LatAm Tech Weekly
#197: SPECIAL EDITION - Separating what matters from the noise in ARTIFICIAL INTELLIGENCE
Happy Sunday!
One of my main goals with this newsletter is to cut through the noise. We’re bombarded with headlines every day, and it’s getting harder to spot what’s actually worth our attention. My usual approach is simple: pick the best reports, read them cover-to-cover (no AI shortcuts here), and explain why they matter. I normally kick off each edition that way and then deliver the main pieces of news—this week, I’m flipping the script.
Over the past few weeks I’ve read a lot on AI, from The Economist deep dives to OpenAI’s latest white papers. The result? Classic information overload. I’d absorbed a ton, yet couldn’t pinpoint the handful of takeaways I should keep front-of-mind. So I hit pause, and this Sunday I reread the pieces that truly resonated, and distilled them into the essentials. In today’s issue, you’ll get those highlights: why each source stands out, what it actually means, and how the dots connect.
Writing this every Sunday is my built-in learning loop, and—self-serving as it sounds—today’s “AI special” is my way of turning my personal crash-course into something useful for you, too.
Let’s dig in.
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.
Last week’s Economist cover was all about AI—and while I know it’s behind a paywall, it’s one of those editions I genuinely think is worth tracking down.
Let’s start with a big-picture rewind. Before 1700, the global economy grew at just 8% per century. If anyone had predicted what was about to come, they would’ve sounded completely insane. But then came the Industrial Revolution, and for the next 300 years, growth exploded—averaging 350% per century. Lower mortality, higher fertility, more people, more ideas… the flywheel slowly started turning. Eventually, richer societies had fewer kids, which meant better living standards and consistent growth around 2% per year…
BUT - AI doesn’t have to wait for babies to grow up.
That’s what makes this moment so wild. Technology, especially AI, doesn’t rely on demographics to scale. Sam Altman, CEO of OpenAI, predicts that by next year, AI could already be generating novel insights. Not just repeating back what it learned, but producing original thought. As I’ve written before, one of the craziest (and most exciting) things about this technology is that it’s now training itself to improve. By 2028, some believe these systems will be overseeing their own upgrades. A self-reinforcing cycle, speeding up discovery faster than we’ve ever seen.
That’s why people are talking about a potential second economic explosion. If machines can drive technological breakthroughs without human bottlenecks—and if we keep reinvesting the returns into even more powerful compute—we could see wealth accumulate at a pace that blows today’s numbers out of the water. According to Epoch AI, once AI handles just 30% of tasks, global GDP growth could hit 20% per year. That’s not a typo.
Now, I know that sounds extreme. Historically, the safest bet has always been “more of the same.” But sometimes, the future breaks the mold. And honestly, I think we’re approaching one of those moments. In just a few years, AI could outperform the average human on virtually every cognitive task. Sure, it still makes mistakes. Yes, we still need humans in the loop. But that’s been true for every transformational technology. The difference with this one? It gets better on its own—every day.
And while it’s easy to be skeptical, let’s not ignore the scoreboard. Just this year, large language models from OpenAI and Google DeepMind hit gold medal level on the International Mathematical Olympiad—18 years ahead of schedule, according to expert forecasts from just a few years ago. Meanwhile, the race for compute is turning into a global arms race. By 2027, it could take 1,000× more computing power to train a frontier model than what was used to build GPT-4.
It’s happening faster than we thought—and we’re all going to have to decide how we respond.
Okay—here’s where the skeptics usually jump in: What about compute and energy? Can the world actually support the massive power demands that AI will require?
The data is pretty sobering. According to a RAND Corporation study published in January 2025, a single frontier-model training run could consume up to 1 gigawatt of power by 2028—and as much as 8 gigawatts by 2030 if current compute scaling trends continue. That’s the equivalent of eight nuclear reactors… just to train one model.
The International Energy Agency backs this up, projecting that global electricity consumption from data centers will double between 2022 and 2026, with AI alone accounting for roughly 90% of that growth.
The takeaway? The energy demands of AI are real—and growing fast. But I’ll explain why I still think we’ll rise to the challenge.
Google recently revealed that embedding generative answers directly into search results multiplies the floating-point operations per query by 10×—essentially turning search itself into an AI inference engine.
Still, I remain optimistic that supply will rise to meet demand. Since 2020, Amazon has secured more than 10 GW of new renewable energy capacity. China, on its end, is repurposing underutilized hydro and solar energy to power inland “Western Compute” hubs. And in the U.S., regulators in Virginia are fast-tracking small modular nuclear reactors and grid upgrades specifically designed for hyperscale data centers.
Meanwhile, hardware keeps getting smarter. Each new GPU generation delivers around 20% more performance per watt. Add to that the custom AI chips from AWS, Google, and Microsoft—designed to squeeze even more efficiency out of every inference—and you get a powerful cocktail: smarter chips, smarter infrastructure, and smarter energy planning.
Put all of that together—efficiency gains, geographic arbitrage, and dedicated clean-power buildouts—and I think we’ll have enough power not just to keep the lights on, but to fuel the next wave of AI breakthroughs.
Shifting to a more local perspective, the recent interview with Pedro Moreira Salles and Roberto Setubal—co-chairmen of Itaú Unibanco’s board—put Brazil’s largest private bank firmly in the AI spotlight. While the conversation touched on legacy, regulation, and a century’s worth of crisis management, both leaders were unequivocal about one thing: artificial intelligence is the defining force for Itaú’s next 100 years.
They described AI as “the fastest-adopted technology in our history,” and made it clear that its impact will be “far greater and arrive much sooner than we imagined a year ago.” To prepare, they laid out three strategic priorities: (1) shift from mass segmentation to hyper-personalized, real-time offerings powered by cloud-native models; (2) streamline decision-making so AI pilots can move from idea to implementation in days, not months; and (3) while partnering with best-in-class cloud providers and pursuing selective M&A, invest in building core AI capabilities in-house—because owning your tech stack means owning your future.
Speaking of financial institutions, QED’s recent report on AI agents and the future of agentic payments is well worth a read. The paper makes a compelling case: as autonomous bots start shopping on our behalf, they could fundamentally upend the familiar four-party card model. Once an AI agent is inserted between consumer and merchant, the flow doesn’t just add complexity—it expands to at least six, and potentially nine different players, depending on whether the agent holds payment credentials directly or operates through a staged wallet.
Right now, bot-driven purchases are still niche—mostly experimental, sometimes clunky, and often just for fun. But that’s changing fast. Heavyweights like Visa, Mastercard, PayPal, Coinbase, Amazon, OpenAI, Google, and Perplexity are already launching SDKs that let bots transact natively. That land grab introduces two immediate challenges: first, merchants need to trust bot-driven transactions won’t drive up fraud or chargebacks; second, consumers need better tools to set spending limits and manage credentials across agents. The bottom line? Agentic payments are still in their infancy, but with the biggest players on board and real infrastructure coming into place, QED believes adoption could scale quickly—once security, trust, and compliance catch up.
Current News
It’s been another breathless week in AI, with headlines dropping faster than anyone can keep up. Rumors are swirling that GPT-5 could launch as early as August—potentially bundling ChatGPT, Codex, DALL·E, and Whisper into one unified, much cheaper model. If true, it would mark a huge leap in multimodal AI—and a clear signal that the next wave of capability is just around the corner.
The capital markets are already betting on it. Anthropic is reportedly finalizing a $5B round at a $170B valuation (with ARR 4x’ing to ~$4B this year), while Groq and Cohere are raising new rounds at $6B+ valuations. On the infra side, Microsoft is said to be nearing a deal that would secure long-term access to OpenAI’s tech—even beyond 2030—locking in a strategic advantage just as the game shifts from “build LLMs” to “build agentic ecosystems.”
Meanwhile, the product side hasn’t slowed down either. OpenAI’s new “Study Mode” aims to calm educators’ nerves by guiding students with Socratic prompts instead of handing them answers. AWS is launching an agent marketplace, giving developers a direct line to enterprise customers eager to automate with bots. Even the IPO market is catching the fever: Figma’s $1.2B listing was reportedly 40× oversubscribed, highlighting just how strong public appetite remains for AI-native productivity tools.
And all of this is happening against a surprisingly supportive macro backdrop. With the Fed holding rates steady for the fifth consecutive meeting (at 4.25%–4.50%, decision out on July 30) the cost of capital remains unusually friendly for an industry sprinting to redefine the stack.
Put it all together—rumored model upgrades, massive funding rounds, strategic infra bets, product launches, and IPO heat—and the message is loud and clear: everyone serious about AI is retooling fast. The question now is whether the rest of us will keep up—or get left behind.
The Human Question
So here’s the big question: can we handle what’s coming?
We’ve seen growth spurts before—think Industrial Revolution—but that era didn’t exactly unfold in a democratic, digitally connected world. The Luddites, famous for smashing machines, didn’t have Twitter (or the vote). This time around, the pace of change is faster, and the tools are more powerful. If AI drives a massive leap in productivity, it could absolutely lift average wages—but it could also deepen inequality, trigger new political fault lines, and give states unprecedented tools to surveil or manipulate populations.
In short: more tension, more volatility, and a lot of pressure on governments to rethink how they tax, educate, and protect citizens in a world where intelligence itself is no longer human-only.
But for all the uncertainty, I still think there’s room for awe.
Dario Amodei, CEO of Anthropic, told The Economist that he believes AI will help us treat diseases we once thought untouchable. That’s a powerful frame: this moment doesn’t have to be a rupture. It could be the next chapter of a centuries-long miracle—where progress compounds because we dare to embrace disruption.
Yes, the thought of being outpaced by machines is unsettling. But that doesn’t mean we stop. It means we double down on what makes us human: empathy, ethics, and yes—wisdom. AI might outthink us in many ways, but we’re the ones who decide how to use it.
Let’s just make sure we’re ready!
You will see below an updated list of events, including for the second half of the year. If I forgot your event, and you want to include it - please sure to send those my way asap!
Cubo Conecta 2025
Date: September 2025 (exact dates to be announced)
Location: São Paulo, SP
Description: Celebrating its 10th anniversary, Cubo Conecta is the flagship event of Cubo Itaú, bringing together thousands of entrepreneurs, investors, and innovation leaders from across Latin America. The event offers over 90 hours of content and facilitates more than 5,000 digital connections, highlighting the potential of the technology and innovation ecosystem in the region.
Brazil Climate Summit 2025
Date: September 13–14, 2025
Location: New York, NY, USA
Description: The Brazil Climate Summit focuses on Brazil's role in the global green transition, emphasizing the importance of private sector involvement and international capital to accelerate low-carbon businesses. The event gathers leaders from various sectors to discuss sustainable development strategies.
More infoGartner CIO & IT Executive Conference 2025
Date: September 22–24, 2025
Location: São Paulo, SP
Description: A gathering of CIOs and IT leaders to discuss digital transformation, organizational leadership, and innovation strategies across industries.
More infoMoney 20/20 USA 2025
Date: October 26–29, 2025
Location: Las Vegas, NV
Description: One of the world’s leading fintech and payments conferences, Money 20/20 gathers global leaders across banking, payments, tech, and startups to explore the future of money and financial services.
More infoWeb Summit Lisbon 2025
Date: November 10–13, 2025
Location: Lisbon, Portugal
Description: One of the world’s most influential tech conferences, Web Summit Lisbon connects 70,000+ attendees from startups, enterprises, and media to discuss trends shaping the tech industry.
More infoBrazil Tech Summit 2025
Date: December 9, 2025
Location: São Paulo, SP
Description: Part of the Global Startup Ecosystem Series, Brazil Tech Summit brings together entrepreneurs, government leaders, and investors to foster tech-driven innovation across Latin America.
More info
See the hyperlinks above!
“I believe AI will play a crucial role in eliminating diseases like cancer and Alzheimer’s, while also extending human lifespan.” Dario Amodei