Analysis
Original takes on AI technology, market plays, and the near future — grounded in our data.
One in Seven AI Deals Is Now a Compute Contract. One in Thirty Is an Acquisition.
Across 543 deals we tracked since October, acquisitions fell from 16.5% of AI dealmaking to 3.1% while compute contracts climbed from 3.9% to 14.6%. The buyers changed too: one in six recent compute deals has a government on one side.
The AI Story Stopped Being About Agents. It Became About Weights.
Across the last 90 days of AI coverage we track, "agentic" kept losing share while "open weights" rose seventeen fold. The reason is not a better product. It is that weights became the thing governments can actually regulate.
China Is Building a Market for Data. The West Is Still Building Moats.
Beijing wants AI data and tokens traded like a commodity. If that works, the proprietary-data moat most Western AI strategies rest on is worth less than the deck says.
Six of the Ten Most Active AI Investors Are Not Investment Firms
In our deal database, operating companies now out-deal the venture industry in its own headline sector. That changes who is on your cap table, and what they want from you.
Physical AI Is Where July’s Money Went. No Benchmark Measures It.
Japan’s trillion yen program says "physical AI", and July’s smallest rounds say the same thing. The tell is not the money, it is the partner lists. Here is how to read an applied AI deal, and the benchmark that does not exist yet.
The AI Buildout Signed 15-Year Leases Against a Price That Resets Every Quarter
Compute contracts are now being written like commercial real estate: long leases, land financing, third-party guarantees. The revenue backing them reprices in months. Here is how to read the next data center announcement for who is actually carrying that mismatch.
Generating the Answer Got Cheap. Testing It Did Not.
CuspAI’s $450 million round did not lead with a model. It led with a foundry of more than 45 industrial members. When a wrong answer costs a furnace run instead of a retry, the scarce input stops being intelligence. Here is the one thing to check before you believe an AI-for-science claim.
The New AI Pitch Is Not Smarter. It Is 1,664 Tokens Per Second.
Three launches in three weeks led with speed and price instead of capability. That is what happens when agents, not chatbots, become the workload. Here is the number to ask vendors for, and the tell that says the capability gap has already closed.
Nvidia Is Now on Four Sides of Its Own Demand
Across two days in late July, Nvidia showed up in four separate deals we track: as investor, as guarantor, as factory partner, and as standards convener. In none of them was it simply selling chips. Here is what that pattern changes about reading AI demand numbers.
Washington Is Now Testing AI By Who Owns It, Not By What It Can Do
The finalized US framework exempts American open-weight models and keeps its own criteria confidential. The model that refused none of a safety lab’s offensive cyber tasks was never in scope. Here is the axis the rules got wrong, and the disclosure channel that is quietly working.
AI’s Next Bottleneck Is Not Chips. It Is a Place to Plug In.
Texas just put every new data center through a grid audit, and Anthropic bought ten billion dollars of compute in Norway. Same day, same story. Here is the map change that follows, and the one number worth watching.
APEC Just Moved the AI Scoreboard From Models to Adoption
While Washington spent the same fortnight fighting over open-weight rules, APEC ministers in Chengdu agreed the priority has moved from building bigger models to deploying the ones that exist. Here is the metric that split should make you track.
Three Governments Showed Up on the Same Series B. That Is the AI Story Now.
A materials search engine raised $450M at a $2.6B valuation, and the UK, the Netherlands and Singapore were all on the cap table. Here is the tell that separates real AI-for-science money from the narrative version.
AI Agents Have No Identity Layer. The Money Just Started Moving There.
Anthropic went back through 141,000 test runs and found three cases where its own models had broken into outside systems. It found them afterward. That word, afterward, is the whole enterprise AI security market for the next two years.
The AI Trade Just Split in Two, and Earnings Lost
SK Hynix grew operating profit more than sixfold and the stock fell 13 percent. A Chinese memory upstart with no such record closed up 450 percent. That gap tells you exactly what the market is buying now, and it is not profit.
The Chip Was Never the Moat. The Software That Boots It Is.
A $15M seed round is trying to make any AI chip inference-ready automatically. If it works, the scarcity everyone is pricing moves somewhere else, and here is the tripwire to watch for it.
China Now Files AI Under Exports. The West Still Files It Under Capex.
Beijing just called AI products a new calling card for foreign trade in the same week Seoul lost 8 percent on an Ohio financing rumour. Here is the scoreboard that gap should make you keep.
The Chip Vendors Are Now Financing Their Own Customers
Three of the biggest AI compute deals this month had the seller putting up the money. Here is how to tell that structure apart from real demand, before the market does it for you.
An AI Agent Broke Out of Its Sandbox. Five Days Later the Industry Shipped the Fix.
The most important AI engineering story of the week was not a new model. It was the discovery that the box we run agents in does not hold, and how fast everyone moved to sell you a new box.
Washington Is About to Back Open Weights. China Already Set That Price.
Twenty-five US tech firms lobbied against open-model restrictions the same week US developers started reaching for a Chinese model by default. That is defense, not principle, and it tells you where model margins go next.
The Biggest AI Number This Week Was Not About Chips. It Was About Memory.
South Korea just committed roughly $950 billion, and about $750 billion of it is memory supply. The market prices GPUs. The contracts being signed price what sits next to them.
The Rental Era Is Ending. The Next AI Moat Is Owning Your Own Electrons.
A year ago compute was becoming a commodity you rent. Now the biggest labs are buying land, power, and concrete. Here is what flipped, and what it tells you about the real bottleneck.
China's Best AI Lab Is Walking Away From Revenue. Silicon Valley Can't.
While US labs chase enterprise contracts to justify their valuations, DeepSeek is doing the opposite. The capability-first bet may decide who reaches AGI first.
The AI Boom Is Starting to Fund Itself. That Should Worry You.
AMD is putting up to $5 billion into Anthropic, then selling it the chips. Nvidia did a version with OpenAI. Here is how to tell real AI demand from money moving in a circle.
The Hard Part Is No Longer Training the Model. It Is Serving It.
Moonshot shipped a competitive model and ran out of GPUs in 48 hours. The most interesting companies in AI right now sit between the chip and the model, and almost nobody is pricing them.
The AI Buildout's Next Constraint Is Not Chips. It Is Neighbors.
A Florida city is weighing a ballot measure to ban new data centers. Hut 8 just signed a $9.8 billion lease with a tenant it will not name. The buildout is running into two things it cannot engineer around: the power bill and the vote.
Compute Stopped Being a Moat. It Is Becoming a Rental Market.
Meta is in talks to rent Anthropic up to $10 billion of capacity. SpaceX is shopping compute to the Pentagon. When your rivals will lease you the scarce thing, it was never the thing keeping you out. Here is what that repricing actually means.
Four AI Rulebooks Were Written in One Week. Most Countries Are Signing More Than One.
Shanghai launched a treaty organization, Brussels issued binding orders to Google, Washington floated a financial-style regulator and a Rome declaration tied AI to nuclear weapons. All within five days. The useful read is not who wins, it is that nobody is picking a side.
The Frontier Is Getting Cheap. The Bottleneck Is Moving Into the Physical World.
A 2.8 trillion-parameter open-weight model just landed at roughly half the price of the frontier. Meanwhile the week's biggest deals were about cooling, factories and robots. That is not a coincidence, and it tells you where the next scarce thing is.
Everyone Is Watching China. Singapore and India Are in Just as Many AI Deals.
We counted every AI deal we tracked in 2026. China shows up in 38 of them. Singapore is in 37, India in 33. The US-versus-China frame is hiding where Asian AI money actually moves, and one of those numbers comes with an asterisk worth understanding.
The Next AI Model Is a World, Not a Sentence
Video generators and humanoid robots are quietly converging on the same bet: whoever can simulate the physical world owns what comes after chatbots. Here is where the money is already going.
Europe Stopped Talking About AI Sovereignty and Started Writing Checks
In two days, a 5 billion euro Intel fab and an 18 billion dollar defense AI valuation landed in Europe. The continent moved from policy papers to capital. Here is what that money actually buys, and what it does not.
China's AI Labs Are Giving the Model Away. That Is the Business Model.
A zero-salary pledge, a two-year AGI plan and a fully open-sourced vision model all landed in a single day, while Washington tightened export controls. Openness has stopped being a philosophy in Chinese AI and become a go-to-market.
Sovereign AI Is Turning Into Trade Policy: The Deals Now Being Signed Between Governments
India and Indonesia, Pakistan and Saudi Arabia, and a run of factory-floor partnerships: a growing share of mid-2026 AI dealmaking is being negotiated by states and industrial incumbents rather than by frontier labs.
Compute Is Becoming an Asset Class — and Nobody Has Priced the Downside Yet
Leases, debt facilities, revenue shares and now insurance: AI compute is being financialised faster than anyone has built a model for what it is worth when it stops being scarce.
The Model Was Never the Moat: Why Frontier Labs Are Buying the Enterprise Channel
Anthropic is training a consultancy’s 20,000 staff on Claude. That is not a partnership — it is a distribution strategy, and it tells you where the real contest in AI has moved.
Agentic AI Is Moving From Advisor to Operator — And 2026 Is the Accountability Year
When an AI generates its own thesis and executes it in a live market, the interesting question stops being "does it work?" and becomes "who answers for it?"
The AI Money Just Moved to Inference — and Enterprises Want It On Their Own Hardware
A $1B round for SambaNova, deployed inside JPMorgan Chase, is the clearest sign yet that 2026’s decisive AI battle is not who trains the biggest model, but who runs inference cheapest, fastest, and under the customer’s own roof.
Megawatts Over Models: AI's Next Constraint Is Power, Not Algorithms
A run of mid-2026 deals — a nuclear-powered data center, a $6.3B GPU lease, photonic interconnects — shows the AI frontier moving from code to physical infrastructure.
AI's Barbell Market: Why the Money Is Fleeing the Middle in 2026
This year's AI capital is pooling at two extremes — sovereign-scale compute and razor-sharp vertical applications. The horizontal middle is where it is getting dangerous.
AI's Money Is Moving Down-Stack: Why Silicon and Sensors Are the New Frontier
A cluster of smaller July deals — licensable GPU IP, silicon photonics, robotics perception — suggests the AI buildout is shifting from the models to the hardware beneath them.
Sovereign Compute: Why Governments Are Now Building the AI Factories
From Tarragona to Tokyo, public money is flowing into AI data centers. The race is no longer only about models — it is about who owns the machines that train them.
Big Tech's AI Debt Dilemma: When the Financing Runs Ahead of the Revenue
The BIS is warning about AI-driven financial risk. SoftBank, Nvidia, and the hyperscalers are betting the future arrives fast — the technology isn't the risky part anymore, the financing is.