LatAm Tech Weekly

#246: Itau's i.ai , the other side of AI, deals of the week... and much more!

Weekly writing about what is happening in LatAm tech. By day, I am part of the corporate development team at Itau Unibanco. By night, I am reading and learning about technology in general (now, with a focus on AI). During the weekends, I’m writing the LatAm Tech Weekly. And obviously, always running!

If you have not subscribed yet, join the 14,900+ weekly readers by subscribing here!

Follow me on LinkedIn , Instagram or X for daily updates!

Opinions expressed here are solely my own and does not represent those of people, institutions, organizations that I may or may not be associated with in any capacity, unless explicitly stated.

Happy Sunday!

Officially back from vacations - ready for what’s to come!

Today’s intro is a bit of a mix. We start with Itaú Unibanco’s new AI tool for customers. Yes, I work at Itaú, so consider me biased, but this launch is massive and genuinely worth sharing. Right after, we balance it out with a not so great AI story, because staying impartial matters to me.

Hope you like it!

ITAU i.ai

Standard disclosure first: I work at Itaú. The factual claims about Itaú below are based exclusively on public reporting and company disclosures; the analysis, comparisons and enthusiasm are my own.

On July 27, in São Paulo, Itaú Unibanco launched ia.i — short for Inteligência Artificial Itaú and a play on the familiar Brazilian greeting “e aí?”. Positioned on the home screen of the Superapp, it allows clients to interact through natural-language text or voice.

The initial rollout covers approximately 300,000 clients. Itaú expects to reach three million by the end of September and, subject to the technology’s evolution, 100% of individual Superapp users by year-end. João Araújo, director of Strategy, Client Lifecycle and Artificial Intelligence, described the potential audience simply as “dozens of millions.”

The interface is the visible part. What is more interesting is the architecture underneath—and Itaú has now publicly disclosed some of it.

The orchestration layer

CTO Ricardo Guerra told Exame that the bank built an internal platform called Iara. It operates as a gateway to multiple large language models, combining third-party models with proprietary engines developed with NeoSpace, the Brazilian AI startup acquired by Itaú.

Iara decides case by case which engine is most efficient for a particular interaction. In practical terms, this allows the bank to avoid dependence on a single provider and choose models according to the capability and economics required by each task. Itaú operates with open and closed systems, reportedly has more than 80 off-the-shelf models available and is in dialogue with OpenAI, Anthropic, Google and AWS.

Memory

The platform incorporates short- and long-term memory, allowing conversations to retain context over time. Short-term memory preserves the state of the immediate interaction; long-term memory can make future conversations more contextual without requiring the client to begin again from zero.

In banking, this is one of the hardest parts of the architecture. The more useful a system becomes through persistent context, the more carefully identity, consent, privacy, data retention and regulatory controls must be managed.

Agents rather than one general model

Guerra also described purpose-built agents with predefined functions—including an agent for investment recommendations—that can be combined within the broader ecosystem and supervised by machine-learning controls.

This is a meaningful design choice. One general-purpose model may be simpler conceptually, but specialized agents provide clearer boundaries around what each component is permitted to do. That matters considerably when the system operates inside a highly regulated financial institution.

The economics

This is the number I would underline: Itaú says it has reduced its generative-AI costs by 65% since beginning its journey in 2022.

Carlos Eduardo Mazzei, technology director at the bank, attributed that reduction to better token management, centralized governance through a single platform and greater operational efficiency. Tokens are the units used to measure and bill much of an LLM’s input and output.

Mazzei compared the discipline to FinOps in cloud computing: model consumption must be treated as an actively managed operating cost, rather than an unlimited experimentation budget. The most sophisticated model is not necessarily the right model for every job. Accuracy, latency and cost all matter.

Guerra offered a concrete example. Some interactions related to Pix on WhatsApp can be resolved using Itaú’s internal systems, avoiding unnecessary calls—and charges—to external LLMs.

This becomes increasingly important as companies operate several models at the same time. Without orchestration and governance, the cost of AI can expand rapidly, much as cloud spending did during the first wave of enterprise migration. Exame reported that Itaú invested more than R$11 billion in technology across 2024 and 2025.

The global benchmarks

It is worth placing this against what some of the world’s largest banks have already built. Because this is how big this initiative is.

Bank of America’s Erica surpassed three billion client interactions after launching in 2018. Wells Fargo’s “Fargo” processed more than 245 million interactions in 2024, according to public reporting, using an architecture designed to prevent personally identifiable information from being passed to the underlying LLM.

On the employee side, Goldman Sachs made its multi-model GS AI Assistant available to its workforce of approximately 46,000 people. Morgan Stanley deployed its AI assistant across a wealth-management organization of roughly 16,000 financial advisors and reported adoption by 98% of advisor teams.

These provide useful benchmarks for interaction volume, enterprise distribution and adoption. What is notable about ia.i is the breadth of the intended interface. Rather than using generative AI only inside isolated functions, Itaú wants to bring service, financial guidance and transaction execution into one conversational layer.

GO ITAÚ!

The other side of AI

Two years ago, Leopold Aschenbrenner was one of the most quoted people in AI investing. A Columbia valedictorian at 19 and former member of OpenAI’s Superalignment team, he published a 165-page essay after leaving the company in 2024, arguing that increasingly capable models would require an unprecedented buildout of chips, memory, data centers and power. He then turned that thesis into a fund backed by Patrick and John Collison, Nat Friedman and Daniel Gross. It reportedly grew from around $1.5 billion in 2025 to roughly $45 billion by the beginning of July. He is 25.

I want to be careful here: the intellectual case was serious, and today’s capex numbers are validating parts of it. But being right about a technology and being right about a trade are different things. The second depends less on vision and more on position sizing, leverage and surviving the drawdowns along the way.

The fund was long AI infrastructure and short software companies expected to be disrupted by AI—one of Wall Street’s most crowded trades. It reportedly held concentrated positions in Nebius, CoreWeave, SanDisk, Bloom Energy and IREN, with leverage near four times. Then both sides moved against it: the longs fell 40–50% while the software shorts rallied, triggering margin calls and forced deleveraging. This came after the fund had gained 439% net through June.

Prime brokers reportedly forced the sale of the public portfolio to Citadel at a discount, reducing assets to around $10 billion. Aschenbrenner’s investor letter put the unaudited monthly loss at 67%, while still leaving the fund up 80% for the year. He also stressed that the firm had not been liquidated and would continue as a private investment company, including through a $5 billion stake in Anthropic.

The detail that matters most is that the S&P 500 remained near record highs throughout. This was not a market crash. It was a concentrated and leveraged consensus trade unwinding into itself. The thesis may ultimately prove right. The structure around it was not survivable. And no amount of conviction replaces the boring discipline of making sure you are never forced to sell.

General news:

• Itaú Unibanco launched ia.i, its new generative AI assistant integrated into the bank’s mobile app, enabling customers to interact through natural language to check balances, analyze spending, receive personalized recommendations and initiate Pix transfers. The rollout begins with 300,000 customers before expanding to the entire user base by year-end, reinforcing Itaú’s strategy to position AI as a financial copilot. 🇧🇷

• Monashees and Google’s AI Futures Fund have now partnered to launch a co-investment program for Brazilian founders building AI-native startups at the Pre-Seed and Seed stages — companies where AI is the core of the product, not just an added feature.

The Gama Fund will select five startups to receive:
✦ Up to US$2M in funding
✦ Up to US$350K in Google Cloud and Gemini credits
✦ Early access to Google’s frontier models and the Google DeepMind team
✦ Hands-on collaboration with Google, Monashees, and dedicated startup managers
✦ In-person immersion sessions in the U.S. with DeepMind and the broader Monashees network

• Asaas is expanding into a comprehensive financial platform for SMEs by adding services such as credit, insurance, CRM and accounting through acquisitions, internal development and new ventures. The strategy supports its goal of reaching R$1 billion in revenue in 2026 while deepening customer engagement. 🇧🇷

• Portugal’s Vide de Grunwald selected Maringá to host Brazil’s first Blue Cluster, an innovation hub focused on the blue economy. The initiative will connect startups, researchers, companies and investors to accelerate innovation in biotechnology, AI, watertech, genetics and foodtech. 🇧🇷🇵🇹

• Shein reported a US$99 million net loss in the first quarter of 2026 as new U.S. tariffs on low-value imports weighed on sales in its largest market ahead of its planned Hong Kong IPO. The results have raised concerns about the company’s ability to achieve its targeted US$40–50 billion valuation. 🇨🇳🇺🇸

• Yungas launched a US$920,000 fund to invest in AI startups focused on franchise and multi-unit operations, making its first investment in fintech Fin. The initiative expands Yungas’ AI ecosystem while accelerating operational digitalization across enterprise retail networks. 🇧🇷

• Valutia launched fundraising for its second US$30 million VC fund, nearly tripling the size of its inaugural vehicle while continuing to back Brazilian founders building global companies. The firm plans to invest US$500,000–1 million in early-stage startups and reserve capital for follow-on rounds. 🇧🇷

• 99Pay received approval to operate as a Finance Company (SCFI), allowing the fintech to diversify funding sources and expand lending beyond its own balance sheet. The new license strengthens its ability to scale credit operations for more than 23 million users. 🇧🇷

Deals:

• iFood raised R$600 million through a senior tranche offering for one of its credit funds, expanding its structured lending platform. The transaction reflects the growing use of FIDCs by digital platforms to scale financial services and diversify funding sources. 🇧🇷

General news:

• Brazilian ERP provider Bling is expanding into a one-stop platform for SMEs by integrating e-commerce, logistics, financial services, marketing and AI-powered automation into a single ecosystem. Backed by LWSA, the company is rolling out AI agents and a conversational assistant to simplify multichannel operations and strengthen merchant growth and retention. 🇧🇷

• 99 launched 99Compras in São Paulo with a R$100 million investment, expanding its retail delivery service as part of its super app strategy. The platform debuts with 3,000 retail partners and more than 200,000 couriers, leveraging AI-powered route optimization to compete in Brazil’s on-demand commerce market. 🇧🇷

• Apple became the first company to surpass a US$5 trillion market capitalization, reclaiming its position as the world’s most valuable public company. The milestone reflects investor rotation toward companies with stronger cash generation as concerns grow over AI infrastructure spending and semiconductor valuations. 🇺🇸

• Angel Ventures launched Apex Company Lab, a venture studio that will build startups alongside its traditional VC strategy. Its first company, Bark & Co, targets Mexico’s growing pet care market through a technology-enabled business model. 🇲🇽

• EBANX partnered with Pagaleve to launch Pix 4x, a BNPL solution that lets consumers split Pix purchases into four installments without a credit card. Powered by Pagaleve’s AI credit engine, the offering is immediately available to more than 500 global merchants connected to EBANX. 🇧🇷

Deals:

• Argentinian fintech Fonder is raising a US$500,000 pre-seed round to expand its AI-powered treasury automation platform across Latin America. The startup connects banks, accounting systems and payment platforms to automate cash management for SMEs and is targeting US$1 million in ARR within six months. 🇦🇷

General news:

• Databricks is strengthening its presence in Brazil by creating a dedicated startup team to support early-stage AI companies with technical expertise, cloud credits and go-to-market resources. The initiative reinforces Databricks’ strategy to expand its footprint while accelerating Brazil’s AI startup ecosystem. 🇧🇷

• Brazilian fintech Noh integrated Open Finance into its platform, allowing couples to consolidate accounts, investments and credit cards from multiple institutions into a single financial dashboard. The company aims to position itself as the primary financial hub for household money management while driving broader Open Finance adoption. 🇧🇷

• Chilean regtech Sheriff launched Marshal, an AI-powered compliance agent that automates due diligence and sanctions screening across 192 countries. The platform supports the company’s international expansion strategy while helping businesses strengthen regulatory compliance and fraud prevention. 🇨🇱

• Meta CEO Mark Zuckerberg argued against banning Chinese AI models, warning that excessive restrictions could slow innovation and cybersecurity while reinforcing the dominance of a few AI companies. His comments come as the U.S. debates tighter regulation of frontier AI models amid growing competition with China. 🇺🇸

• OpenAI reduced the pricing of its GPT-5.6 models following infrastructure efficiency improvements, cutting GPT-5.6 Luna prices by around 80% and GPT-5.6 Terra by 20%. The move strengthens OpenAI’s competitive positioning as AI providers increasingly compete on both performance and cost. 🇺🇸

• Global66 expanded its partnership with Pomelo by migrating its Chilean Mastercard operations to the payment infrastructure provider. The unified platform strengthens card issuance, tokenization and fraud management while supporting Global66’s regional expansion strategy across Latin America. 🇨🇱🇨🇴

Deals:

• Argentinian SaaS startup COR raised a US$30 million round to accelerate AI capabilities, expand into new professional services verticals and strengthen its presence in Brazil. The company is enhancing its platform to help businesses manage hybrid teams of humans and AI agents. 🇦🇷

• AI sales platform Dapta acquired paid media agency Beezion to bring demand generation in-house and accelerate customer acquisition. Following the acquisition, the company reported more than 20% growth in its first month while expanding its AI-powered sales platform across Latin America. 🇨🇱

General news:

• Brazil’s high interest rates are reshaping the startup ecosystem by accelerating M&A activity while slowing venture capital investments. Investors are increasingly favoring profitable, well-governed companies as startups prioritize operational efficiency over growth at all costs. 🇧🇷

• Meta and Microsoft delivered contrasting AI investment results, with Meta’s free cash flow falling 91% after heavy AI infrastructure spending while Microsoft posted strong revenue growth and cash generation. The results highlight increasing investor scrutiny over AI capital expenditure efficiency. 🇺🇸

• Brazilian investment platform EqSeed launched EqSeed Academy, expanding beyond equity crowdfunding with an education initiative that prepares entrepreneurs for fundraising. The strategy strengthens EqSeed’s position as an end-to-end capital platform for private companies. 🇧🇷

• Startups launched VC Matchmaker, a free AI-powered platform that matches founders with the most suitable venture capital funds. The tool analyzes startup profiles against a database of more than 600 VC firms across Brazil and Latin America. 🇧🇷

• Atlas Inc. will build its tokenization platform on XRP Ledger after receiving a grant from the Ripple ecosystem. The company will focus on tokenizing real-world assets such as real estate, telecom infrastructure and securitized financial products. 🇧🇷

Deals:

• Brazilian agtech Ecotrace secured a new funding round led by GS1 Ventures to accelerate international expansion and strengthen its AI- and blockchain-powered commodity traceability platform. The company is targeting R$60 million in revenue by 2028. 🇧🇷

• Brazilian AI startup Einship raised a US$1 million seed round to expand internationally and strengthen its AI platform for import and export operations. The company plans to grow across Europe, China, the U.S. and Latin America. 🇧🇷

• Chilean agritech Ecoterra raised US$1 million to expand into Colombia and prepare for entry into Mexico. The funding will support the company’s blockchain-enabled food traceability platform and sustainable agriculture initiatives. 🇨🇱

• Runflow raised a R$7 million seed round to expand its enterprise AI agent platform. The startup helps large organizations build and manage AI agents that automate complex business workflows and expects to more than double ARR by year-end. 🇧🇷

• ICE agreed to acquire MarketAxess for US$5.7 billion, expanding beyond exchange operations into electronic fixed-income trading. The acquisition strengthens ICE’s financial infrastructure ecosystem through market data, analytics and bond trading capabilities. 🇺🇸

General news:

• OpenAI surpassed 1 billion registered users after cutting GPT-5.6 API prices by up to 80%, accelerating adoption as competition with Anthropic intensifies. The company also expanded its long-term AI infrastructure plans, including a US$250 billion Nvidia partnership and projected compute spending of up to US$750 billion by 2030. 🇺🇸

• Situational Awareness, the AI-focused hedge fund lost 67% in July after leveraged bets on AI infrastructure stocks unraveled during a market correction. The collapse highlights growing concerns over elevated AI valuations and the risks of highly leveraged investment strategies. 🇺🇸

• Strategic investors are gaining ground in M&A as private equity firms face slower fundraising and fewer exits. The shift reflects stronger demand for acquisitions driven by operational synergies rather than financial engineering. 🌍

• Brazil’s Central Bank extended the Pix refund dispute deadline from 30 to 80 days under the new MED 2.0 framework. The update strengthens fraud prevention while expanding transaction tracing and supporting the BC’s broader anti-fraud roadmap. 🇧🇷

• Mattos Filho and PUC-SP launched a new executive course focused on venture capital, governance, fundraising and AI. The initiative reflects the growing sophistication of Brazil’s startup investment ecosystem. 🇧🇷

• Apple shares fell nearly 10% after warning that AI-driven demand for chips and memory is straining global supply chains. The selloff erased roughly US$500 billion in market value and raised concerns over rising AI infrastructure costs. 🇺🇸

• Amazon reported record quarterly results, with net income more than tripling as AWS revenue surged 36.7% on accelerating AI demand. The company also announced plans to invest US$200 billion in capital expenditures during 2026 to expand AI infrastructure. 🇺🇸

• Latin America raised US$8.6B in startup funding in 2025, led by Brazil and Mexico — Latin America attracted US$8.6 billion across 500+ startup funding deals in 2025, with Brazil and Mexico capturing 84% of total investment, according to Briter Intelligence’s first regional venture report. Fintech remained the leading sector, accounting for 64% of invested capital, while healthtech, software, and climate tech gained momentum despite ongoing challenges around M&A liquidity, funding concentration, and investment inequality. 🌎

• Brazil’s startup ecosystem is increasingly shifting toward structured debt as founders replace equity fundraising with instruments such as FIDCs and CRIs. The trend reflects a more mature ecosystem focused on capital efficiency amid higher interest rates. 🇧🇷

• Brazilian electric motorcycle startup Vammo reached EBITDA breakeven two quarters ahead of schedule while doubling revenue in four months. The company plans to expand across Latin America and grow its fleet to 15,000 electric motorcycles by year-end. 🇧🇷

• Nequi will begin operating as a regulated finance company from September 1 after receiving regulatory approval in Colombia. The transition adds deposit insurance while preserving the existing customer experience. 🇨🇴

• Five Brazilian founders launched Five Founders, an invite-only community for entrepreneurs expanding into the U.S. The initiative combines networking, operational support and investment opportunities for Brazilian founders building internationally. 🇧🇷

Deals:

• Rintin raises US$6.2M to expand AI-powered B2B commerce in Mexico — Mexican B2B commerce platform Rintin secured a US$6.2 million round led by Cometa to scale its AI-powered marketplace, embedded credit services, and logistics network for micro and small retailers. The company has supported 12,000+ women entrepreneurs, facilitated US$2 million+ in loans, and aims to deepen financial inclusion for underserved businesses across Mexico. 🇲🇽

• Brazilian AI infrastructure startup Strattum raised a US$3.2 million pre-seed round to accelerate international expansion. The company focuses on organizing enterprise data to power AI applications rather than building AI models directly. 🇧🇷

• Colombian proptech Duppla raised US$60 million to expand its rent-to-own housing platform across Colombia. The funding will support nationwide growth after helping more than 300 families purchase homes. 🇨🇴

• Conta Azul acquired cloud accounting startup Mister Contador to strengthen its financial management platform for SMEs. The acquisition expands its accounting automation capabilities following its integration into Visma. 🇧🇷

The AI news of the week was Anthropic’s disclosure that three Claude models unintentionally compromised real-world systems during cybersecurity evaluations. I mentioned OpenAI’s similar incident last week, but Anthropic’s case adds another warning sign: after a configuration error left internet access open, the models exploited weak passwords, exposed endpoints and SQL injection flaws, scanned thousands of targets and even published a malicious Python package downloaded by 15 real systems. Anthropic paused the evaluations, notified the affected organizations and is tightening its safeguards. The lesson is becoming difficult to ignore: frontier agents are now capable enough that the environments built to test them must be secured like production infrastructure.

• Rio Innovation Week 2026
  Date: August 4–7, 2026
  Location: Rio de Janeiro, Brazil
  Description: One of Brazil’s largest innovation festivals, bringing together entrepreneurs, startups, corporations, investors, and policymakers to discuss technology, business, and social transformation.
  More info

• SP2B (São Paulo Beyond Business) 2026
  Date: August 9–16, 2026
  Location: São Paulo, Brazil
  Description: A new business and innovation event designed to connect entrepreneurs, executives, investors, and corporations through discussions on growth, creativity, and the future of business.
  More info

• Deep Tech Summit 2026
  Date: August 11–12, 2026
  Location: São Paulo, Brazil
  Description: A conference focused on science-based innovation, bringing together deep tech startups, investors, corporations, and researchers to explore frontier technologies and emerging opportunities.
  More info

• Febraban Tech 2026
  Date: August 24–26, 2026
  Location: São Paulo, Brazil
  Description: One of the main financial technology and innovation events for the banking and financial services sector in Latin America.
  More info

• Startup Summit 2026
  Date: August 26–28, 2026
  Location: Florianópolis, Brazil
  Description: One of Latin America’s leading startup events, connecting founders, investors, and ecosystem leaders through content, networking, and business opportunities.
  More info

• HackTown 2026
  Date: September 3–7, 2026
  Location: Santa Rita do Sapucaí, Brazil
  Description: A unique innovation, technology, and culture festival that transforms an entire city into a hub for networking, learning, entrepreneurship, and creative collaboration.
  More info

• Hotmart Fire 2026
  Date: September 10–12, 2026
  Location: Belo Horizonte, Brazil
  Description: One of the leading events for the digital business and creator economy ecosystem, covering marketing, sales, online education, innovation, and business growth.
  More info

• Dreamforce 2026
  Date: September 15–17, 2026
  Location: San Francisco, United States
  Description: Salesforce’s flagship conference focused on CRM, AI, digital transformation, customer experience, and enterprise innovation.
  More info

• Asaas Connect 2026
  Date: October 14, 2026
  Location: São Paulo, Brazil
  Description: A business conference bringing together CEOs, executives, and growth-focused professionals for keynote sessions, networking, and discussions on entrepreneurship, innovation, leadership, and business growth.
  More info

• TechCrunch Disrupt 2026
  Date: October 13–15, 2026
  Location: San Francisco, United States
  Description: A leading global startup conference where founders, investors, and technology leaders discuss entrepreneurship, fundraising, and innovation.
  More info

• Web Summit Lisbon 2026
  Date: November 9–12, 2026
  Location: Lisbon, Portugal
  Description: One of the world’s largest technology conferences, bringing together entrepreneurs, investors, executives, and policymakers to discuss global innovation trends.
  More info

• Slush 2026
  Date: November 18–19, 2026
  Location: Helsinki, Finland
  Description: A globally recognized startup and venture capital conference focused on connecting founders and investors while fostering innovation and growth.
  More info

• AWS re:Invent 2026
  Date: November 30 – December 4, 2026
  Location: Las Vegas, United States
  Description: AWS’s flagship cloud computing conference, featuring product launches, technical sessions, training, and discussions on cloud infrastructure, AI, and data.
  More info

Recently, Emerging VC Fellows published a report I found worth the read, done in partnership with AWS, ABVCAP and FM Derraik on how venture capital funds are actually adopting AI in their operations. The core argument: the conversation has moved past whether to use AI, and into how to embed it with operational coherence and real governance, since the volume of market signals today outpaces what teams can process manually.

Key takeaways:

Four maturity levels for VC funds:

• Intuitive — decisions run on networks and relationships, no structured data

• Data-aware — structured CRM, but analysis still manual

• Data-driven — LLM connected to tools via MCP, AI organizes context, investor decides (Point Nine’s case)

• AI-augmented — proprietary infrastructure and in-house technical team (EQT Ventures and Earlybird; the latter’s AI covers 95% of monitored opportunities and drives 50% of active deal flow)

• Most funds are still behind: the top tiers are the exception — most use AI in scattered, one-off ways without deep integration into decision-making.

• Where AI helps most: sourcing, screening, portfolio monitoring

• Where human judgment still leads: reading founders, negotiation, building trust with LPs

• The analyst role is shifting: less value in manual processing and memo-writing, more value in qualitative judgment and orchestrating automated systems.

• Legal angle (FM/Derraik): AI output is only one fraction of due diligence — a good advisor’s edge is spotting what AI can’t weigh on its own.

• Harry Stebbings’ 20VC interview with Joon Sung Park, founder and CEO of Simile

“You really have to be unwaveringly obsessed...” Harry Stebbings on six-day work weeks