You can start small and expand as your project actually demands it
Em um mercado onde o acesso à computação de alto desempenho sempre exigiu capital abundante e compromissos de larga escala, a Claro apresentou no Web Summit Rio uma alternativa que redistribui esse poder: a contratação fracionada de GPUs para projetos de inteligência artificial. A iniciativa permite que startups, instituições públicas e centros de pesquisa aluguem apenas a capacidade de que realmente precisam, pagando em reais e expandindo conforme a demanda cresce. É uma tentativa de democratizar o acesso à infraestrutura que, até agora, favorecia quem já tinha recursos para desperdiçar.
- O mercado tradicional de GPUs obrigava empresas a comprar pacotes fixos de oito unidades, mesmo que usassem apenas 20% da capacidade contratada nas fases iniciais — um desperdício estrutural que travava projetos menores.
- Startups e instituições de pesquisa enfrentavam uma barreira de entrada financeira desproporcional, forçadas a investir pesado antes de validar qualquer hipótese com inteligência artificial.
- A Claro lança um modelo de GPU-as-a-Service que permite contratos fracionados, cobrança em reais e expansão gradual conforme o projeto evolui, eliminando a exposição cambial e o compromisso inicial excessivo.
- A empresa se posiciona como o primeiro parceiro NCP da NVIDIA na América Latina e ancora o serviço em parcerias com a USP e a FAPESP, sinalizando ambição além do mercado corporativo.
- O movimento insere a oferta de GPU em um ecossistema de nuvem no qual a Claro já investiu um bilhão de reais, transformando infraestrutura bruta em plataformas de IA acessíveis a empresas, governo e academia.
A Claro chegou ao Web Summit Rio com uma resposta a um problema que tem freado silenciosamente quem tenta construir com inteligência artificial sem grandes reservas de capital: o mercado de GPUs sempre exigiu compra em bloco, mesmo quando o projeto precisava de apenas uma fração dessa capacidade.
O serviço anunciado permite contratar poder computacional por fatias — sem a obrigação de adquirir pacotes completos de hardware. Rodrigo Assad, diretor de inovação e produtos B2B da beOn Claro, foi direto ao ponto: GPUs eram vendidas em conjuntos fixos de oito unidades, exigindo investimento pesado mesmo de organizações ainda em fase de testes. Em muitos casos, apenas 20% da infraestrutura contratada era efetivamente utilizada nesse período. O novo modelo permite começar pequeno, validar o conceito e escalar conforme a demanda real.
Há também uma dimensão financeira relevante. Mário Rachid, diretor executivo de soluções digitais da empresa, destacou que a cobrança é feita em reais — o que elimina a exposição cambial e oferece previsibilidade de custos para startups e instituições que operam com margens apertadas. Para muitos projetos, essa certeza pode ser o que separa o lançamento do arquivamento.
A Claro também anunciou ser o primeiro parceiro Cloud da NVIDIA na América Latina, com acesso a infraestrutura de computação acelerada para treinamento de modelos, visão computacional e análise de dados. A parceria se estende à pesquisa acadêmica, com projetos desenvolvidos em conjunto com a USP e a FAPESP nas áreas de IA e redes 5G.
No fundo, o que a empresa está fazendo é abaixar o piso de entrada. O mercado de GPUs sempre favoreceu quem tinha capital para comprometer antecipadamente. Ao permitir contratos fracionados e expansão gradual, a Claro abre espaço para atores menores — e os próximos meses dirão se isso se traduz em uma nova onda de projetos de IA no Brasil.
Claro walked onto the stage at Web Summit Rio this week with a solution to a problem that has quietly frustrated anyone trying to build with artificial intelligence on a budget: the GPU market forces you to buy in bulk, even when you need only a sliver.
The Brazilian telecom announced a fractional GPU service—essentially the ability to rent computing power by the slice rather than the whole pie. Companies, startups, and government agencies developing AI applications can now contract only the capacity they actually need for training and running machine learning models, without the traditional requirement to purchase complete hardware blocks. The service supports everything from large language models down to smaller specialized systems, and it scales with you as your project grows.
The friction point Claro is addressing is real and measurable. Rodrigo Assad, the company's director of innovation and B2B products at beOn Claro, laid it out plainly: GPUs have historically been sold in fixed packages of eight units, demanding substantial upfront investment even for organizations still in early testing phases. In some cases, Assad noted, companies end up using only about 20 percent of the infrastructure they've paid for during those initial stages. The waste is baked into the model. Claro's approach lets teams start small, prove their concept works, and expand capacity as their workloads actually demand it.
There's a financial dimension too. Mário Rachid, the company's executive director of digital solutions, emphasized that billing happens in Brazilian reais rather than dollars. That matters more than it might sound—it eliminates currency exposure and gives clients predictable costs in their home currency. For a startup or research institution operating on tight margins, that certainty can be the difference between launching a project and shelving it.
Claro positioned this service as part of a larger cloud infrastructure push. Last year the company committed a billion reais to expanding its cloud platform for corporate clients. The GPU offering slots into that ecosystem, complementing a growing portfolio of AI services delivered as platforms rather than raw hardware.
The company also announced it is NVIDIA's first Cloud Partner in the Latin American region, a designation that gives it access to accelerated computing infrastructure for everything from model training to computer vision to data analysis. That partnership extends into research: Claro is supporting projects conducted jointly with the University of São Paulo and the São Paulo Research Foundation, focusing on artificial intelligence and 5G networks.
What Claro is really doing here is lowering the floor. The traditional GPU market has always favored organizations with capital to spend upfront and the scale to justify it. By allowing fractional contracts and gradual expansion, the company is opening the door to smaller players—startups testing ideas, universities running research, government agencies exploring what AI can do for public services. The barrier to entry just got lower. Whether that translates into a wave of new AI projects in Brazil, or simply makes existing ones more efficient, will become clear over the coming months.
Notable Quotes
Historically, one of the main barriers to widespread GPU adoption has been the way they're sold—usually in closed blocks of eight units, requiring substantial investment.— Rodrigo Assad, director of innovation and B2B products at beOn Claro
Fractional GPU contracts enable proof-of-concept projects with lower initial investment, and billing in reais reduces currency exposure and brings greater financial predictability.— Mário Rachid, executive director of digital solutions at Claro
The Hearth Conversation Another angle on the story
Why does it matter that Claro is selling GPUs in fractions instead of whole blocks?
Because right now, if you want to train an AI model, you have to commit to buying eight GPUs at once, even if you only need one or two while you're testing. Most companies waste most of what they buy in the early stages.
So this is really about reducing waste?
It's about reducing waste and risk. If you're a startup with limited capital, you can't afford to spend heavily on infrastructure you might not fully use. Claro lets you start small and pay only for what you actually consume.
How does billing in reais change the equation?
It removes currency risk. If you're a Brazilian company and you're paying in dollars, you're exposed to exchange rate swings. Billing in reais means your costs are predictable in the currency you actually earn in.
Who benefits most from this?
Startups testing new ideas, universities running research, government agencies exploring AI applications. Anyone who can't justify a million-dollar hardware purchase but has a real problem they want to solve with machine learning.
Is Claro the only company offering this?
Not globally—cloud providers like AWS and Google have similar services. But in Latin America, Claro is positioning itself as the first major player doing this at scale, backed by NVIDIA's infrastructure.
What's the catch?
You're renting, not owning. Over time, if you're running heavy workloads constantly, buying your own hardware might be cheaper. But for most organizations, especially in the early stages, the flexibility and lower upfront cost outweigh that.