Good morning. Today the business of super intelligence steps into the open, with a draft IPO filing that shows its true scale, a $5.66 billion marketing deal, and super intelligence moving deeper into drug development.

Anthropic’s draft IPO prospectus shows $16.3 billion in first half revenue. The Wall Street Journal reported on October 20 on a draft filing that puts first half 2026 revenue at $16.3 billion, up from $1.1 billion a year earlier, with third quarter revenue reaching $16.8 billion. The same draft shows an $80.7 billion net loss, though about $72 billion of that is a non cash accounting charge tied to financing instruments, alongside nearly $9 billion spent on compute in six months and future commitments above $600 billion. Why it matters: this is the clearest public look yet at the economics of a frontier lab, and investors will price the whole sector against it when Anthropic lists.

Salesforce agrees to buy Braze for about $5.66 billion. Announced on October 20, the definitive deal pays Braze shareholders $47.50 in cash per share, a premium of about 58 percent, and is expected to close in Salesforce’s fiscal 2028. Braze serves about 2,770 customers, including Canva, Instacart and Venmo, with tools that send personalised messages in response to customer behaviour in real time. Why it matters: Salesforce is buying the real time engagement layer its Marketing Cloud has lacked, and folding it into Data 360 and Agentforce tells us where enterprise software thinks growth now lives.

OpenAI acquires Orchestra Bio and moves into drug development. OpenAI will combine Orchestra Bio, a platform that helps biopharma teams coordinate studies, decisions and timelines, with Rosalind, its models and tools for life sciences research. The move takes OpenAI beyond finding promising molecules and into the long operational work of turning a discovery into a treatment. Why it matters: the value in life sciences sits in the years between a result and a therapy, and OpenAI now wants its software inside that process.

DeepSeek open sources DeepInfer as Google releases TorchTPU. On October 20, DeepSeek released DeepInfer, the in house inference engine that serves its V4.1 Flash model, as open source under the Apache 2.0 licence, with support for Huawei Ascend hardware built in. The same day, Google worked with Meta’s PyTorch team to release TorchTPU, which lets existing PyTorch code run on Google TPUs. Why it matters: both releases attack the same bottleneck from opposite sides, making capable models cheaper to serve and easier to run on hardware beyond Nvidia.

Meta’s Muse passes 10 million U.S. downloads in its first 40 days. Sensor Tower data reported by TechCrunch on October 20 puts the AI companion app among only about a dozen apps ever to reach that post launch mark, a list previously dominated by mobile games, Disney+ and Meta’s own Threads. The app is also holding users well against rival companions and chatbots. Why it matters: consumer super intelligence is becoming an app store race, and distribution at this speed is a moat no model release can match on its own.

A lifestyle program improved cognition 55 percent more than basic health advice. The LatAm-FINGERS study followed older adults at risk of dementia across 11 Latin American countries for two years, combining physical activity, healthy eating, cognitive training and social engagement, adapted to local cultures rather than simply translated. The structured program delivered the strongest gains, results presented at the Alzheimer’s Association International Conference 2026 in London and published in The Lancet. Why it matters: it is strong evidence that everyday habits, coached consistently, can protect the brain, and that the model works across very different communities.