World Bank's AI Leapfrog: A Beautiful Slogan. A Missing Grid.

January 2025. World Bank drops Global Economic Prospects. Global growth scraping a 30-year floor. Buried inside: developing economies should "rapidly adopt" AI. Not develop. Adopt. Skip the research phase. Skip the model-building phase. Rent the intelligence. Import the outputs. Run the state on someone else's compute. The report acknowledges the risks — inequality could widen, foreign dependency could lock in — then steps on the accelerator anyway. The press release omits something critical. This is the same leapfrog narrative that sold crypto to the Global South. Same structure. Same missing infrastructure. Same unasked question: who pays? I have watched adoption stories die on bad infrastructure for twenty-nine years. The 2017 Paragon sprint taught me speed wins the story. The 2022 Terra collapse taught me fast narratives lose when the collateral is fake. This recommendation carries red flags on every block. The World Bank's policy DNA is consistent: low capital intensity, high leverage, technical assistance over physical buildout. That is why the recommendation targets application-layer adoption — generative AI bolted onto existing public services — instead of sovereign foundation-model development. The math explains the choice. Training a 10-billion-parameter model costs between $1 million and $10 million in compute alone. That exceeds most low-income nations' entire annual AI budgets. So the route is predetermined: buy the intelligence. Do not build the brain. This is the mobile-money logic applied to artificial intelligence. In the 2000s, mobile payments let entire economies skip the credit-card era. Kenya's M-Pesa did not wait for a national banking grid. It went straight to phones. The World Bank sees AI skipping the PC era, the enterprise-software era, the whole legacy stack. But the leapfrog analogy hides a dirty secret. M-Pesa worked because basic infrastructure already existed. Cell towers were up. Demand was local. Regulatory space was permissive. AI leapfrogging has none of those guarantees. And here is the crypto parallel no policy shop wants to draw: Africa's mobile-money leapfrog worked because the rails were domestic. Crypto's leapfrog failed because the rails were offshore. AI's leapfrog has the same ownership problem — except the rails are not just offshore. They are opaque. You cannot audit a black box living in a foreign data center. Translation for readers of this newsletter: this is a growth play wearing a development hat. The Global South is not being invited to build AI. It is being invited to consume it. Consumption is the fastest route to measured GDP impact. It is also the fastest route to unmeasured dependency. The report's own hedges — inequality, dependence — point toward caution. Its action item points toward the vendors. Watch which one wins the implementation phase. Start with the ledger. ITU data from 2024 puts internet penetration in low-income countries around 36 percent. Sub-Saharan Africa's electricity access sits below 50 percent. Of the roughly 800 hyperscale data centers on Earth, Africa hosts under 2 percent. You cannot rapidly adopt a cloud-dependent stack in an economy where the grid collapses weekly. Generative AI inference happens in remote data centers. The end-user terminal is a smartphone. That is the architecture. And the architecture is the bottleneck. The thin-client path is the one saving grace. Smartphone penetration exceeds 60 percent globally. Mobile-first. Cloud inference. No local compute required. A farmer in Niger gets an AI crop advisor through a $50 Android device. No data center needed. No power plant needed. Just signal. Except the signal is the catch. Every query to a foreign AI API transmits local data across borders. This is data colonialism in its purest form: raw data exported, processed intelligence imported, value extracted at the border. The World Bank flags foreign technology dependence as a risk in one sentence. Then it recommends the strategy that guarantees that dependence in the next. Now watch what the headline coverage drops entirely: the open-source route. The economics are unambiguous. Closed API pricing is a perpetual tax. Developing countries cannot sustain dollar-denominated license fees while their own currencies bleed. Only open-weight models — Llama, Qwen, the broader open ecosystem — make adoption without subordination mathematically possible. The report implies the preference. It refuses to state it. That silence is strategic. A consensus machine cannot name open-source stacks without picking sides in the US-China AI standoff. Then there is the governance vacuum. Stanford's AI Index 2024 found roughly 10 percent of African countries have any national AI strategy. No frameworks. No guardrails. No accountability. Speed in a governance vacuum does not produce adoption. It produces extraction with a development-agency smile. The deeper constraint is absorptive capacity. Technology transfer only functions when institutions, data pipelines, and skilled labor can absorb it. The World Bank's own economics literature is dense with absorptive-capacity caveats. The policy headline drops them. Public institutions in the Global South lack structured data. They lack procurement capacity. They lack cybersecurity baselines. Failure costs run higher in fragile environments. A hallucinating crop advisor in a food-insecure region is not a bug. It is a policy outcome. Let me bring in my own crisis file. May 2022. Terra is a liquid black hole. I am on-chain tracking stETH exposure at three hedge funds that over-leveraged liquid staking derivatives as collateral. Every liquidation cascade visible in wallet data. Every analyst reading the narrative instead of the addresses gets destroyed. The lesson I carried out: when the collateral layer sits outside your control, your risk model is fiction. The Global South is being asked to put public services on collateral it cannot audit. Foreign compute. Foreign models. Foreign data centers. In a crisis, who rescues that stack? The World Bank's record compounds the problem. Its digital-infrastructure push redirected billions in the 2010s. Its financial-inclusion push built the rails for microcredit — which became debt traps in parts of South Asia. Policy endorsement moves capital. It is not a delivery mechanism. Three fault lines run through this recommendation. Fast versus stable. External versus internal. Optimism versus material reality. Call it rapid adoption or call it reckless exposure. The difference is whether the infrastructure ledger closes. Watch how this lands in practice. A government under fiscal pressure signs a cloud contract it cannot exit. The exit costs — data migration, retraining, procurement rework — lock the relationship in. In crypto we call that liquidity withdrawal. In development finance, it is called capacity building. Here is the angle the mainstream coverage is missing. This is not a development policy. It is a customer acquisition strategy for hyperscalers. Developing economies represent roughly 40 percent of global GDP at purchasing-power parity, with near-zero AI penetration. Every percentage point of penetration opens billions in addressable market. The World Bank committed over $100 billion in 2024 alone. When a lender that size endorses a direction, the wording migrates into loan covenants. AI readiness becomes a financing condition. And who supplies the readiness? AWS. Azure. GCP. Alibaba Cloud. Huawei Cloud. Consumption stimulus. Not supply-side buildout. The mandate creates users. It does not create producers. Around that mandate, a whole "AI for Development" industry is assembling — consultants, NGOs, tech vendors packaging "help the Global South adopt AI" into service bundles. Rent-seeking layer on top of a rent-seeking design. The unanswered commercial question sits underneath this entire recommendation: who pays? Low-income governments are broke. Aid budgets trend downward. Private-sector demand stays thin outside a few export hubs. The World Bank assumes a payment layer that does not exist. Same blind spot that killed half the DeFi projects I have audited — build the abstraction, skip the revenue loop. AI adoption without local revenue becomes another subsidy. Another dependency. Another report. The local tech sector gets squeezed in the middle. Local startups cannot compete with globally subsidized model access. Capital flows to foreign vendors while domestic capacity atrophies. That is not a leap. That is capture. Anyone who watched DeFi's liquidity-mining era understands the pattern. Subsidies inflate the headline numbers. Then they vanish. TVL printed by incentives. Users gone when the rewards stop. The World Bank's adoption push is the same architecture: incentives without foundations. The part that should make every crypto native uncomfortable: the only adoption that stuck in the Global South was not ideology-driven. It was inflation-driven. Venezuelans and Argentines did not adopt stablecoins for blockchain philosophy. They adopted them because local currencies melted. Survival drives adoption. Not narrative. AI adoption will move as fast as the pain demands — not as fast as the policy memo wants. And the paradox. The World Bank's dependency warning hands the Web3 crowd a ready-made narrative. Centralized AI creates structural dependency. Decentralized infrastructure — compute markets, data sovereignty layers, provenance rails — gets marketed as the Global South's alternative to hyperscaler capture. The honest read: most of those decentralized solutions cannot meet production load yet. But the marketing writes itself. Expect the AI-crypto crossover story to get much louder over the next twelve months. My 2021 Bored Ape liquidity work taught me to separate green flame from mechanism. The narrative here is green flame. The mechanism is missing electricity, missing bandwidth, missing governance. The report is a signal. Not a solution. Follow the money. Track three markers. Whether the World Bank opens a dedicated AI financing window within twelve months — words trapped in a report change nothing. Whether India, Indonesia, Nigeria, or Vietnam write AI adoption into budgets, not vision documents. And what share of new multilateral financing carries AI readiness conditions. Adoption without infrastructure is a sloganeer's dream. Slide decks do not survive power outages. The Global South does not need a leapfrog lecture. It needs a ledger that accounts for electricity, bandwidth, training, and governance — in that order. Until that ledger balances, rapid adoption accelerates one thing only: extraction. Wake me when the money follows the memo.