Enterprise AI Newsletter Archive

8 min read

The GPU landlords start to fall

Industry insiders now expect only a handful of the hundreds of AI cloud firms renting out GPUs to survive, and Nvidia is propping up the rest with its own money.

The Physical Layer

Amazon, NVIDIA, AMD, and the CIA’s investment fund announced today they backed a world-model startup with $310 million. World models — AI systems trained on physics and object relationships rather than language — are the foundation for robots that train in simulation, factories that test production changes without halting operations, and logistics networks that model real-world constraints before deployment. The language model stack that enterprise leaders have spent three years building is not being replaced. It is getting another layer beneath it.

The Condition of Access

The past week has produced a specific new framework for enterprise AI: model access is explicitly conditional. Fable 5 is offline on government terms — fix a named jailbreak vulnerability, then access will be restored. OpenAI extended GPT-5.5-Cyber to the EU through a formal compliance program. GitHub's consumption billing moved enterprise AI usage to the token level. Microsoft's Work IQ API went generally available today. AI tools are no longer simply on or off. They come with terms.

The 72-Hour Model

Claude Fable 5 launched Tuesday. The US Department of Commerce ordered it offline Friday. Every customer worldwide — enterprise and individual, US-based or not — lost access simultaneously, with no advance warning and no operational continuity provisions. As of this morning, both Fable 5 and Mythos 5 remain offline. The first government recall of a deployed frontier AI model defines a risk category that did not exist in enterprise AI planning frameworks ten days ago.

The Public Market Phase

SpaceX began trading on Nasdaq this morning. Anthropic, which this week disclosed it pays $1.25 billion per month to rent xAI compute, is simultaneously worth $965 billion, has crossed $47 billion in annualized revenue, and has confidentially filed for an IPO. OpenAI is behind it in the same pipeline. The AI companies enterprise leaders have built their strategies around are entering the public market. The information environment is about to become significantly more legible — and the compliance clock is ticking.

The Infrastructure Ledger

SpaceX priced the largest IPO in recorded history today — $1.75 trillion, $75 billion raised. The S-1 it filed is also the most complete public accounting of AI compute economics ever produced. Google pays $920 million per month for compute. Anthropic pays $1.25 billion per month. On the same day, OpenAI plugged into Oracle's enterprise procurement network, and JPMorgan disclosed the enterprise AI productivity benchmark the market has been waiting for. The infrastructure economics of AI have been opaque. Today they are in the public record.

The Capability Threshold

Anthropic released Claude Fable 5 yesterday — the first publicly available Mythos-class model, the tier above Opus that has existed only in restricted access since April. Stripe reported it compressed months of engineering into a single day. The same week, Google is paying SpaceX $30 billion for GPU capacity because there is not enough compute in the world to meet demand, and OpenAI filed its IPO. The capability crossed a new threshold. The infrastructure race to run it at scale is just as consequential.

The Compute Scarcity

Google is paying SpaceX $920 million per month — $30 billion total — for GPU access at xAI's Colossus data centers. Google has its own vast infrastructure. It is still renting compute from a competitor because there is not enough AI compute in the world to satisfy demand through internal build-out alone. Meta is spending $125-145 billion on infrastructure this year while cutting 8,000 employees. OpenAI filed its confidential S-1 targeting a $1 trillion-plus valuation. The compute scarcity that was described as theoretical 18 months ago is now showing up as nine-figure monthly payments to rivals.

The Rationalization Phase

Glean hit $300 million in annual recurring revenue by selling AI budget cuts — enterprises are consolidating tools, not expanding them, and Glean wins by connecting AI to organizational knowledge. AlphaSense crossed $600 million ARR with 70% of the S&P 500 as customers by solving the same problem for financial intelligence. DeepSeek is about to raise $7.4 billion, proving open-source AI is no longer cheap. The enterprise AI market has entered its rationalization phase: the experiments that didn't connect to organizational knowledge are being cut, and the platforms that did are compounding.

The Device Layer

Apple opened WWDC today by announcing that Siri has been rebuilt on Google Gemini — a custom 1.2 trillion-parameter model licensed at approximately $1 billion per year. iOS 27 ships to 1.4 billion active iPhones. Users can choose between Gemini, ChatGPT, and Claude as their AI engine. AI is no longer a tool employees choose to adopt — it is now a layer in the operating system they already carry. The enterprise AI adoption question just changed character entirely.

The GPU landlords start to fall

The Physical Layer

The Condition of Access

The 72-Hour Model

The Public Market Phase

The Infrastructure Ledger

The Capability Threshold

The Compute Scarcity

The Rationalization Phase

The Device Layer

Spearhead

Company

Work

Programs

Insights