Even companies worth $1 trillion don’t want to pay Nvidia’s prices

I hope that everyone has made tons of money taking my investing advice and doing the opposite. Loyal readers may remember that I’ve been skeptical of Nvidia becoming worth more than all of the European Union combined and, therefore, never purchased NVDA (current market cap: $5+ trillion). My theory was that 50 percent profit margins for Nvidia would motivate other companies to develop chips and then the AI nerds would recompile their code (or ask AI to recompile their code). Google did this a long time ago with its Tensor Processing Unit. Intel tried and apparently failed with Gaudi (but they were very successful in DEI!). Elon Musk has his AI5 chip and, eventually, his own fab.

OpenAI is the latest entrant into this mix, apparently. “OpenAI Claims Its New Chips Can Outperform Nvidia Processors in Tests” (Bloomberg, yesterday):

The company plans to start using the new chips to support its artificial intelligence models later this year, which should significantly reduce costs for OpenAI as it rolls out more widely.

OpenAI, losing enough money every year to be worth almost $1 trillion, created Jalapeno in a partnership with Broadcom Inc., which makes custom chips for a variety of clients. The two companies, which announced their pact last year, touted the speed at which the processor was developed in June, saying it came together in record time.

Even if this thing succeeds, there will still be a huge market for Nvidia, apparently:

Jalapeno wasn’t tested against the new generation of Nvidia chips, Vera Rubin, which just began shipping. It’s also not designed for training AI models, an area where Nvidia technology excels. Jalapeno is intended for the inference phase of AI — the stage when models have already been trained and can begin responding to prompts and handling tasks.

(Why is the chip named for Vera Rubin, someone who was neither an electrical engineer nor a software developer? Google AI:

Honoring Women in STEM: The company has a specific practice of naming key platforms after pioneering women scientists. Previous examples include Grace Hopper (the Grace CPU) and Ada Lovelace (the RTX 40-series architecture)…

I guess we can’t expect the William Shockley CPU/architecture any time soon, despite the man’s Nobel for the invention of the transistor, but why not one for Jack Kilby? Kilby died in 2005, prior to the full flowering of Rainbow Science, and therefore Kilby’s gender ID is unknown (we could say the same regarding Vera Rubin, who died in 2016, some years before the trans-enhanced Rainbow Flag flew at the White House))

How much do journalists and editors need to know about electrical engineering in order to cover cutting-edge AI hardware?

Because the chip is achieving strong results at low voltage — 700 watts — Jalapeno will allow OpenAI to save money when running its data centers, where power is a key cost, he said.

Speaking of high voltage, let’s have a look at the cooling system for Google’s latest TPUs (blog entry, which Google’s massive AI brain couldn’t figure out should have a date on it; I think it is from April 2026):

It looks like they got retired Pratt PT6 engineers to work on this!

(Actually, this suggests a career path: instead of fighting it out on the crowded software front, a young person could study to become a packaging/cooling engineer.)

Vaguely related… let’s see if we can figure out where Nvidia’s passion for “Honoring Women in STEM” comes from. Perhaps nerds identifying as “women” were the principal designers of Nvidia’s hardware and software?

Prompt to Gemini (to run on Google’s TPUs so that the results aren’t biased by running on Nvidia chips):

Who were the principal engineers of Nvidia’s historically important GPUs and software architectures, including CUDA? In other words, who are the people who are primarily responsible for making Nvidia as valuable as it is today?

Answer (not in quote style):

Key Technical Leaders Behind Nvidia’s Rise

1. Curtis Priem & Chris Malachowsky (Co-Founders & Founding Hardware Architects)

• The Role: Co-founded Nvidia in 1993 alongside Jensen Huang.
• Their Contribution: Priem was the chief architect of Nvidia’s early graphics chips (including the NV1 and the breakthrough RIVA 128), while Malachowsky led system engineering and hardware design operations.
• Impact: They established Nvidia’s core hardware engineering practices and designed the early pipelines that allowed Nvidia to survive the “3D accelerator wars” of the late 1990s and create the original GeForce 256—the chip Nvidia marketed as the world’s first true “GPU.”

2. Dr. David Kirk (The “Father of CUDA” & Chief Scientist 1997–2009)

• The Role: Chief Scientist and VP of Architecture.
• Their Contribution: David Kirk was the primary visionary behind transforming Nvidia’s fixed-function 3D graphics hardware into programmable, general-purpose parallel processors.
• Impact: In the mid-2000s, Kirk persuaded Jensen Huang to mandate that every single Nvidia GPU include dedicated general-purpose execution units—a massive financial gamble at the time. This decision laid the groundwork for CUDA (Compute Unified Device Architecture), launching general-purpose GPU computing (GPGPU) and making GPUs usable for scientific computing and later AI.

3. Ian Buck (VP of Accelerated Computing & Original CUDA Architect)

• The Role: Creator of the BrookGPU programming language (at Stanford) who joined Nvidia to lead the CUDA project.
• Their Contribution: While pursuing his PhD at Stanford, Buck created Brook, an extension of C for streaming coprocessors. Jensen Huang hired Buck to lead the software and systems team that built the original CUDA software stack and runtime environment, released in 2006.
• Impact: Buck turned the raw, complex parallel hardware engineered by Kirk’s team into an accessible software platform for C/C++ developers. Without Buck’s software stack, AI researchers in 2012 (like Alex Krizhevsky and Geoffrey Hinton) would not have been able to train AlexNet on Nvidia GPUs—the spark that triggered the modern deep learning boom.

4. John Montrym (Chief Architect, GeForce/Tesla Architecture)

• The Role: Senior Architect and Fellow.
• Their Contribution: Montrym was a key architect behind the NV40 (GeForce 6 series) and the landmark G80 (Tesla architecture) released in 2006.
• Impact: The G80 was the first unified shader architecture in a PC GPU and the hardware home where CUDA was born. It abandoned separate vertex and pixel pipelines in favor of unified streaming multiprocessors (SMs), setting the blueprint for every modern GPU architecture designed since.

5. Jonah Alben (Senior VP of GPU Engineering)

• The Role: Longtime leader of Nvidia’s hardware silicon development.
• Their Contribution: Alben joined Nvidia in 1997 and has spent over two decades directing the physical architecture, design execution, and engineering teams behind generations of GPU microarchitectures: Fermi, Kepler, Maxwell, Pascal, Volta, Ampere, Hopper, and Blackwell.
• Impact: Alben’s team managed the scaling and execution challenges as GPUs grew from millions to over 200 billion transistors on single silicon packages. His team integrated specialized hardware engines directly into the GPU pipeline—most notably Tensor Cores (introduced in Volta/Ampere) and Transformer Engines (in Hopper/Blackwell)—which gave Nvidia a 10x-100x efficiency advantage in AI workloads over traditional CPUs and competing chips.

6. Dr. Bill Dally (Chief Scientist & Head of Nvidia Research)

• The Role: Former Chairman of Stanford’s CS Department, appointed Nvidia Chief Scientist in 2009 (succeeding David Kirk).
• Their Contribution: A pioneer in computer architecture and high-speed interconnect networks, Dally leads Nvidia’s long-term research division.
• Impact: Dally pivoted Nvidia’s research toward optical interconnects, low-precision AI math (FP8, FP4 formats), ray tracing hardware cores (RT Cores), and multi-chip network scaling (NVLink). His architectural research predicted the massive bandwidth bottlenecks of large AI models years before ChatGPT emerged.

You can see the full answer and also one from Grok.

Fun fact from Oshkosh/AirVenture 2026: I ran into a mid-level Nvidia employee. He had shown up in his new airplane, a perfectly restored P-51 Mustang ($5 million?).