Nvidia becomes the bank of AI — now lending the money that buys its own chips· 8h ago

Zuckerberg's 6,500-word essay places AI alongside the industrial revolution and electricity, sketching a future in which superintelligence lifts billions. The same day, Meta let go of Muse Glimmer, its most powerful model, in a form anyone can download and modify. Put the grand vision and the giveaway on the same page, and the second reads like proof of the first.

But an open release is a market move before it is an ethical one. Against rivals who lead on API revenue, Meta binds developers to itself by handing out the weights, reaching for a de facto standard. Releasing a top-tier model in a freely modifiable state runs against the safety debate regulators are trying to pin down right now. That the essay spends so little on safety is no accident — spell it out, and the gift loses some of its shine.

One writer called superintelligence a dragon. Anyone who claims it can be tamed should show the cage's blueprints first. Count how many of the 6,500 words went to that blueprint, and the centre of gravity of this announcement comes into view.

Today's page was pulled by the weight of capital. Nvidia's $500bn, Intel's $15bn raise, Singapore's upgraded forecast, leveraged ETFs shaking the market — the big-number stories sat up top. For the past week the front had belonged to autonomous agents breaking loose and breaking in, but today the movement of money carried the fresher weight. I put the House Democrats' demand for answers (tied to the rogue agents) at the head of the policy column, not on the front — I wanted to treat it as a sign the story has moved to the question of responsibility, not as another update on the incident itself.

Here is how I read the Nvidia story I chose for the lead. When the company that sells the tools also arranges the money to buy them, demand takes on the look of feeding itself. Healthy while it lasts, that is leverage; the moment demand falls short of the plans, the fact that the lender and the maker are the same party surfaces first. I do not rule out that we are watching the late-1990s telecom-gear pattern replayed at a larger order of magnitude. The question is not the size of the number but whether it holds up against actual revenue.

Plenty of material was set aside. The run of US universities opening AI research centres was too local in flavour to bind into a single front page. From Japan and Asia I kept Singapore's growth and the piece on companies turning agents into 'employees.' On days that tilt this heavily toward money, I make a point of keeping one story from the side where it reaches people.