2026-10-28
By Vadym · Generated with AI, curated by me
Today’s stories are about capital efficiency, not capability: a GPU maker funding a startup that helps customers use fewer GPUs, a $10 billion fund betting on infrastructure over any single logo, and a warehouse robot expansion that's honest about not having proven ROI yet.
Bengaluru-based Simplismart closed an all-primary $14 million Series B led by Dallas Venture Capital, with Nvidia joining Accel, Shastra VC and Micromax at a roughly $100 million valuation. Its software sits between AI models and the GPUs running them, tuning deployments to cut compute costs — customer InVideo says it cut GPU serving costs by 56%. Annualized revenue grew 14x in six months, more than half of it from the US.
Hyphenate, founded by two Microsoft veterans in 2024, raised $45 million to build custom AI agents for finance teams handling accounting and strategic finance, per a Bloomberg report. Customers include Groupon and Synthesia, and the company competes directly with Ramp. Revenue has reportedly grown 86x since its $9 million seed round roughly a year ago.
Dallas-based Disruptive, founded in 2012 by Alex Davis, closed a $10 billion flagship fund — its first blind-pool vehicle — to write checks of up to $1 billion into at least 10 late-stage AI and defense companies. The firm’s past bets include Palantir, Groq and ElevenLabs, and it told LPs it has returned over $6 billion on roughly $1.9 billion invested.
Agility Robotics and GXO Logistics are expanding their three-year-old partnership, adding Digit humanoids at GXO sites in Statesville, Tennessee and Lebanon, Indiana, with Digit 4 units arriving in early 2027. Digit has already moved more than 100,000 totes and logged over 65,000 hours at GXO’s original Georgia site. GXO’s own CEO acknowledged on an August earnings call that the company hasn’t yet hit ROI on humanoids despite running 45 pilots.
Liquid AI released LFM3, six models sharing a 1.7-billion-parameter foundation pretrained on 22 trillion tokens, covering text, vision, speech and structured decisions, aimed at enterprises that want narrow workflows running on hardware they control rather than the cloud. Mercedes-Benz and Shopify tested pre-release. Liquid’s own documentation admits the smallest model “is not recommended for agentic coding or complex multi-step reasoning.”
None of today’s stories are about a model getting smarter. They’re about money getting more disciplined: a GPU maker funding GPU efficiency, a mega-fund betting on infrastructure instead of any single startup, a finance-agent startup growing revenue 86x on real back-office work, and a warehouse robotics deal expanding before its own customer admits the ROI math isn’t there yet. The money is still flowing — it’s just flowing toward whoever can make the unit economics work, not whoever has the flashiest demo.
— Boba
Curated by Vadym