Reallocated

A Blackwell AI Chip Sells for $40,000. An RTX 5090 Sells for $2,000, When You Can Find One.

A single AI datacenter chip from Nvidia sells for somewhere between $30,000 and $40,000. A gaming graphics card from the same company sells for $299 to $1,999 — when you can find one at all. Nvidia's own CEO has said the quiet part out loud, and the math explains everything about why your next GPU upgrade is going to be expensive, late, or both.

Same Company, Two Customers, One Obvious Favorite

Jensen Huang told CNBC exactly what a Blackwell B200 AI chip costs a buyer: $30,000 to $40,000 per unit. A GeForce gaming card, built by the same company on adjacent production lines, tops out under $2,000. Nvidia's own financial reporting makes the imbalance concrete — in fiscal year 2025, the company's data center business pulled in $115.2 billion in revenue, while gaming brought in $11.4 billion. Data center now accounts for something like 87% of Nvidia's total revenue. Gaming, the business that built Nvidia's brand for two decades, is a rounding error next to it.

An Nvidia GeForce RTX 5060 Ti graphics card. Photo: FreeMediaKid!, Wikimedia Commons, CC BY-SA 4.0.

The Factories Are Booked Solid, and Not for You

None of this happens in a vacuum — Nvidia's AI chips are manufactured by TSMC, whose advanced packaging capacity has become the tightest resource in the entire tech industry. TSMC's CEO told shareholders in June 2026 that this capacity was "extremely tight and sold out through 2026," with the company's advanced-packaging facilities reportedly booked with year-plus lead times into 2027. TSMC's high-performance-computing segment — the AI-anchored one — now accounts for roughly two-thirds of the company's entire wafer revenue. Every wafer dedicated to a $35,000 AI chip is a wafer that isn't going toward the far less profitable business of making sure a gamer can buy a graphics card at MSRP.

January 2025: The Cards Existed for About Ten Minutes

The consumer end of this showed up exactly as you'd expect. When the RTX 5090 and RTX 5080 launched on January 30, 2025, at official prices of $1,999 and $999, both sold out almost instantly — a "paper launch," in the press's own preferred term for a product you can look at but not buy. Scalpers had RTX 5090s listed on eBay for over $6,000, with some listings reaching $7,000. In fairness to Nvidia, that specific launch also had a genuine, admitted manufacturing defect — Huang later acknowledged "a design flaw in Blackwell" that separately hurt yields — so not every bit of that particular shortage is a straight line back to AI. But the underlying incentive problem doesn't need any one bad launch to prove it: when a company can sell a chip for $35,000 or $300, one of those numbers is going to get more attention from the factory floor.

Nvidia's Own Warning, in Its Own Words

Nvidia doesn't dispute any of this — it's said so on its own earnings calls. CFO Colette Kress told investors in February 2026: "We expect supply constraints to be a headwind to gaming in the first quarter of fiscal 2027 and beyond... As much as we would love to have additional more supply, we do believe for a couple quarters, it is going to be very tight." That's a company confirming, on the record, that the product line gamers actually want is the one it's least equipped — or least motivated — to keep fully stocked. Nvidia will keep making graphics cards. It's just not going to lose any sleep over how few of them show up on shelves, as long as the factory has somewhere better-paying to put its capacity.

Filed from public reporting:

A Hand-Tightened Blind Flange and a Botched Shift Handover Killed 167 Men on Occidental Petroleum's Piper Alpha Platform in 22 Minutes. No One at the Company Was Ever Criminally Charged.

At around 10 p.m. on July 6, 1988, a pressure safety valve that had been pulled for maintenance that morning let gas condensate leak past a blind flange that had only been hand-tightened. Within 22 minutes, the Piper Alpha platform, 120 miles northeast of Aberdeen in the North

Netflix Spent Six Years Telling Customers Love Is Sharing a Password. Once It Started Charging $7.99 a Month Per Shared Password Instead, Subscriber Growth Hit Its Best Year Since the Pandemic.

On March 10, 2017, Netflix's official Twitter account posted four words that would follow the company around for the next six years: "Love is sharing a password." It was a promo tweet for a show called Love, but it was also, unmistakably, a policy statement — an

New York Paused Congestion Pricing for 'Affordability' Weeks Before a 2024 Election, Then Relaunched It at a Lower Price Once Voting Was Over. A Year Later, Traffic Was Down 11 Percent and the Federal Government Was Still in Court Trying to Kill It.

On June 5, 2024, Governor Kathy Hochul stood before reporters and "indefinitely paused" New York's Central Business District Tolling Program, the long-planned congestion charge for driving into Manhattan below 60th Street. The launch was three weeks away. Hochul cited affordability concerns for working- and middle-

Twitter Cut Off the Third-Party Apps That Built Its Developer Ecosystem in January 2023 With No Warning. One of Those Apps Coined the Word Tweet and Designed Twitter's Bird Logo Before Twitter Had Either.

On January 13, 2023, Twitter's API went dark for third-party client apps with no announcement and no explanation. Tweetbot, Twitterrific, Fenix, and others simply stopped working, all at once, for everyone. Developers spent days guessing whether it was a bug, an outage, or deliberate. Four days later,