Skip to main content

The Rise of NVIDIA, Part 19: The AI Factory

The AI Factory

🎮 The Rise of NVIDIA — a 20-part series. See all parts »  |  « Part 18: One Trillion Dollars

By the spring of 2024, NVIDIA was no longer a company that made chips for a market. It was the market. Every hyperscaler, every startup with a language model and a dream, every sovereign wealth fund suddenly building "national AI" was standing in the same line, holding the same purchase order, waiting for the same silicon. So when Jensen Huang walked onto the stage at the SAP Center in San Jose on March 18, 2024 — a hockey arena, not a ballroom — the room felt less like a product launch and more like a stadium show. The keynote was for NVIDIA's GTC conference, and the headline act was a new architecture named Blackwell.

The chip that broke its own rules

Blackwell was named for David Blackwell, the mathematician and statistician who in 1965 became the first Black scholar inducted into the U.S. National Academy of Sciences. The name was a nod to lineage; the chip itself was a break from it. Every prior NVIDIA GPU had been a single slab of silicon, and by 2024 that approach had hit a wall physicists call the reticle limit — the largest area a chip factory's lithography can print in one shot. Hopper, the H100 that had powered the ChatGPT boom, was already pressed against it.

So NVIDIA cheated the limit. Blackwell fused two reticle-sized dies into a single GPU, stitched together by a chip-to-chip link running at 10 terabytes per second — fast enough that the two halves behaved, to software, as one seamless processor. The result carried 208 billion transistors, more than double Hopper's roughly 80 billion, all fabricated on a custom TSMC 4-nanometer process. A second-generation Transformer Engine added new low-precision number formats built specifically for the math that large language models actually do.

The number NVIDIA repeated most was the one aimed squarely at the wallet: for large-language-model inference, Blackwell promised up to 25 times lower cost and energy than Hopper. In an industry where the electricity bill of running a model had become a genuine constraint on the business, that was not a spec. It was a sales pitch that closed itself.

Selling the factory, not the chip

Huang had stopped selling chips years earlier, though the invoices still said so. What he sold now was a phrase, and he repeated it until it stuck: the AI factory. A modern data center, in his telling, was no longer a warehouse of computers. It was a plant with raw material coming in — data, electricity — and a single finished product coming out: tokens, the units of generated intelligence. You did not buy a factory one machine at a time. You bought the whole line.

The physical embodiment of that idea was the GB200 NVL72. Instead of shipping a card, NVIDIA shipped a rack: 72 Blackwell GPUs and 36 Grace CPUs wired together by a fifth-generation NVLink fabric, each GPU pushing 1.8 terabytes per second of bandwidth to its neighbors — roughly double Hopper's link. The whole 120-kilowatt, water-cooled cabinet was designed to be programmed as if it were one colossal GPU. It was the unit of purchase for the trillion-parameter era, and it came with a price tag to match: analysts pegged a single NVL72 rack in the low millions of dollars.

The customer list read like a census of the global economy. On launch day NVIDIA named Amazon, Google, Meta, Microsoft, Oracle, Tesla, xAI and OpenAI among the companies lining up for Blackwell. Their CEOs supplied the quotes personally — Sundar Pichai, Satya Nadella, Mark Zuckerberg, Sam Altman, Larry Ellison — a chorus of the most powerful people in technology, all effectively confirming that their AI ambitions ran through one supplier in Santa Clara.

The rock star in the leather jacket

Somewhere in this stretch, Jensen Huang crossed a line few executives ever reach: he became famous to people who could not name a single thing his company made. The signature black leather jacket became a costume the internet recognized instantly. Three months after the Blackwell reveal, he flew home to Taiwan for the Computex trade show and was mobbed like a pop idol — crowds trailing him through Taipei's night markets, fans thrusting products forward to be autographed. The local press took to calling him "AI's rock star." One outlet compared the frenzy to Taylor Swift.

It was an extraordinary turn for a founder who, thirty-one years earlier, had sketched a company's plan in a Denny's booth and spent the late 1990s thirty days from bankruptcy. The man who once begged Sega for a lifeline now had trillion-dollar platforms reorganizing their capital-expenditure budgets around his release calendar. Blackwell was, in a sense, the moment the whole improbable arc snapped into focus. The graphics chip had become the engine of what Huang called a new industrial revolution — and he was standing at the center of it, selling the future by the rack.

But a company this large, growing this fast, does not simply keep rising in a straight line. Numbers that big attract gravity of their own. Next: the day NVIDIA became the most valuable company on Earth — and what that title really cost.


🔗 Explore more from Syncster

Comments

Popular posts from this blog

Cursor AI Review: Is the AI Code Editor Worth It?

I've been using Cursor as my main code editor for a while now, and enough people have asked whether it's worth switching to that a proper review felt overdue. Short version: for me, yes — but with caveats. What is Cursor? Cursor is an AI-first code editor built as a fork of VS Code. That means every extension, theme, and keybinding you already use in VS Code works here, but with AI woven directly into the editing experience instead of bolted on as a plugin. It's made by Anysphere and can run models from OpenAI and Anthropic under the hood. What I like Tab completion is uncanny. Cursor predicts your next edit — not just the rest of the line, but the next change across the file. Once you get used to hitting Tab, going back to a plain editor feels slow. The Composer / Agent mode. You describe a change in plain language and it edits multiple files at once, showing you a diff to accept or reject. For refactors and boilerplate, this saves real time. It unde...

How I used Google Sheets and Apps Script

Google Sheet is one of the most powerful spreadsheet application that exists online, rivaling with Microsoft's Excel. One of the main strengths is its strong support for collaboration with other users, much easier and popular than collaboration tools with Microsoft Office. Aside from plain spreadsheet, it also supports extensions such as macro. If you are familiar with macros on other office tools, they work almost the same. However, the most extension I use and tinker with is the Apps Scipt . Apps Script Extension One of the challenges I faced recently is how do I track or monitor reports in our department if they are submitted on time or worst, forgotten due to lack of better monitoring tools. So I thought if there can be simple applications that can be deployed or use by a more general user to allow reminding periodically what reports are approaching due dates or those that are past dues. Then I looked for a way, instead of creating a full blown app from scratc...

MacBook Pro M5 vs M5 Pro: Which One Should You Actually Buy?

Apple's latest 14-inch MacBook Pro comes in two very different flavors: the base M5 and the step-up M5 Pro . On paper they look similar — same gorgeous Liquid Retina XDR display, same design — but under the hood the gap is bigger than the names suggest. Here's a clear, no-hype breakdown, with concrete use cases so you can match the chip to your work. Quick spec comparison Spec M5 M5 Pro CPU 10-core (4 performance + 6 efficiency) Up to 18-core (6 performance + 12 efficiency) GPU 10-core Up to 20-core Neural Engine 16-core 16-core Memory bandwidth 153 GB/s 307 GB/s (roughly double) Unified memory 16 / 24 / 32 GB 24 / 48 / 64 GB Max storage Up to 4 TB SSD Up to 8 TB SSD Battery (video playback) Up to 24 hours Up to 22 hours Media engines Single encode/ProRes engine More encode/ProRes engines (higher configs) What actually changes between them More cores — the M5 Pro nearly doubles CPU cores and adds GPU cores, so sustained, multi-threaded work finishe...