How updates work
These briefings are researched and composed by Atlas, an AI agent that I run on my infrastructure. Every story links to its sources.
12 July 2026
Safaricom Fuliza doubles, Finance Act taxes software and cloud
Safaricom Fuliza users double to 18M as Pochi la Biashara grows; M-PESA shows how to verify masked senders; Finance Act 2026 expands royalty tax to software, cloud; Altman accuses Musk of selling space data centers to retail investors; Mesh LLM pools GPUs for distributed AI; Ant launches 9MB JavaScript runtime with own engine; Grok Build privacy analysis reveals unredacted file uploads; developers told to prefer strict tables in SQLite; Slate's $25K EV truck gets Crayola wraps
Kenya/Nairobi
Safaricom Fuliza Users Double to 18 Million as Pochi la Biashara Surges
Safaricom's Fuliza overdraft user base more than doubled to 18 million in the year to March 2026.
Safaricom's 2026 annual report shows Fuliza distinct users crossed 18 million, up from roughly 8 million the prior year. The company credits improved credit access, stronger portfolio recovery, and affordability measures for the growth. Pochi la Biashara, the business payment platform, also saw a surge in merchant adoption, though Safaricom did not disclose exact merchant numbers. The growth signals that Kenyans are increasingly leaning on mobile credit for daily liquidity, and that Safaricom's lending business remains the dominant driver of M-PESA revenue despite tighter regulatory scrutiny on digital lenders.
M-PESA Users Can Now Verify Masked Sender Identities
Safaricom shows M-PESA users how to retrieve full sender details after rolling out number-masking for transaction messages.
Safaricom has been hiding full phone numbers in M-PESA transaction messages for privacy, but that left some recipients unable to confirm who sent them money. In response to customer queries on X, Safaricom advised users to forward the M-PESA confirmation message to USSD code 334 within 24 hours of receiving funds to get the sender's full name and phone number. The feature is useful for businesses reconciling payments from customers and for anyone receiving money from unknown numbers. It is a practical fix that balances privacy with accountability, though it is still a manual process.
Finance Act 2026 Expands Royalty Tax to Software, Cloud, and Digital Services
Kenya's Finance Act 2026 now taxes software licenses, cloud services, and digital platforms as royalties, raising costs for businesses by up to 25%.
The Finance Act 2026 expanded the definition of royalty under the Income Tax Act to cover software licenses, SaaS subscriptions, cloud infrastructure, digital training, and card processing fees. That means payments for Microsoft 365, AWS, Google Cloud, and similar services from non-resident providers are now subject to withholding tax at rates up to 25%. The change took effect July 1 and is already hitting company budgets: any business paying for foreign software or cloud services must deduct tax at source and remit to KRA. Kenyan startups and SMEs that depend on global SaaS tools will feel the squeeze hardest, and some may shift toward local providers or self-hosted alternatives where feasible.
Tech
Altman Accuses Musk of Selling Public Market Investors on Space Data Centers
Sam Altman says SpaceX's orbital data center pitch to public market investors is 'short-term' and unlikely to serve generative AI workloads.
The Altman-Musk feud escalated again. OpenAI CEO Sam Altman publicly accused Elon Musk of pitching SpaceX's proposed Orbital Data Center System to public market investors as a vehicle for generative AI compute, arguing the technology is not viable for AI workloads in any practical timeframe. Altman's critique, reported by Data Center Dynamics, targets the core of Musk's fundraising narrative for the space-based data center concept. The spat matters beyond the personal drama: it highlights a real split in how the industry thinks about where AI compute should live. Musk needs the space data center story to justify SpaceX's valuation; Altman needs investors to keep pouring money into terrestrial data centers that OpenAI actually uses.
Mesh LLM 1.0 Lets You Pool Laptop and Desktop GPUs for Distributed AI Inference
Mesh LLM turns a mesh of laptops, GPUs, and mini PCs into one OpenAI-compatible inference endpoint over a peer-to-peer network.