$95B Burned, $500B Promised, Zero Returns Yet

So, What Actually Happened?

So I was reading quarterly filings on Saturday morning, which is not how I planned to spend it, and one line stopped me. Four US technology giants burned through $95 billion in cash in a single quarter, and the reason is AI capital spending. We scanned 190,000 articles this week so you don't have to. Then Goldman put a number on the next leg, telling clients to expect more than $500 billion of AI investment across 2026. Fine, big companies spend big money. But in the same window a broker opened a pre-IPO platform listing Databricks, so private AI exposure is being packaged and sold while the spending meant to justify it is still going out the door. I kept looking for the part where the returns land. It is not in this quarter.

The Bottom Line: The spending is being financed forward and the exit is being sold backward. Nothing in the middle has cleared yet.

The Tracks That Matter

1. Four Tech Giants Burned $95 Billion in a Single Quarter

The number that matters this weekend is not a valuation, it is a cash outflow: four of the largest US technology companies bled $95 billion in Q2, driven by AI infrastructure. These are the companies with the strongest balance sheets on earth, which is exactly why it registers. Goldman's read is that the pace continues, with the sector heading toward over $500 billion of investment this year. Capital spending on that scale is not a bet on a product, it is a bet on a decade. And the awkward part for everyone downstream is that these firms are your suppliers. When a supplier is spending ahead of revenue, the price you pay eventually carries some of that gap.

Here's what works: Ask your cloud and model vendors what their per-unit pricing assumes about their own capital recovery. The answer shapes your 2027 renewal, not your 2026 one.

2. Databricks Shares Go on Sale Before the IPO Does

While the giants burn cash publicly, private AI paper is quietly getting a storefront. Clear Street launched a pre-IPO platform with Databricks among the listed names, which turns a company most buyers only meet through a procurement portal into something you can hold a position in. This is market plumbing, so it gets covered as a broker story rather than an AI story, and that is the miss. Secondary markets forming around a private company means the people closest to it want liquidity before a public price exists. That is not a scandal. It is a signal about where confidence sits on the timeline, and it arrived the same week the capex forecasts went up.

Here's what works: If a vendor of yours starts trading on secondary platforms, read it as a countdown. Renegotiate multi-year terms before a listing resets their pricing power.

3. Google Retires the Keyword and Advertisers Lose the Receipt

Quietly, and with far less noise than a model launch, Google is replacing Dynamic Search Ads with AI Max, moving paid search from matching words to inferring intent. ”This evolution is about moving from static tools to a system that truly understands intent,” said Brandon Ervin, and I believe him, which is the problem. A keyword is auditable. You can point at it, defend it in a budget review, and kill it when it underperforms. An intent model is a black box sitting between your money and your customer, and the only lever you still control is the quality of what you feed it. Marketing just joined the club data teams have been in for years.

Here's what works: Before the migration completes, export your keyword-level performance history. It becomes the only baseline you will ever have for judging what the model does next.

Quick hits:

• Korea's chip trade turned into a market event. The Kospi surged 18% on AI chip optimism, a reminder that AI capex lands as national economic policy somewhere before it lands as your invoice.
• Private credit just got its own AI startup. Repeat founder Ryan Williams raised a $10 million seed for tooling aimed at private credit managers, which is where the least-digitized money in finance currently sits.
• Researchers asked whether agents can do research. A new paper tests whether AI agents can run open-ended research, and the early results are the honest kind: useful at the edges, not yet at the core.

Signal vs. Noise

🟢 Signal: data quality. Data quality gained real ground across the industry this week while louder categories lost it, and the AI Max change shows why: when a system infers intent instead of matching a keyword, the inputs become the only thing you still control. Most coverage is scoring model releases against each other and missing that the leverage moved back to the feed.

🔴 Noise: the phrase ”agentic AI.” It pulled heavy volume again while thinning out on substance, and the tell is what got published alongside it: buyer's guides, category directories, and standards checklists rather than deployments. When a term generates more vendor lists than case studies, it has become a shelf label.

From the 190K

We scanned 190,000 articles this week. Here's what no one's talking about:

Four of the biggest technology companies burned $95 billion of cash in three months, Goldman told clients to expect over $500 billion of AI investment this year, and a broker opened a pre-IPO window on Databricks in the same week.

Read separately, each lands on a different desk and gets filed as routine. The earnings desk covers cash flow. The research desk covers a capex forecast. The markets desk covers a broker launching a product. Put them on one morning and the shape changes: the industry is financing an enormous build-out on future returns while simultaneously creating ways to sell exposure to it before those returns exist. Both of those things happen in healthy markets. They rarely happen this close together.

What changes on Monday is a question you can ask out loud without sounding alarmist. Every AI supplier in your stack is currently spending ahead of what they earn. That gap gets closed by revenue growth, by price increases, or by consolidation. Only one of those three is good for you, and none of your contracts specify which one you are betting on.

By The Numbers

• Four US tech giants bled $95 billion in cash in Q2 — the cash drain is AI infrastructure, and it is happening at the companies with the deepest reserves in the industry.
• Goldman Sachs projects over $500 billion of AI investment in 2026 — not a forecast of demand, a forecast of spending, which is a very different thing to underwrite.
• The Kospi surged 18% on AI chip optimism — an entire national index moving on the semiconductor leg of the AI trade.
• Blue Origin was valued at $130 billion after SpaceX's record IPO — a rival's listing lifted the comparable instead of denting it, which tells you how thin the private pricing anchors are.
• Ryan Williams raised a $10 million seed for private credit AI — early money moving toward the least-digitized corner of finance.
• See what's rising in our 190K-article corpus this week →

Deep Dive: The Tour That Loses Money

Every band I knew lost money on the road. Diesel, hotels, a sound engineer who drinks your margin, and a merch table that never quite covers it. Nobody thought that was a failure. Touring was how you built the thing that eventually paid, and the ones who survived knew precisely which part of the loss was buying something and which part was just bleeding.

The loss is supposed to buy an audience
Ninety-five billion in a quarter is a tour budget. The question is not whether it is large, it is what it purchases. Compute capacity that compounds, or capacity that depreciates on a three-year clock while a cheaper chip ships. Both look identical on a cash flow statement in July.

The merch table opened early
A pre-IPO listing while the spending is still climbing is the band selling t-shirts before the album. Perfectly legitimate, and it tells you something real about who wants liquidity and when. Watch which insiders reach for the merch table, not what the tour poster says.

Somebody is paying for the diesel
Suppliers spending ahead of revenue close the gap eventually, and enterprise buyers are the nearest source of cash. It arrives as a price change, a repackaged tier, or a feature moving behind a higher plan. Rarely as an announcement.

What Actually Works

1. Read your vendors' cash flow, not their launches: the release notes tell you what they built, the cash statement tells you what they need from you.
2. Price in a step change at renewal: assume one repricing event per major AI supplier in the next 24 months and budget the range now.
3. Keep one workload portable: not all of them, that is fantasy. One, chosen deliberately, proves your switching cost is real and gives you a number to negotiate with.
4. Separate compounding spend from depreciating spend: your own AI budget has both. The tour only works if you can tell them apart.

Every band on the road is losing money. Only some of them are buying an audience with it.

What's Coming

Capex Discipline Shows Up in Vendor Pricing

The $500 billion investment projection does not stay on the supplier side of the ledger. Expect tier restructuring and usage-based repricing to arrive dressed as product improvements over the next two quarters. Read release notes with a procurement eye.

Private AI Names Get a Public Price Signal

With pre-IPO platforms listing private AI companies, the valuations of your unlisted vendors stop being private. That visibility cuts both ways, and it will start showing up in negotiations before it shows up in headlines.

Governance Buying Shifts From Frameworks to Controls

The arrival of a full AI governance vendor landscape marks the moment a discipline stops being a discussion and becomes a purchase order. Expect budget to move from policy documents toward auditable controls, and expect the phrase itself to keep losing meaning as that happens.

For Your Team

Monday's meeting prompt: ”Every major AI supplier we use is currently spending more cash than it earns. That gap closes through revenue growth, price increases, or consolidation. Which one are our contracts written for, and what happens to us if it's one of the other two?”

Share-worthy stat: Four of the largest US technology companies burned through $95 billion in cash in a single quarter, and the driver was AI infrastructure. The interesting part is not the number. It is that the companies with the strongest balance sheets in the world chose to run them down on purpose.

Go deeper: Track where AI capital and vendor pricing are moving, in real time →

The Track of the Day

”This evolution is about moving from static tools to a system that truly understands intent.”
— Brandon Ervin, on Google's AI Max replacing Dynamic Search Ads

A system that truly understands intent is also a system you cannot cross-examine. Keep a copy of what the old one told you.

We scanned 190,000 articles this week so you don't have to. Data Pains → Business Gains.

Published: August 2, 2026 | Curated by Yves Mulkers @ Ins7ghts