• MercadoLibre's Q3'26 shareholder letter says its AI investment grew by almost $30 million quarter on quarter; PYMNTS rounds this to $30 million.
• The company says AI usage is growing much faster than AI spend, but the letter gives examples rather than a cost-per-outcome figure.
• Mercado Pago's credit portfolio reached almost $18 billion, up 62% year on year, per the letter.
• NIMAL, the letter's net-interest-margin measure after losses, was 20.1% versus 21.0% a year earlier, so lending growth came with a thinner margin.
• For challenger-bank watchers, the useful test is whether AI savings appear in disclosed cost-to-serve and loss metrics, not only in anecdotes.

MercadoLibre raised its AI investment by almost $30 million between the second and third quarters of 2026, according to the shareholder letter it filed on November 4. The same letter reports that Mercado Pago's credit book grew 62% to almost $18 billion. Read together, they raise a question for any bank or challenger building on AI: whether efficiency gains are large enough to protect margins as lending scales.

What the letter says about AI spend

The letter states that investment in AI grew by almost $30 million quarter on quarter, and that usage is growing much faster than spend. PYMNTS reports the chief financial officer repeating the $30 million figure on the earnings call. The company credits model selection by task and more open-source models for the efficiency. These are management assertions: the letter does not publish a total AI budget or a cost per resolved interaction, so the efficiency claim cannot be independently checked from it.

Where the company says AI is paying back

The letter offers four examples of return. It says revenue uplift from AI in search more than offset model costs, customer service needs fewer representatives by a multiple, and the Mago assistant holds 19 million conversations per month. It also says AI freed almost 20% of finance team hours. Each is a company-reported claim without a disclosed baseline, so they indicate direction rather than measured savings, and none is tied to the fintech segment's cost line.

Credit growth came with a thinner margin

Mercado Pago's credit portfolio reached almost $18 billion in Q3'26, up 62% year on year, while NIMAL fell to 20.1% from 21.0% a year earlier, according to the letter. Fintech monthly active users reached 93 million, up 29% from 72 million. Total payment volume was about $108.8 billion, up 53% in dollars. The company reported net income of $448 million, up 6% from $421 million, so profit grew far more slowly than lending.

What this means for challenger-bank analysis

The case is a useful template for assessing Gulf and global challengers that cite AI as a cost lever: lending volume grew several times faster than profit, and margin narrowed. Readers should look for AI savings in disclosed service cost, credit losses and servicing headcount rather than in conversation counts. The letter covers Latin America only and contains no Gulf data, so the lesson is about disclosure discipline, not regional comparison.

What to watch next

Three checks will show whether the efficiency claim holds. First, whether NIMAL stabilizes or keeps falling as the portfolio grows. Second, whether the company discloses AI spend as a share of operating costs. Third, whether the 19 million monthly Mago conversations translate into measurable service-cost reductions. Until those appear, the $30 million increase is best treated as a reported input, with the claimed output still unverified.