Open source isn't a nice-to-have for personal AI assistants. It's the only structure where "your data is yours" is a technical fact instead of a clause in a policy document.
We're entering an era where assistants know more about us than most people in our lives do. Handing that to a single vendor's closed stack isn't innovation — it's just deferred risk.
AI is becoming part of everyday life. But whose everyday life are we talking about?
Some markets still lack products that fit their needs. Low adoption gets mistaken for low demand. Investment then flows toward markets already gaining traction, while underserved markets stay overlooked. And the cycle continues...
Frontier research matters. So does expanding who benefits from it.
Markets we call "small" or "not ready" may simply need the right products. They could hold some of the biggest opportunities in AI.
Which markets do you think the AI industry is underestimating?
8,800 to 1.
For every 8,800 English pages on the open web, there's just one in Assamese, a language spoken by 15 million people in Northeast India. It shows exactly why AI struggles with low-resource languages. Not because the languages are complex. Because the training data barely exists.
Three months in, our team ranked #1 across all six evaluation metrics in four of WMT26's most data-scarce language translation tracks. A look at the problem we're solving: https://t.co/4yaWdG0yNw
#NLProc #LowResourceNLP #WMT26
The AI infrastructure narrative: hyperscalers spending trillions on data centers.
What's missing: 70% of the world's population still lives in markets where cloud access is limited, expensive, or intermittent.
The next billion AI users won't arrive via a data center. They'll arrive via a $200 phone with a capable on-device model in their pocket.
That's not a low-end play. It's the biggest addressable market in AI.
#EmergingMarkets #AIForAll
"AI phone" is often dismissed as a marketing gimmick in mature markets. In emerging markets, it's a necessity.
The logic:
• Average income → no disposable GPU budget
• Patchy electricity → cloud inference is unreliable
• Poor latency → real-time AI must live on device
AI phones in Africa, South Asia, LATAM aren't "premium features." They're the first affordable AI infrastructure millions will ever touch.
#OnDeviceAI #EmergingMarkets #AIInfrastructure
Africa has less than 1% of the world's data centers.
The instinct is to build more cloud. The smarter bet is on-device.
When connectivity is unreliable and cloud latency is real, the edge IS the infrastructure. That's why AI phones matter more in emerging markets than anywhere else — not as a premium add-on, but as the primary delivery layer.
The race isn't who builds the biggest data center in Lagos. It's who ships the most capable device.
#AIinEmergingMarkets #OnDeviceAI
The new era breaks induction — past data no longer predicts the future. But deduction isn't dead; its bar just got higher. What really fails is "induction + inertial extrapolation," the industrial-era default OS. The replacement: anti-fragile experiments, signal sensing over prediction, and scenario planning that maps multiple mutually-exclusive futures — each with its own leading indicators. Sense and respond, don't predict and pray.
I run AI at Transsion — the company that sells more smartphones in Africa than anyone else. So I think about this daily.
The "leapfrog" narrative says: Africa skipped desktop straight to mobile, and now it'll skip straight to mobile-first AI. Beautiful story. Partially true — but mostly self-serving for Silicon Valley.
Here's the reality: The $100-150 devices most first-time smartphone buyers use don't have the NPU horsepower for on-device AI that matters. A Snapdragon 8 Gen 3 pushes ~40 TOPS. Entry-level chips? Forget it. You're running quantized 1-3B models that feel like a toy compared to what premium users get.
The uncomfortable truth is that "mobile-first AI" often becomes a euphemism for "good enough for them." M-Pesa was a true leap because the infrastructure (SMS + agent network) already existed and the use case was simple. AI is a full-stack problem — compute, data, latency, localization. Cut any corner and you ship a demo, not a product.
I'm not saying mobile AI is wrong. I'm saying leapfrog narratives conveniently absolve the industry of asking: are we shipping genuine capability, or just a differentiated price point dressed as innovation?
What makes me optimistic: the hardware race is real. Dimensity and Snapdragon are pushing AI capabilities down the stack fast. The question is whether the software stack keeps pace — and whether we're willing to call out the gap rather than paper over it with storytelling.
Everyone's talking about the Anthropic export ban as a US-China story.
The biggest loser isn't Beijing — it's Lagos, Nairobi, and Jakarta.
June 12, the US Commerce Dept ordered Anthropic to cut foreign access to Claude Fable 5 & Mythos 5. JPMorgan already blocked Hong Kong staff.
Frontier AI is becoming a licensed good, not a public utility.
For Global South users with 200ms latency, 5GB downloads, unreliable cloud — this changes nothing today. Everything tomorrow.
The counterintuitive truth:
→ The ban doesn't slow down China's AI.
→ It accelerates the Global South's shift to edge-first architecture.
Africa's $240B mobile economy doesn't need Fable 5. It needs a model that fits in a pocket, speaks Hausa, and works offline.
Are you architecting for sovereignty from day one — or betting on API access that can be revoked overnight?
Everyone's hyping Midjourney Medical's full-body ultrasonic CT scanner. 60 seconds. Radiation-free. AI-powered. Spa-like experience.
Cool tech.
But it's not democratizing healthcare. It's luxury healthcare for people who already have good healthcare.
The real AI opportunity in emerging markets for medical imaging isn't building a better scanner. It's making the $5,000 ultrasound that's already sitting in a rural clinic actually useful—by giving it an AI radiologist that doesn't need a $300K salary and a city apartment.
That's where the impact is. Not in the spa pod.
Africa's mobile economy hit $240B in 2025 — 7.8% of GDP. On track for $290B by 2030.
But here's the part most miss:
The next $50B won't come from laying more fiber. It'll come from closing the "usage gap" — the millions already within coverage who aren't online yet.
Affordability. Digital literacy. Relevance of services.
This is where on-device AI becomes the bridge, not infrastructure.
A $100 smartphone with local-language AI can do more for inclusion than another $10B in undersea cables.
The Global South's AI story isn't about frontier models. It's about the device in someone's pocket.