O que aconteceu
BNamericas reports that Claro Empresas, América Móvil’s Brazilian corporate unit, is preparing to install 256 GPUs at its São Paulo data center for a GPU-as-a-service offering. The service is already available to clients through infrastructure operating abroad, while the Brazilian deployment remains dependent on equipment imports and facility upgrades. Claro says the offering can reduce AI processing costs by up to 50%.
BNamericas reports that Claro Empresas, the corporate arm of América Móvil’s Brazilian carrier, is building a GPU-as-a-service business with Nvidia and Oracle. The company says the service is intended to reduce AI processing costs for Brazilian companies by as much as 50%. The offering lets customers rent computing capacity rather than buy complete GPU systems.
BNamericas reports that the service is already available to clients, but the infrastructure currently running for those customers is abroad rather than in Brazil. BNamericas reports that Claro is importing 256 GPU cards for installation at its São Paulo data center. The import has not yet been completed, and the carrier is adapting the facility to receive the equipment. The company plans to make the capacity available to customers on a first-come, first-served basis and expects to order additional equipment as installed capacity is consumed.
The report does not provide an expected completion date, a confirmed launch date for the Brazilian infrastructure or evidence that all 256 GPUs have arrived. The reported commercial model includes fractional GPU capacity, billing in Brazilian reais and a spending cap intended to make variable AI-token costs more predictable. BNamericas reports that Claro’s control panel can alert a customer when projected consumption is likely to exceed a configured limit, using R$10,000 as an example. Claro also says it maintains an internal team to help customers design use cases before scaling them. The report does not disclose standard prices, service-level commitments, minimum contracts, supported models or the technical definition of a fractional GPU allocation.
BNamericas reports that Claro launched Claro STT, a speech-to-text product for the financial sector, at the same Febraban Tech event where it described the infrastructure strategy. The company says it tested the transcription system in its own call centers and processed all daily calls, compared with manual sampling of roughly 5% to 10% of operations. Claro also attributes a 10% increase in sales conversion to the pilot. Those figures are company-reported results; the source supplies no independent evaluation, methodology, baseline period or details about whether other operational changes contributed to the result.
Leia a fonte primária: bnamericas.com ↗
Por que isso importa
The plan could give Brazilian companies access to AI computing without purchasing or operating their own GPU infrastructure. Local billing, fractional capacity and spending caps address practical adoption barriers, while in-country processing could matter to organizations managing Brazilian data and regulatory requirements. The reported benefits remain company claims that have not been independently confirmed.
The reported offering targets one of the most concrete barriers to AI adoption: access to expensive accelerated computing. Many organizations may want to test speech, language or other AI workloads without committing to a full hardware deployment. Fractional capacity could lower the entry point, while billing in reais may reduce currency exposure for Brazilian customers. These are practical design choices, but their value will depend on how capacity is allocated during periods of high demand and whether customers can obtain predictable performance rather than only nominal access.
The local infrastructure plan also has a data-governance dimension. BNamericas reports that Claro is positioning processing inside its Brazilian cloud as relevant to LGPD obligations and central-bank regulation, especially for government and financial customers. Claro’s government executive is quoted as saying that meeting those requirements is an obligation rather than a differentiator. The report does not establish that the planned service has received any regulatory approval or that local hosting alone resolves every privacy, security, retention or cross-border data requirement.
The infrastructure is significant because it connects AI software demand with the physical constraints of power, cooling and data-center design. BNamericas reports that Claro’s São Paulo facility already has Tier III certification but requires expanded power capacity, modular cooling changes and reinforcement of its raised floor to support heavier, more energy-intensive GPU racks. The company has not disclosed the investment amount or the facility’s total IT capacity. That information is necessary to judge the scale of the deployment and the economics of the service.
The move also illustrates a broader shift in which telecom operators and other infrastructure companies seek revenue from AI computing, rather than treating data centers only as support facilities for connectivity. BNamericas reports that Claro sees the São Paulo deployment as a pilot for operations across Latin America and that other regional companies are pursuing GPU-rental models. The report says Claro is the first carrier in the region to offer Nvidia’s GPU-as-a-service, according to an Nvidia executive. That distinction is an attributed company statement, not an independently established market ranking.
O que assistir a seguir
The key tests are whether Claro completes the GPU import and São Paulo upgrades, what capacity and pricing customers actually receive, and whether the promised savings are achieved across real workloads. Independent evidence will also be needed for Claro’s reported call-center conversion gains, the scope of local processing, and any later expansion to other Brazilian facilities.
The first verification point is physical delivery and readiness. BNamericas reports that Claro is still importing the 256 GPUs and adapting the São Paulo site. Follow-up reporting should establish how many GPUs are installed, whether they are operational for Brazilian customers, what workloads they support and whether the service remains dependent on Oracle-operated infrastructure abroad during the transition. The source does not provide a firm timetable, so the Brazilian rollout should not be treated as complete.
The second is whether the promised economics survive real usage. Claro’s claimed cost reduction of up to 50% needs a defined comparison: against which cloud provider, hardware configuration, workload, utilization rate and service tier. Customers will also need to understand how token consumption is estimated, how spending alerts are calculated, what happens when a cap is reached and whether extra usage requires explicit authorization. None of those contract and measurement details are provided in the report.
The Claro STT results warrant similar scrutiny. BNamericas attributes the reported 100% call transcription rate and 10% sales-conversion increase to Claro’s internal pilot, but the source does not identify the sample size, evaluation period, transcription accuracy, language conditions, staffing changes or controls used to isolate the AI system’s effect. Independent testing would help determine whether the product’s value comes primarily from improved transcription, broader call coverage, workflow changes or other factors.
Finally, the reported expansion plans should be distinguished from confirmed deployments. BNamericas says Claro may eventually adapt data centers in Rio de Janeiro and Brasília, while the company is initially concentrating the GPU-as-a-service structure in São Paulo. An executive also signaled a possible future connection between the enterprise GPU business and a consumer-facing ChatGPT partnership, but the source says there is no joint commercial offering yet. Neither expansion nor convergence should be presented as launched until Claro, Oracle or another authoritative source provides concrete implementation details.
