Monday, and the week opens on the money, not the models. The theme today isn't a new capability — it's the gap between how much capital the AI buildout is consuming and how much of that spend has actually shown up in anyone's P&L yet. Three stories put that gap in sharp relief, and if you sign the invoices at a real company, it's the most important thing on the board this morning.
Anthropic files to go public — and shows the price tag
The headline event is a filing. Anthropic unveiled a public IPO filing, bringing the AI lab a step closer to a historic offering, with the filing confirmed in Monday's market coverage. The number that stops you is in the disclosure: Anthropic reported a loss of $80.7 billion in the first half of this year, and the S-1 lays out its risk factors in the open for the first time.
The operator's take: read the headline number, then go read the footnotes — a figure that size on an AI lab's books is a signal about how the instruments are structured, not just how much cash went out the door. But the strategic point stands regardless: the company whose model may be answering your customers is now a public entity carrying losses at a scale that makes its pricing a function of capital markets, not just compute. When your core vendor goes public, its margin pressure becomes your roadmap risk. Keep your inference layer portable and assume today's generous pricing is a growth-stage subsidy that an IPO prospectus eventually has to justify.
The "zero payoff" note nobody at the AI labs wants framed
Now the other side of the ledger. Apollo Global Management's chief economist Torsten Slok argued that despite the investment and hype, AI has delivered "zero payoff" to corporate America outside of technology. Looking at what he calls the S&P 493 — the index minus the Magnificent Seven — he found revenue per worker rising for the Mag 7 but flat for everyone else, with the same true of profit margins and no sign yet of AI boosting productivity growth. Put plainly, the technology has shown no financial payoff for corporate America outside of tech so far.
The operator's take: this is not an argument to stop — it's an argument to stop confusing activity with return. "Flat revenue per worker" across 493 companies means most AI deployments are landing as cost, pilots, and vibes, not as margin. If you're spending on AI, you owe your board a number: which process got cheaper, which cycle got faster, which headcount you didn't have to add. Pick two or three workflows, instrument them before and after, and kill anything that can't show the delta. The labs can run on narrative for a few more quarters; your operating budget can't.
Meta lines up $95B, SpaceX $40B — the capital intensity is the story
So who's paying for all of it? Keep spending. Meta plans to raise a total of $95 billion to fund its investments in AI infrastructure and superintelligence, per Monday's coverage. In the same report, SpaceX said it had secured $40 billion to pay for GPUs for AI, with first shipments expected in the first half of 2027. Two of the most capitalized companies on earth are borrowing at nine- and ten-figure scale just to buy the inputs.
The operator's take: when the giants are raising debt to fund the compute, you are not going to win by out-spending them, and you shouldn't try. Their capital intensity is your opportunity to be capital-light — buy the capability as a service, rent the infrastructure, and spend your scarce dollars on the proprietary data and workflow that they can't replicate. The buildout is someone else's balance-sheet problem. Your job is to turn their subsidized compute into a margin the S&P 493 note says most companies haven't found yet.
Also on my radar
- The delivery clock gets expensive. Data center tenants are now negotiating late-delivery penalties of $5M or more per day, up from roughly $1M earlier this spring — a clean tell that capacity, not demand, is the binding constraint heading into 2027, and that leverage has swung hard to the buyer.
- Nvidia floats a recurring toll. Jensen Huang discussed a possible annual software license of about $2,000 per GPU for running AI agents — it's a pricing signal, not a product yet, but it's exactly the kind of per-seat-for-silicon fee you should model into any agent TCO before you commit.
- Another lab, another breach. Quest Diagnostics confirmed a cyberattack and data breach affecting sensitive patient information in its Quanum lab system, disclosed in an 8-K filing — regulated health data, breached through a billing-connected system, a reminder that your integrations are your attack surface.
The throughline today is a timing mismatch: the capital going into AI is enormous and accelerating — an IPO, a $95B raise, a $40B chip loan — while the payoff outside the companies selling the picks and shovels is, by at least one hard-nosed read, still zero. That doesn't mean the technology is a dud. It means the market is financing a future that operators still have to actually build, one measured workflow at a time. Spend like the return is your job to prove, because it is. That's the Signal for today.
Paul Sapio is the CIO of Mikhail Education and a full-stack AI engineer. Open to contract work in security, networking, AI, and SaaS development — reach out.