🕹️ Last week the head of the world's most valuable company stood in a Tokyo arcade, surrounded by cabinets from the year 2000, and told a story he has told many times: Sega gave Nvidia $5 million when Nvidia was dying, and Nvidia never forgot. Every outlet on earth ran it. Almost none of them mentioned the part that makes it interesting, which is that the man being thanked had helped choose the technology that failed.

On July 15, 2026, Jensen Huang appeared at GiGO Akihabara Building 3, on a floor lined with arcade machines, for an invitation-only event announced only days earlier. A few dozen fans got in through a social media contest. Sega chairman and CEO Haruki Satomi and COO Shuji Utsumi were there. So was Yu Suzuki, the designer behind Virtua Fighter and Shenmue. And so was Shoichiro Irimajiri, who Japanese reporters noted had not stood on a Sega stage in a long time.

Huang took the microphone and talked for about twenty minutes, mostly about 1994.

He came to Japan for the arcade, and then he wouldn't leave the arcade

Nvidia was founded in 1993. By its own founder's account, the company was supposed to be building a chip, and instead he was in game centers every day.

What had him hooked was Sega's arcade hardware. Virtua Racing, Virtua Fighter, and then Daytona USA, which reached the United States in 1994. Huang told the Akihabara crowd that the motion was so smooth he suspected it couldn't be real 3D, and that when he looked closer he found fully polygonal geometry with texture mapping on top. Nobody was making 3D games on PCs. As far as he could tell, arcades were the only place real 3D existed at all.

So in 1994 he flew to Japan to meet Suzuki and Irimajiri and ask Sega for help.

This is where the standard telling starts to smooth things over. In the version that traveled around the world this week, Nvidia signed a deal to supply a graphics chip for Sega's next console, the deal collapsed, and a generous Sega executive bailed the startup out anyway. Clean. Moral. Slightly boring.

The Japanese record is messier, and better.

Four sides or three

To draw a 3D object, a computer breaks its surface into small shapes. The industry was converging on triangles. Nvidia had decided to use quadrilaterals, paired with a method called forward texture mapping. Huang's assessment in 2026 was blunt: the correct answers were triangles and inverse texture mapping, and Nvidia picked neither.

At the time, though, the quad approach didn't look like a mistake. It looked like an edge. Nvidia's first chip, the NV1, shipped in 1995 through Diamond Multimedia as the Edge 3D, and it was a genuinely strange and ambitious product that handled graphics, audio, and even a game controller port. What it did not handle was compatibility with Microsoft's emerging DirectX standard, and that turned out to be the whole ballgame. As the Japanese tech outlet TechnoEdge put it, the NV1 never made it out of the industry's side channels.

Now here is the sentence that almost no English coverage has carried.

Irimajiri picked it.

In an interview with the Japanese financial outlet Kabutan, published in January 2025, Irimajiri described the origin himself, and he dates the decisive meeting a little later than Huang does. Around 1995, Huang came to Sega to sell him on the NV1. Irimajiri listened to the pitch about quadrilaterals producing smoother motion than triangles, decided the technology was novel and interesting, and acted on it twice: he put the NV1 behind the PC version of Virtua Fighter, and he commissioned Nvidia to develop the graphics chip for the next-generation console Sega was preparing, the machine that became the Dreamcast.

He was not a bystander who got unlucky. He was the buyer who evaluated an unconventional architecture, found it exciting, and bet a console on it. Yu Suzuki, according to TechnoEdge, was the Sega side's technical channel into the project, relaying what the hardware actually needed to do.

Sega had reasons to move fast. The Saturn had launched against Sony's PlayStation and was already losing; internally the company had already written off its own console's future and was reaching for the next one. Which is the exact state in which firms grab the most exciting option on the table.

Then, nine months in, the chip didn't work.

He flew back to say it did not work

He got on a plane, went to Japan, and told Irimajiri that the technology did not work and that Nvidia could not finish the project. He asked to be released from the contract. Coming from a company with almost no cash left, this was close to announcing your own funeral. Famitsu described it as walking yourself to the gallows.

And then he asked for money anyway.

A note on titles, because the coverage blurs this. Irimajiri was not Sega's president when he made the decision. He was vice president, and he also ran Sega's U.S. operation; the presidency came in 1998. Most outlets this week called him Sega's former president, which is true of the man today but easy to read as the rank he held then. He was senior enough to be heard, and junior enough that he had to go back and sell it to a board.

He sold it. Sega put $5 million into Nvidia. Reports differ on the mechanism, with some describing money Sega already owed under the contract being converted into equity, and others describing an additional investment on top of the contracted payment. What is not in dispute is the size and the effect. Nikkei Business reported that the $5 million amounted to nearly everything Nvidia had. It bought roughly six more months.

Irimajiri's own explanation, given to the Wall Street Journal in 2024, was that he still believed in Huang.

RIVA 128, and the bill that came due

Nvidia threw out the architecture, rebuilt around triangles, and shipped the RIVA 128 in 1997. It sold about a million units in four months. The company went public in 1999, and the line from there to GeForce, to general-purpose GPU computing, to the chips currently underwriting the entire AI industry, is unbroken.

Sega sold its Nvidia stake after the IPO for roughly $15 million. Tripling your money in about three years on a company whose only distinctive technology had just been declared worthless is, by any normal standard, an excellent trade.

It has since become a punchline. Huang has said in the past that if Sega had held on, the stake would be worth a trillion dollars, and this week that number was repeated more or less everywhere. It is worth noting that TechnoEdge ran the arithmetic (original share count, subsequent splits, dilution, current market capitalization) and landed closer to ¥10 trillion, or roughly $60 billion at current rates. Enormous. Not a trillion dollars. Their read is that Huang is speaking colloquially rather than quoting a figure, which sounds right.

Either way, Sega could not have held. Irimajiri became president in 1998, took the Dreamcast into the fight against the PlayStation, and lost. A third straight year of losses moved him out of the presidency in June 2000, and he left Sega that December. The stake had been sold to keep the lights on at a company that was running out of them.

Nobody in this story is a saint, which is why it's a good story

The anniversary itself is, gently speaking, constructed. Nvidia and Sega marked thirty years of partnership, and TechnoEdge pointed out what nobody at the event did: Sega has never shipped an Nvidia GPU in any of its home consoles. Not one. Nvidia silicon reached Sega arcade boards eventually, but that was the 2000s. Nvidia dates the relationship to the NV1 and the PC port of Virtua Fighter. TechnoEdge dates it to the rescue. Either way, the thing being celebrated grew out of a collaboration that never delivered what it was for.

More interesting is what Huang actually said on that stage, which is more precise than the headlines it produced. Per the speech as reported by Famitsu: He took the entire technical failure onto himself: Nvidia chose the wrong approach. And then he said that Nvidia had the wrong technology, but that Irimajiri had seen that Nvidia had the right people.

It's a gracious line. It's also, quietly, not quite accurate, and the inaccuracy runs in Irimajiri's favor. Irimajiri hadn't merely spotted good people attached to bad technology. He had looked at that same technology a year or two earlier and liked it enough to build a console around it. When he walked into the room to decide Nvidia's fate, he was not a wronged customer weighing whether to be generous to a stranger. He was a man deciding what to do about a mistake with his own fingerprints on it.

That is a harder decision. The easy move is to bury the evidence, take the contractual write-off, and never say the vendor's name again. Instead he went to his board and argued for putting more money into the company that had just failed him, on an argument he could not prove.

Huang said in Akihabara that the episode taught him something about how Japanese companies operate: that even when two companies are doing business, not everything is only business, and that friendship and mutual support matter too. He has earned the right to read it that way. But a less sentimental reading is available, and it is the more useful one. Irimajiri didn't reward loyalty. He separated a bad bet from the people who made it, including himself, and he was right about which one to keep.

The event closed with a demo of Virtua Fighter Crossroads, the series revival due in 2027, running on a PC. Sega has not announced its platforms, and Huang said even he was seeing it for the first time. Thirty years on, the series that started everything was being shown off at an Nvidia event.

So here's the question worth arguing about. When a supplier fails you because of a decision you signed off on together, what does your company do: write it off and walk, or work out which half of the wreck was worth keeping? Has anyone where you work ever made that second call?

参照