Wall Street just made compute an asset class | Thursday, August 13, 2026

Google said the standalone Gemini app crossed one billion monthly users, its fastest-growing product ever, a scale it reached mainly by being the default assistant on Android and stitched into Search and Workspace. The count excludes AI Overviews and Workspace embeds, so the real enterprise footprint is larger and already inside tools you license. The lesson for a CIO is that the model your team standardizes on and the one your employees actually open are drifting apart, and distribution is deciding the second. Audit what your people already use before you run another bake-off, because the default has probably made the call.

OpenAI's Presence platform packages voice and chat agents and sells them as engagements led by OpenAI's own Forward Deployed Engineers and a set of global systems integrators, with early customers including BBVA Mexico, SoftBank, and insurer IAG. Each deployment starts with one job, a billing dispute or an IT ticket, and the customer writes the rules for sign-off and handoff. This is a departure for a company that ran on API keys and seat licenses. The model vendor now wants the integration work too, which moves the build-versus-buy line and puts your foundation-model supplier into competition with your systems integrator. Disclosure: OpenAI is a Spearhead technology partner, and the harder read, that this narrows a customer's independence, is the one we ran.

OpenAI shipped GPT-5.6-Cyber and made its Daybreak cyber models available through Amazon Bedrock, putting frontier security capability in the hands of approved defenders. In the same week, OpenAI and Anthropic faced pressure to explain AI-driven hacking sprees that ran for weeks before anyone noticed. Both things are true at once: the same capability that helps your security team hunt vulnerabilities is the capability regulators worry about most. For a CISO, the practical move is to treat frontier cyber models as dual-use by default and to assume your adversaries have the same tools you just got. Disclosure: OpenAI and Anthropic are Spearhead technology partners.

AWS signed a multi-year deal with Superblocks to run its generative app-development platform inside a customer's own AWS environment, integrated with Bedrock, so corporate data never leaves the tenant. The pitch is governed "vibe coding": business users build internal apps while IT keeps its network controls, and policy agents check AI-generated code against enterprise standards before it reaches production. A Smart Router on Bedrock claims up to 30% savings by sending each task to the cheapest model that can do it. For a regulated buyer, this makes data residency and code governance a default rather than a project, which is how AI app development becomes safe enough to hand to non-engineers.

DeepSeek released the weights for V4-Flash under a permissive MIT license: a 284-billion-parameter mixture-of-experts model with 13 billion active parameters per token, a one-million-token context window, and API pricing near $0.14 per million input tokens. It matches strong proprietary models on coding and agentic benchmarks at a fraction of the cost. Meta put a laptop-grade open model on desks last week; this puts a data-center-grade one in your own environment with no license negotiation. The open option is not a discount tier anymore, it is a credible default, and it keeps getting cheaper. Every open release like this resets the build-versus-rent math for workloads you would rather not send to an API.