives can also hand foreign firms generous terms without guaranteeing local jobs, local procurement or lasting spillovers. That concern is fair, and it should shape how the incentives are written.
Why power is the right starting point
Still, the two stories are not the same. Stargate was a demand-side headline attached to capacity that did not yet exist. Pampa's pitch begins with the scarce input, which is firm electricity near cheap gas. AI data centers are limited by megawatts, cooling and grid interconnection long before they are limited by clever regulation. A generator that can offer power at a site it already controls removes the first and hardest bottleneck.
That favours a light-touch policy. The state does not need to pick a champion or design an AI-specific licensing regime. It needs to make a large, capital-intensive, long-lived investment bankable. A data center is a 15-to-30-year asset. Investors price in the risk that tax and currency rules will change before they recover their capital.
What the incentive framework offers
The existing regime is RIGI, created by Ley 27.742. The Economy Ministry's RIGI explainer lists a general investment floor of US$200 million, a fixed 25% income tax rate in place of the graduated general scale, and 30 years of stability for the incentives from adhesion. The program's official page is the starting point for the project registry.
The pending "Súper RIGI" raises the threshold and the prize. La Nación reported that the Chamber of Deputies approved it on 24 June 2026 by 130 votes to 106, with 7 abstentions. It covers investments above US$1 billion, with a reduced 15% income tax rate, against 25% under RIGI and 35% in the general regime, plus 30 years of tax and currency stability. It then went to the Senate for final approval. Nothing I could verify shows the Senate has finished its work, so the bill remains pending.
The tension inside Super RIGI
There is a design question the bill's backers should answer. A US$1 billion floor suits a hyperscale campus. It excludes the mid-sized, modular and regional facilities that would host local startups, universities and public-sector workloads. Pampa AI may or may not reach that threshold, and the 800 MW it describes could support facilities of very different sizes. Sound policy would ensure that the lower RIGI tier remains workable for AI infrastructure below the billion-dollar line, rather than leaving a gap that forces everything into one giant-project model.
The opposite error is also real. A 15% rate for 30 years is a large fiscal commitment made for a technology whose economics are changing quickly. Proportionate design would pair it with transparency. That means a public registry of enrolled projects, clear energy-supply rules so that data centers do not crowd out residential and industrial users, and a review clause tied to the bill's own five-year initial validity.
What to watch
The test is concrete and measurable, and it does not depend on rhetoric:
- Senate action on Super RIGI. Until it becomes law, the 15% rate is a promise rather than a term an investor can rely on.
- A named anchor tenant for Pampa AI. Capacity without a customer is only a power plant with a new brand.
- A first RIGI or Super RIGI enrollment for any AI facility. Stargate Argentina's lack of one after a year is the benchmark to beat.
- Grid and gas-supply arrangements. These determine whether the 800 MW is real for AI loads or only available on paper.
Argentina has the ingredients that matter most for compute: abundant gas, generation capacity and a government willing to offer long-term stability. What it lacks is a signed project. The right response is neither to subsidise hype nor to regulate AI before the buildings exist. Pass a clear, stable incentive law, keep the rules open to inspection, and let builders prove the case with steel in the ground.