AI News Roundup — Week of July 19, 2026: SME Contractors Get Tier-One Leverage, RealPage Buys the Data Layer, and China Crashes the Frontier

UK startup Prolo closed an oversubscribed £4.2 million seed round led by Triple Point, with participation from an Andreessen Horowitz scout fund, Anamcara Capital, Concrete Ventures, and several other early-stage investors. Prolo operates as an outsourced, AI-powered procurement team for small and mid-sized construction firms: contractors place orders by WhatsApp, email, or phone, and the platform combines AI with human procurement specialists to source pricing from a network of more than 185 suppliers — including trade rates on bulk materials and specialist plant hire normally reserved for the largest contractors.

CEO James Morris-Manuel framed the problem plainly: SME contractors "often pay higher prices because they lack the purchasing power of tier-one contractors."

RealPage, the property management software giant, completed its acquisition of Cherre, the real estate data intelligence company that built its business connecting fragmented property data into a single governed layer for institutional owners. RealPage CEO Dirk Wakeham said the deal means "every customer, whether they manage one property or a global portfolio, gets access to a stronger, more trustworthy foundation for their decisions." Cherre founder L.D. Salmanson described the industry's moment as moving "from reporting to reasoning."

Moonshot AI's newly released Kimi K3 entered the top tier of global AI performance, surpassing leading US models on some coding and text benchmarks while costing substantially less. The Chinese lab also plans to release Kimi as an open-weight model, letting organizations customize and run it on their own infrastructure. Its arrival challenges the assumption that China trails the US frontier by months — and puts direct pressure on American providers' pricing.

Two data points landed this week that turn "AI visibility" from theory into arithmetic. First, Invoca's analysis of more than 70 million phone conversations found that calls originating from ChatGPT converted into sales leads 49% of the time — roughly 10 percentage points better than every other measured channel. Volume is still small, but callers arriving via AI are demonstrably further along in their buying decision. Second, LinkedIn published its first platform-specific playbook for getting cited by AI search systems: open with a clear keyword-rich first sentence, skip hashtags in the first line, target 200–300 words for posts and 800–1,200 for articles, and favor educational content over promotion.

Mira Murati's Thinking Machines released Inkling, its first open-weight foundation model, positioned not as a benchmark-topper but as a base that organizations fine-tune with their own expertise through its Tinker platform — the bet being that a model trained on your knowledge beats a general-purpose one for your workflows. The same week, Microsoft CEO Satya Nadella warned that firms risk gradually surrendering proprietary knowledge to AI providers through everyday prompts, corrections, and workflows — a "reverse information paradox" in which the vendor accumulates your institutional expertise over time.