Three moves this morning are all about leverage over the stack you build on — who you're allowed to build with, whose model becomes the default, and where the next billion flows — and each hands you a same-day job. OpenAI is cutting off Cursor's access to its models because SpaceX now owns it. Salesforce made Claude the default AI across Slack and its own products. And the biggest new fund of the week is a16z's $1.1B bet on AI hardware, not software. Here's the whole edition in one screen, and the one thing to do about each:

The through-line: none of these is a new model or a new benchmark. All three change who holds leverage over the ground your product stands on. Here's what each means, and what to do before the week ends.

1. OpenAI is pulling its models from Cursor — because a rival owns it now#

On Aug 28, 2026, OpenAI notified SpaceX that it will stop serving its models to Cursor on Nov 12, the maximum notice its contract allows. The trigger was ownership: SpaceX's all-stock acquisition of Cursor-maker Anysphere closed in August, and OpenAI now competes, through xAI and Musk's other ventures, with the company that controls one of the most popular AI coding tools.

OpenAI's stated reason is trust, not capability. It says it "cannot be confident" SpaceX will use its models within OpenAI's terms of service, and it pointed to a track record: X breached a contract after Musk acquired it, and Musk acknowledged under oath this year that xAI had violated OpenAI's terms. This is two companies on opposite sides of a public rivalry, and the coding tool is caught in the middle.

What it means: Cursor does not break — it still serves Anthropic's Claude, Google's Gemini, and xAI's Grok, and its CEO put direct OpenAI access at roughly 5% of the tool's traffic. Anthropic said the same day it would add compute so Claude runs well there. But the lesson is bigger than one tool: if a workflow you depend on is pinned to a single model provider inside someone else's product, a boardroom feud you have no part in can revoke it with 60-plus days' notice. Keep your coding setup model-portable — favor tools that let you swap providers in a dropdown, and don't hard-code prompts or evals to one vendor's quirks. We keep a running comparison of the terminal and IDE coding agents if you want to know what your fallbacks actually are.

2. Salesforce made Claude its default — and put itself inside Claude#

Salesforce and Anthropic announced "Claudeforce" on Aug 26–27, 2026, and the word "default" is the whole story. Per Salesforce's announcement, Claude becomes the default model for Slack AI, Slackbot, Agentforce Coworker, and Claude Code across Salesforce's engineering organization, and it is the first LLM fully integrated inside the Salesforce Trust Boundary. The partnership also ships "Salesforce in Claude," a plugin with 37 prebuilt sales skills that let sellers reason over live CRM data and take governed actions from inside Claude — in pilot now, with open beta expected in September 2026.

The money makes the commitment concrete: Salesforce plans to spend about $300M on Anthropic tokens in 2026, on top of an existing roughly $300M equity stake. Its stock jumped after the news.

What it means: If your customers or your own team live in Slack and Salesforce, a frontier model is now reading and acting on that data by default — and the era of "we're model-agnostic" enterprise AI is quietly ending at the top of the market. Two same-week actions. First, check what your platforms' "default AI" now is and what it can see; a default that ships on is a default most teams never change. Second, if you sell into the Salesforce ecosystem, the surface for your product may now be a Claude plugin or an Agentforce skill, not only a native app — worth knowing before you build your next integration against last year's assumptions.

3. a16z's $1.1B fund bets on silicon, not software#

On Aug 28, 2026, Andreessen Horowitz announced a $1.1B "Machine Age Fund" aimed at the physical buildout of AI: semiconductors, memory, networking, storage, data centers, robotics, and even home AI appliances. For a firm that built its name on software, that is a pointed move — capital flowing toward the compute and physical layer rather than another application tier.

What it means: Read it as a signal about the fundraising climate you're pitching into. The rotation toward infrastructure means hard-tech, devices, energy, and compute startups have a fresh, deep-pocketed buyer for their story. The flip side is the warning for everyone else: a thin software wrapper around someone else's model — exactly the kind of product the first two items in this edition just showed can lose its model access or get absorbed into a platform default — is a harder sell than it was a year ago. If that's you, the pitch that lands now is the durable thing you own: proprietary data, a workflow moat, a distribution edge, or margins that survive your model provider changing its terms. The cost of that compute is still the number under everything; our GPU rental price map tracks where it's headed.

The common thread across all three: the models keep improving, but this week the news was about leverage — who can revoke your access, whose model becomes the default under your customers, and which layer the smart money now thinks is scarce. Those are the terms a founder has to plan around, whatever model you're prompting this morning.