SatoshiGold 🦇🔊

Bitcoin miner since 2012. CEO of LIGHTNINGASIC. Founder of

[SatoshiGold 🦇🔊 @realSatoshiGold]

Don't trust PHA .It's co-founded by scammer @bgmshana .I hired him as lead dev of BTG. Give him title. He faked info to promote PHA.
Stay away from scammers.

[Phala @PhalaNetwork]

2024 was the year AI, blockchain, and TEE converged. From scaling to setting new benchmarks for decentralized AI, Phala has solidified its place as the go-to AI coprocessor for the decentralized web. 🧵

Read Phala's Year in Review: https://t.co/Ryj0NwXMLw

[SatoshiGold 🦇🔊 @realSatoshiGold]

@cnfinancewatch 换股。

[SatoshiGold 🦇🔊 @realSatoshiGold]

@BrianRoemmele smart. now @ethercoinorg will do same thing for AI.

[SatoshiGold 🦇🔊 @realSatoshiGold]

@FGuzmanAI @karpathy ASIC is the best choice for local reasoning.

[SatoshiGold 🦇🔊 @realSatoshiGold]

@_0xghost_ https://t.co/CPwtnywSwB

[SatoshiGold 🦇🔊 @realSatoshiGold]

@BarrySilbert Yes. https://t.co/IIUXD8COPU.

[SatoshiGold 🦇🔊 @realSatoshiGold]

The first shipment—10 sets of solar-powered smart microgrid systems—was dispatched from Shenzhen, destined for Rwanda. June 13, 2026.
https://t.co/lz2bERqMPg, light up every home.

[SatoshiGold 🦇🔊 @realSatoshiGold]

@RealDanODowd @elonmusk 100% agree.

[SatoshiGold 🦇🔊 @realSatoshiGold]

@0xSisyphus agree.

[SatoshiGold 🦇🔊 @realSatoshiGold]

为浏览率买单，为梦想窒息。

[SatoshiGold 🦇🔊 @realSatoshiGold]

币圈不合规交易所，号称拿了 SpaceX IPO 的额度卖给币圈人，大半个币圈人都掏钱参与打新了，现在IPO了，没额度，退款还要给几百刀的手续费。
强监管的成熟市场，创新的币圈不合规交易所没有成功的机会。
创新总在意想不到的地方。

[SatoshiGold 🦇🔊 @realSatoshiGold]

@mischamartijn Seven years ago Tesla sold me Full Self-Driving with one promise: the hardware in your car is all you’ll ever need. 7 years.
What's the compensation?

[SatoshiGold 🦇🔊 @realSatoshiGold]

@ma_sane39 I was there too, in 2022 for https://t.co/IIUXD8COPU AI computing power.

[SatoshiGold 🦇🔊 @realSatoshiGold]

@davepl1968 do you know how big the ISS is? and the cost of it?

[Elizabeth Warren @SenWarren]

The SEC must delay the SpaceX IPO until investors are protected.

[SatoshiGold 🦇🔊 @realSatoshiGold]

Saylor defeated himself. STRC interest comes from the bitcoin price rise. That's why STRC unplugged when bitcoin price dropped.

[Milk Road @milkroaddaily]

Saylor's acid test for $STRC vs just holding $BTC:

If you bought Bitcoin on Oct 6th of last year, you're down 35-40% with zero dividends.

If you bought STRC, "you would be holding par and you would have collected 7.4% in dividends."

For long-term holders, Bitcoin still wins if you can hold a decade.

But for working capital you need in the next 12-24 months: $STRC pays 4x the money market rate while betting only that Bitcoin doesn't go to zero.

FT @Saylor @Sam_Vadas @SchwabNetwork.

[SatoshiGold 🦇🔊 @realSatoshiGold]

@sudoingX LINUX only.

[SatoshiGold 🦇🔊 @realSatoshiGold]

@ZacksJerryRig No. what just released is just one space IDC center, not include computing part yet.

[SatoshiGold 🦇🔊 @realSatoshiGold]

@michaeljmcnair burning cash reasoning cannot afford the rental cost.

[SatoshiGold 🦇🔊 @realSatoshiGold]

Nobody can beyond TSMS. Let's verify it in 2 years.

[Ming @tslaming]

Elon Musk is essentially saying that, if his claim holds true, Tesla’s upcoming AI6 chip may extract an unusually high amount of working AI computing power from each silicon wafer 🔥

Here is exactly what it means in plain English 🧵

BREAKING DOWN THE JARGON

💿 Wafer: Computer chips aren't manufactured one by one. They are printed by the dozens or hundreds onto a giant, shiny, pizza-sized disc of silicon called a "wafer." Once the intricate printing process is done, this wafer is sliced up into individual square chips.

🏭 Yield: When you manufacture chips on a silicon wafer, not all of them come out usable. A microscopic speck of dust or a tiny printing flaw can damage a chip. "Yield" is the percentage of chips on the wafer that work well enough to be used. Sometimes a chip with a small flaw can still be used by disabling the damaged part or running it at a lower performance level.

🧠 Usable intelligence: This is Elon's phrase, not a standard chip-industry metric. He likely means useful AI processing power: how much real neural-network work Tesla can get from the chips that actually work. It is not "intelligence" in the human sense.

CONNECTING THE PIECES

Think of a silicon wafer like a large baking sheet, and the individual chips as cookies being baked on that sheet.

🍪 Every chip design tries to pack as many transistors—the brain cells of a chip—into a given area as possible to maximize computing power. However, silicon manufacturing is incredibly delicate.

📉 Imagine that every time you bake, a few random drops of liquid soap fall onto the baking sheet. If a drop lands on a cookie, that cookie is ruined. This is the reality of manufacturing defects.

🎯 This creates a massive trade-off problem. A bigger chip hurts you twice: fewer of them fit on each wafer, and each one is a bigger target with a higher chance of being ruined by a random defect. If you make the chips very small, more of them fit on the wafer and yield may improve, but each chip may not have enough computing power for serious AI workloads.

HOW TESLA MIGHT PULL THIS OFF

No one outside Tesla knows the exact AI6 design yet, so this part is best understood as an educated guess.

To maximize real neural-network work from each wafer, Tesla would likely need to optimize several things at once:

1️⃣ First, the chip must be the right size. Public reporting suggests AI5 is not a maximum-size chip, but closer to a more practical size that still leaves room for strong performance. That matters because a chip that is too large may be powerful, but fewer fit on each wafer and each one has a higher chance of being ruined by defects.

2️⃣ Second, the chip can be specialized for Tesla’s own AI workloads instead of being a general-purpose processor. If the hardware is designed mainly around the kinds of neural networks Tesla actually runs in cars, robots, and data centers, more of the silicon can be spent on useful AI work.

3️⃣ Third, Tesla may be able to improve yield by designing in redundancy, meaning a chip with a small defect could still be salvaged and used instead of being thrown away. This is common in advanced chip design, though we do not know the exact AI6 approach.

4️⃣ Fourth, memory bandwidth and power efficiency matter immensely. Public AI5 details suggest Tesla is paying close attention to feeding data quickly into the chip. That is important because an AI chip is only useful if it can keep its AI engines busy without wasting too much energy.

So the achievement would not come from one magic trick. It would come from balancing chip size, specialization, memory, power, and manufacturability better than competitors.

THE BOTTOM LINE

The key point is that Elon’s claim is not just about making the most powerful individual chip. It is about maximizing the total amount of useful AI compute Tesla gets from each wafer.

The rough equation is:

Usable AI compute per wafer = how many chips fit on the wafer × how many survive manufacturing × how powerful each working chip is

So when Elon says AI6 might set a record for "the most amount of usable intelligence from a wafer when factoring in yield," he is saying Tesla may have optimized the whole system: chip size, defect tolerance, AI specialization, memory, power efficiency, and manufacturing yield.

In plain English, the win would not simply be:

"We made a giant chip".

It would be:

"We got the most working AI brainpower out of each expensive silicon wafer".

That is why the yield part matters. A chip only counts if it actually works, and the real engineering achievement is maximizing the amount of useful AI work that survives the manufacturing process.

[SatoshiGold 🦇🔊 @realSatoshiGold]

@iwelsh Till today FSD not delivery yet, 10 years passed.