OpenAI offered Washington 5% of the company
OpenAI proposed handing the US government a 5% equity stake and suggested other American labs adopt the same arrangement, days before the White House is expected to announce voluntary release standards negotiated with OpenAI, Google and Anthropic. After a June in which the government held back GPT-5.6 and briefly froze Fable 5, the labs are choosing to make Washington a shareholder rather than fight it as a regulator. For an operator the read is simple: frontier model availability is now partly a political product. Keep a second provider wired in, because the next launch delay may come from a negotiation, not a lab.
Source: Original source
Anthropic quietly took the revenue crown from OpenAI
Anthropic is running at roughly $47 billion in annualized revenue against OpenAI's reported $25 to $33 billion, passed it in business subscriptions back in May, and ChatGPT's share of chatbot web traffic slipped below half for the first time. The money is following coding and agent workloads, which is where enterprises actually pay, while consumer chat flattens. The practical takeaway is that this is now a real two-vendor market at the top with Google closing fast. If you locked into one provider a year ago out of inertia, you have negotiating leverage today that you did not have then, and it is worth using it.
Source: Original source
Microsoft is betting $2.5 billion that deployment, not models, is the bottleneck
Microsoft launched Frontier Company, a new division backed by $2.5 billion that embeds 6,000 forward-deployed engineers, consultants and industry specialists directly inside client companies to get AI systems actually running. The same week, 8090 Solutions raised $135 million led by Salesforce Ventures for human-supervised agent platforms. The largest software company on earth just said out loud that the constraint is no longer model quality, it is the last mile inside the business. If Microsoft needs 6,000 humans to close that gap for enterprises, the same hands-on deployment work is wide open for smaller operators at local scale.
Source: Original source
The labs are building their own chips to make your tokens cheaper
OpenAI announced a custom inference processor with Broadcom, nicknamed Jalapeno, and Anthropic opened talks with Samsung about a custom chip of its own. This is happening because chip supply is the binding constraint on AI buildout in 2026, and inference cost is the single biggest lever on every lab's margins. Custom silicon takes years to pay off, but the direction is set: the price per token keeps falling, and the labs are spending billions to make sure it falls faster. When you price an AI feature or an automation for a client, price it on the curve, not on today's bill.
Source: Original source