AI raised another $85 billion today — and ran straight into the power bill

Information Technology 5 min 25 sources

Google's parent raised a record $85 billion for AI in a single oversubscribed sale, even as Europe moved to cap data-centre energy use, Americans told pollsters they don't want the buildings, and Amazon's own engineers showed up at city hall to demand limits. The cloud, it turns out, is made of steel, water, and electricity.

Key takeaways

Google's parent Alphabet raised a record ~$85 billion for AI in a single oversubscribed share sale — proof that investor appetite for anything labelled AI is close to unlimited.
That money turns into data centres, which are now colliding with hard physical limits: Europe moved to cap their soaring power use, seven in ten Americans oppose them over water, and Amazon's own engineers protested at Seattle city hall.
AI is splitting in two directions at once — ballooning in the cloud while also shrinking to fit a laptop, as Google's free, offline Gemma 4 model shows.

Two facts from today sit oddly together. Investors handed Google’s parent company a record pile of cash to build more artificial intelligence. And on the same day, the people who live near where that AI gets built — including the engineers who build it — pushed back harder than ever. The AI boom just hit the part of the world that doesn’t scale with a software update: the power grid.

The money

Alphabet, Google’s parent, raised about $85 billion by selling new shares to fund its AI plans [20]. It had planned to sell a first chunk of $40 billion; demand was so heavy it took $45 billion in that round alone [21]. It’s the largest raise of its kind, and the signal is blunt: investors will pour almost any amount into anything labelled AI [21]. CEO Sundar Pichai called the appetite voracious, and the price proved him right.

That money mostly turns into one thing — data centres, the warehouses full of computers that train and run AI. And those buildings are colliding with physical limits that no amount of funding can wish away.

The power bill

Start with electricity. The European Union proposed minimum energy-efficiency standards for data centres, warning their power use is climbing fast [5]. In the EU alone, data-centre demand is set to more than double — from 12 gigawatts last year to 28 by 2030 — pushing their share of the bloc’s electricity past today’s 2.5% [5]. Globally, the International Energy Agency expects data centres to drive a fifth of all growth in electricity demand in advanced economies by 2030 [5]. That’s the catch: if the grid can’t keep up, either power prices rise for everyone or old fossil-fuel plants stay switched on longer.

Then water. Data centres use it to cool the heat their servers throw off — and the numbers are large. Google’s facility in Council Bluffs, Iowa, drank more than a billion gallons in 2024 alone [16]. A US national lab estimates the biggest data centres could consume up to 33 billion gallons of water a year by 2030 [16]. The public has noticed: a Gallup poll found seven in ten Americans now oppose data-centre development in their area, with water scarcity their top worry [16].

The sharpest sign came from inside. Amazon software engineers turned up at Seattle city-council meetings to demand local limits on data centres — believed to be the first time the people building these systems have protested them this publicly [22]. “Let’s not let Big Tech burn Seattle to win the AI race,” one Amazon engineer told a hearing [22]. The thing the industry calls “the cloud” turns out to have a street address, a water meter, and neighbours.

Europe wants its own machine

If AI is infrastructure, then who controls it becomes a national-security question — and Europe spent today answering it. The European Commission unveiled a “made-in-Europe” tech package, including a Cloud and AI Development Act and a “Chips Act 2.0,” aiming to double the EU’s share of global semiconductor-making to 20% by 2030 [3]. Semiconductors are the chips that run everything; today most are made in the US and Asia.

The fear driving it is concrete. “We cannot afford to depend on others for the technologies that keep our hospitals running, our energy grids stable and our services secure,” said Commission President Ursula von der Leyen [3]. The EU’s tech chief warned of foreign “kill switches” that could disable critical services [3]. A specific worry is the US Cloud Act, a law that can force American cloud providers to hand data to US authorities even when it’s stored on European soil [3]. The plan would require providers in sensitive sectors — banking, energy, healthcare — to keep European data under European control [3]. For anyone running services on a US cloud in Europe, the rules of where data must live are about to change.

The model that fits on your laptop

Against all that bigness, a quiet counter-current: AI is also getting small. Google released Gemma 4, a free, open model that runs entirely on an ordinary laptop with 16GB of memory — no data centre, no internet connection required [10][8]. It handles text, audio, and video together, holds a large amount of context at once, and can run offline, which matters for privacy or for working on a plane [8].

The interesting part is the direction. While the headline money flows into ever-larger cloud systems, the tools are also shrinking to fit a device you already own. For a developer or a business, that’s a real fork: some AI work that today means renting time on someone else’s data centre could soon run on the hardware on your desk — free, private, and off the grid that everyone else is fighting over.

Where else the money went

The funding firehose stayed on elsewhere. Meta launched an AI “agent” for businesses — software that doesn’t just chat but takes actions, like booking appointments or closing a sale — and said more than a million businesses already use its earlier versions on WhatsApp and Messenger [2]. It’s Meta’s bid to make money from companies, not just ads, and to keep pace with OpenAI, Anthropic, and Google [2]. AI music startup Suno raised money at a $5.4 billion valuation [6]. Coralogix raised $200 million on a tidy premise: as companies unleash AI agents, someone has to watch what the agents actually do [1].

But the money isn’t blind. Broadcom, a major AI-chip maker, reported quarterly sales of $22.19 billion — a hair below what Wall Street wanted — and its shares fell more than 13% after hours [24]. The company still expects to ship more than 10 gigawatts’ worth of AI chips in 2027; it just didn’t raise its forecast, and in this market, standing still is enough to get punished [24]. Even inside a boom, the bar keeps moving up.

The quiet listing

The under-covered story: quantum computing took a real step toward Wall Street. Quantinuum, the quantum-computing firm majority-owned by Honeywell, raised $1.68 billion in a US stock-market debut, pricing shares at $60 [7]. Quantum machines — which use the strange rules of physics at atomic scale to attempt calculations ordinary computers can’t — are still years from broad usefulness. But $1.68 billion of investor money says the race to own the next computing platform is already on, well before the current one has finished reshaping the grid beneath it.

02 · Lesson · why it matters

Where the weight went

We call it "the cloud" — a word that means light, far away, not here. Today, Google's own engineers stood up at a city-council meeting to talk about the water and power it actually uses. The lesson is in the gap between the word and the thing.

A word doing a quiet job

Think about why we say “the cloud.” Your files, your photos, the answer an AI gives you — we picture them floating somewhere weightless and far off. Nothing about the word suggests a building. Nothing about it suggests a water meter.

But the cloud is a warehouse full of computers, in a real town, plugged into a real power grid, cooled by real water. Today the numbers got specific: one Google site drank over a billion gallons of water in a single year, and seven in ten Americans say they don’t want these buildings near them. The word “cloud” wasn’t a lie. It was just doing a job — making something heavy feel light.

Weightless to you, heavy for someone

This isn’t only about AI. It’s a pattern worth seeing everywhere.

Almost everything that feels frictionless to you has moved its cost somewhere you can’t see. The shirt that’s cheap because it was made by someone far away on wages you’ll never think about. The package that arrives in a day because of a warehouse and a driver running on a clock you don’t see. The “free” app paid for with your attention and your data. The AI answer that appears in a second, drawing power from a grid a stranger lives next to.

The convenience is real. So is the cost. The trick of modern life is that the two are almost never in the same place. What feels weightless to one person is heavy for another — and usually that other person is out of sight.

A cost you can’t see is a cost you can’t weigh

Here’s why this matters beyond noticing it. When a cost is hidden, you make decisions as if it were zero.

If running an AI query felt like turning on a tap that someone else paid the water bill for — which is roughly what’s happening — you’d use it differently than if it felt free. Not necessarily less. Just with your eyes open. The danger isn’t using the convenient thing. The danger is choosing it while genuinely believing it costs nothing, because the meter is in another state.

That’s the deeper error: treating the convenient front and the hidden cost as two separate things, when they’re one thing seen from two ends. The whole system — the grid, the town, the water table — pays for the gap between how light it feels and how heavy it is.

The engineers are the tell

The most interesting detail today is who pushed back. Not activists. Amazon’s own software engineers, the people who build these systems, showing up at city hall to ask for limits.

When the people closest to a thing start naming its weight out loud, it usually means the gap between the word and the reality got too wide to ignore. They can see both ends at once — the elegant system and the warehouse next to the reservoir. That double vision is exactly the skill worth borrowing.

What to carry out of today

You don’t need to feel guilty about using the cloud, or AI, or next-day delivery. Guilt isn’t the point, and it doesn’t help anyone.

The point is to keep both ends in view. When something feels free, instant, or weightless, ask one plain question: where did the weight go, and who’s carrying it? Sometimes the answer is “no one much, it’s fine.” Sometimes it’s a town you’ve never been to, running its air conditioning on a hotter grid so your answer could arrive a second faster. You can’t weigh a trade you can’t see. Seeing it is the whole skill.

03 · Lab · your turn

Weightless?

Judge whether each everyday convenience is genuinely cheap or hides a cost someone else carries — practising the skill of looking past how free something feels.