ludy

This operational chaos is hardly surprising when you look at the decay of Tesla’s local organization in China. The local team has essentially degenerated into a sluggish, do-nothing bureaucracy. Rumored to be run by toxic former joint-venture (JV) executives who imported old-school corporate rot, the local OTA team now functions like a lazy, slow-moving firm. They do nothing until regulatory pressure forces their hand, only to get repeatedly rejected due to poorly prepared compliance. Every minor local adaptation gets bottlenecked by the infamous "Elon email check" loop, while Musk himself rarely visits to resolve the gridlock. Instead of agile tech innovators, Tesla China’s software team has become a sluggish corporate zombie, rubber-stamping delays while the hardware of their loyal users slowly rots.

The clock is ticking Tesla. People are subscribing to FSD for HW3 V14 Lite and wasting money because it’s not gone wide. My buddy on the other hand subscribed to test out v14.3.4 on HW4 and he’s stuck on 14.2.2.5 Fix this. Do something about it. These people are trying to believe in FSD and then shit like this happens. It’s unacceptable with no answers. Either give them a free month, refund, or prorate it.

From a business standpoint, this aligns perfectly with Tesla’s classic strategy of planned obsolescence and incremental updates. By artificially gating features on older hardware, Tesla creates a "second-class citizen" experience for Intel owners, subtly pushing them to pay for expensive MCU hardware upgrades or trade in for a new car. At the same time, Tesla is likely holding the "Intel full-screen toggle" in its back pocket as a future marketing card. They will probably roll it out years later as a "miracle legacy optimization" once most Intel cars have been retired and the remaining user base is negligible, scoring cheap PR points.

There are some small differences in UI between Tesla FSD V14 Lite and HW4 V14. • Starting Self-Driving requires brake to confirm. It’s on by default. • There is no FSD intervention survey, just a voice message. • No Self-Driving stats (hope that can be added in the future). • Arrival options do not appear when approaching the destination. • It does not say “Driving to Whole Foods” when FSD is activated (I hope that’s added in the future as well). Keep in mind, this is a legacy Model Y with an Intel chip. All in all, V14 delivers on the main features from HW4 FSD. It parks in your driveway, in parking lots, at Superchargers, etc. It’s been really incredible. A night-and-day difference compared to V12.

This "revenge story" skips the most critical truth: CXMT didn't build DRAM from scratch. They scavenged Qimonda’s grave and hired Silicon Valley-trained talent. A highly commercialized, hyped market that lacks deep industrial sedimentation is actually toxic to real tech innovation. Acquirers and state capital only chase quick margins and political face ("mianzi"), while completely neglecting the organic cultivation of a local technical knowledge commons. Even today, the default playbook remains unchanged: avoid the painful work of foundational R&D, and instead rely on copycatting, IP poaching, or hostile acquisitions. Without Europe’s industrial graveyard to scavenge, that $20B of Hefei money would have produced nothing but high-tech PPTs.

In 2015, China tried to buy Micron for $23 billion. America said no. That rejection just created a company that made $4.5 billion profit in a single quarter. This is the best revenge story in tech. Here's what happened. China is the world's biggest buyer of memory chips but makes almost none. Total dependence on Samsung, SK Hynix, Micron. So China does the obvious thing: tries to buy Micron outright. Washington blocks it. National security. China being China decided to build its own Micron from zero. The city of Hefei came to Zhu with a wild proposal: build China's DRAM champion. Private investors bailed. DRAM is the hardest business in semiconductors: three giants, four decades of patents, billions burned before your first working chip. But first who the fuck is Zhu Yiming? Tsinghua physics grad. Moved to America for his masters. Spent years in Silicon Valley building memory chips (his chips were literally inside the Nintendo GameCube). In 2005 he went back to China with $100K and a few patents. Built GigaDevice. Today it's worth $21 billion. So when Hefei needed someone crazy enough to take on Samsung, Zhu was the guy. The government took 80% of the risk. $20+ billion committed. Codename: Project 506. One problem. You can't patent your way into DRAM. Samsung, Hynix, and Micron own everything. Four decades of moats. The solution came from a graveyard. Qimonda, Europe's DRAM champion, went bankrupt in 2009. Its patents and engineers were just lying around. CXMT scooped them up, including veterans with 20+ years of DRAM experience. A dead German company became the seed of China's memory industry. Then came the pain. Nine straight years of losses. US sanctions cut them off from the world's best chip machines. Washington banned its own government from touching their chips. They kept building anyway. Then AI happened. Every AI data center needs mountains of memory. The big three couldn't make enough. Prices went vertical, some memory jumped 90% in a single quarter. And CXMT, born from a rejection and a bankruptcy, was standing right there with ready factories. Q1 2026: ~$7 billion revenue. Up 719% in a year. ~$4.5 billion profit. In one quarter. After a decade of bleeding. And now the ending nobody had on their bingo card: Apple, the most American company alive, is testing CXMT's chips and lobbying Washington to let US companies buy MORE Chinese memory. The country that blocked China from buying Micron now has its biggest company asking permission to buy from Micron's Chinese rival. Blocking that deal was supposed to protect American chips. Instead it forced China to build a competitor that owns 11% of global DRAM capacity, heading to 15% by 2028. So, honest question: was blocking the Micron deal the biggest self-own of the chip war, or was China building this anyway? Pick a side.

APPLE HAS STARTED TESTING CXMT’S DRAM FOR APPLE DEVICES SOLD IN CHINA — FT

Zero dependency on external frontier APIs doesn’t mean farming with horses. It means running highly optimized open-source or local models on your own silicon. To do that efficiently, you need world-class local systems optimization and robust classical software. You still get the tractor, but you actually own the keys. Besides, what is easily rented or borrowed from others is rarely where true, enduring value lies. Often, the most precious capabilities are those custom-built and kept offline, rather than the public APIs everyone is chasing. As for Europe, not having a flashy "frontier AI" monopoly today doesn't mean they lack technological depth. Their caution in tech adoption is a deliberate, long-term strategy for societal resilience and systemic sustainability, rather than blindly chasing the hype cycle.

@ludydev @antirez that's like saying a farmer not being reliant on a tractor company and instead using horses is better off despite being at a massive disadvantage. It simply isn't true. AI is a massive force multiplier for talented people

If that happens Europe is officially fucked.

The Summer "AI Hype skepticism" is a cyclical Wall Street pattern. Every July, analysts exploit developments in the Chinese tech sector as short-term "bear narratives" to trigger retail panic-selling right before US Q2 earnings. Let's look at the facts vs. Wall Street's exaggerated claims (2023-2026): 👇 (1/5) 1/ The "Domestic GPU Substitution" narrative: Bears claimed that export restrictions and the rise of local hardware players like Huawei Ascend would permanently cripple NVDA's growth. Reality: Huawei's engineering progress is a massive success for China's local ecosystem. But globally, NVDA's backlog from US/EU cloud giants was so large that global demand continued to outstrip supply, driving profit margins to new highs. (2/5) 2/ The "LLM Price War" narrative: When Chinese tech giants (ByteDance, Alibaba, Tencent) slashed API prices by 90%, bears claimed "AI software is unprofitable, so GPU buying will stop." Reality: Lower API costs democratized AI, sparking an explosion in application usage. This surge in daily active users actually drove higher inference-side compute demand, forcing firms to buy more hardware. (3/5) 3/ The "Custom Chip (DeepSeek/Zhipu)" narrative: Today, reports of local firms designing custom inference chips are used as a short-term trigger to sell off the semiconductor sector. Reality: Designing custom chips reflects brilliant engineering talent in the region. But NVDA's core long-term hyperscaler pipeline remains fully secured. The pullback is a healthy leverage flush-out, not a structural reversal. (4/5) Conclusion: Every summer, Wall Street uses headlines to create short-term volatility, accumulates cheaper shares, and watches tech earnings beat expectations in August. Focus on the long-term fundamentals. 💎🙌 #NVDA #INTC #SPCX #TSLA #NASDAQ #AI (5/5)

While global media frames this as the next stage of the AI chip war, a sober look at the domestic ecosystem reveals five major contradictions: The "Vibe" vs. Actual Adoption: While DeepSeek (V3/R1) won the hearts of researchers globally for its training efficiency, actual developer adoption for core, high-value production workflows still belongs to Claude and GPT-4o. DeepSeek is heavily discussed online, but in production, it's often relegated to a cost-saving secondary layer. Censorship and Compliance Bottlenecks: The biggest bottleneck for Chinese AI isn't just hardware—it’s the regulatory environment. Strict algorithm filings, security reviews, and mandatory real-time content filtering severely compromise the reasoning capabilities of domestic models, preventing a thriving user-facing application ecosystem from taking off. The Government Subsidy Game: In the domestic tech scene, "self-developed AI chips" and "semiconductor self-reliance" are the ultimate keywords to unlock massive local government grants, tax incentives, and state-backed funding. Historically, many of these projects exist on PPT presentations to secure political capital rather than entering mass commercial production. The Secret Sauce is Still Distilled Claude: For actual, unconstrained productivity, many domestic developers rely on workarounds to access Claude or OpenAI models directly, or they use their outputs to "distill" smaller local models. The reliance on frontier Western models for raw reasoning remains high, despite the public PR. The NVIDIA Gravity Well & Supply Chain Sanctions: DeepSeek’s brilliance lies in its software engineers writing highly optimized CUDA/Triton kernels on Nvidia hardware. Abandoning this to build a proprietary hardware stack—while facing severe US export sanctions on advanced fabrication (SMIC limitations), CoWoS packaging, and HBM memory—is a massive technological regression.

CHINA’S DEEPSEEK DEVELOPING ITS OWN AI CHIP, SOURCES SAY -RTRS

The Summer "AI Skepticism" comedy show is on its 4th season. 🎭 Every July, Wall Street plays the exact same script to scare retail out of their shares before Q2 earnings. Here is the breakdown of their fake panic vs. the actual quantitative reality (2023-2026): 👇 (1/5) 🚨 2023: "ChatGPT traffic is declining, generative AI is a fad!" 🔗 reuters.com/technology/cha… ▪️ MOCKERY: Analysts thought a dip in Similarweb browser hits meant AI was dead, ignoring that people actually use API integration. ▪️ ANALYSIS: Simple web traffic decline reflected the drop in retail curiosity, while B2B API token consumption was growing exponentially. Wall Street mistook consumer play-time for enterprise demand. (2/5) 🚨 2024: "GenAI is a $1 Trillion capex bubble with zero ROI!" 🔗 goldmansachs.com/insights/pages… ▪️ MOCKERY: Wall Street expects a massive infrastructure buildout to print billions in software revenue within 6 months of buying GPUs. ▪️ ANALYSIS: Infrastructure has a J-curve utility. Hardware CAPEX must precede software monetization. Tech giants are securing the picks and shovels. Expecting instant ROI during the infrastructure phase is a fundamental macro misunderstanding. (3/5) 🚨 2025: "Scaling Law is dead, AI gridlock due to power limits!" 🔗 bloomberg.com/news/articles/… ▪️ MOCKERY: Bears thought the AI boom would end because we ran out of internet data to pre-train, and gridlocks would shut down the fabs. ▪️ ANALYSIS: The industry pivoted to "Test-Time Compute" (e.g. OpenAI o1/o3 reasoning models), shifting scaling from pre-training to inference. Meanwhile, mega nuclear energy deals solved the power constraint. Tech always routes around bottlenecks. (4/5) 🚨 2026 (Today): "Samsung 19x profit is a 'sell the news' AI Stress Test!" ▪️ MOCKERY: Fabs are printing money, NVDA roadmap is fully intact, yet retail panics over localized Korean leverage liquidations. ▪️ ANALYSIS: Today's pullback is a classic H1-H2 transition rebalancing, amplified by a margin flush in Kospi. The secular hardware demand remains structurally robust. Conclusion: Every summer they drop the market, buy your cheap shares, and watch Q2 earnings crush expectations in August. Don't be their exit liquidity. Hold the line. 💎🙌 #NVDA #INTC #SPCX #TSLA #NASDAQ #AI

@Nick_Vuj @jukan05 Classic retail mindset. They will always spin a bearish narrative to justify their anxiety, no matter how good the numbers are. Honestly, filtering out the noise—especially blocking the toxic Chinese retail community—is the only way to keep your sanity.

Unfortunately it will. Most X “experts” (those with many ignorant followers) will lazily say it’s cyclical or those with simple TA will spout “look at how extended the chart is” stupidity without understanding the structural changes AI is going to effect our lives in every imaginable way, now and in the future. Really disappointing.

Looking at the current market mood, it feels like memory is going down either way. Case 1: If they deliver earnings above market expectations: “Isn’t this peak-out? Do you really think these profits are sustainable? Memory is cyclical, and this is the top.” Case 2: If they miss market expectations: “Memory is over now. Even after excluding bonus provisions, they should have easily printed KRW90tn. Sell on news.”

I just published The frustrating truth about tracking body fat (And why I built a zero-bloat calculator) medium.com/p/the-frustrat…

@BLKMDL3 Since HW3 + Intel can run the full-screen FSD visualization so smoothly, why has the Tesla OTA team been so slow to bring this to older hardware? It’s honestly ridiculous. @Tesla @elonmusk @Tesla_AI @wholemars @SawyerMerritt @28delayslater @BLKMDL3

Doing some more FSD v14 lite drives and testing now before I write up a big overall article of my thoughts on the last 4 days with it.

@kymeuni1 @huoshan007 X炒股，最先屏蔽就是简中的傻逼v

@huoshan007 这已经是很久之前的了，真不明白为什么要把国内的这套还拿在X上面来找流量

兄弟们小心点！这哥们突然开始猛空半导体，味儿不太对。 Leopold Aschenbrenner，新开的空头仓位： $SMH 做空 21.3 亿美元 $AMD 做空 10.6 亿美元 $NVDA 做空 16.6 亿美元 $MU 做空 6.86 亿美元 重点是，这人 24 岁把 2.25 亿美元干到 200 亿美元。 这种级别的人下这么重的空单，很难说只是手痒。 半导体这波，怕是他提前闻到什么味了……。

I just published Naming music projects is surprisingly painful. I built a free tool to fix it. medium.com/p/naming-music…

@JonBbC_TechGeek full screen only for fsd mode.the @Tesla ota team still a joker

V14 LITE vs. DRIVE THRU: I think it's very cool how V14 Lite has inherited some higher level intelligence from its bigger brother. It actually stops at the right places while going through a drive thru now. 🤯This is next level. You have to watch it to believe it. And the crazy part? There was no lead car in front of my 2019 Model 3 to stop it. This is definitely new behavior and it's way different than Full Self-Driving v12.6.4. Congratulations again to the entire @Tesla_AI team for a job well done here. I'm continually amazed at some of the things I'm seeing here in Chicago. Please ship it for wider release, I think it's ready.

Tech stocks dipped today, but this isn’t the ‘AI bubble bursting.’ Meta’s move to sell excess AI compute (stock +10%) is smart monetization of their massive capex—not failure. It adds supply, pressuring pure infra plays, while Fed’s Warsh comments added macro uncertainty. Real AI demand (cloud growth, adoption) remains strong. Pullback = rotation & profit-taking, not collapse. Long-term winners building real value. #AI #Meta #NVDA #MSFT #GOOGL #TechStocks

Exactly. Intel HW3 owners can actually go full-screen and zoom/pan the driving visualization in FSD V14 Lite when moving (speed > 0). The hardware can render it. Gating the parked 3D view is a software/policy decision, not a hardware impossibility. We pay the same subscription; we deserve optimized features, not software locks.

Genuine question for Tesla: Why are those of us with HW3 (Intel/AMD) paying the same monthly FSD subscription as owners with AI4/HW4? We don’t get the same features—no Grok, Intel owners don’t get the full-screen parking view, and other new capabilities are hardware-limited. If we’re paying the same price, shouldn’t we be getting the same value? Or shouldn’t the subscription cost reflect the hardware differences? Curious what everyone thinks.

In this short clip w/ v14 Lite on HW3, I had to intervene to press on the accelerator to move past the construction workers. Unlike my human eyes, the cameras did not recognize the truck driver motioning with his hand for me to drive past him. I pushed it to the last second before I intervened. Shortly after that, I had to disengage to prevent my car from running over a metal fitting at the end of a construction hose that was lying in the road. For those of you who are regularly using v14 (AI4), can you please comment on any different behaviors that may have taken place? How in the world would you maintain an FSD streak in a situation like this? I really want to know because this scenario is not completely abnormal. @DavidMoss @DevinOlsenn @01Ananto @BLKMDL3 @wholemars @DirtyTesLa @AIDRIVR @ChuckCook

Circling hardware plays: #SpaceX (ecosystem), #Nvidia (#NVDA), #Tesla (#TSLA), #Intel (#INTC). AI-driven storage & chip costs exploding (DRAM +40-55%, data centers taking 70% of memory). Hardware suppliers win big. Server-heavy names lose. reddit.com Hardware = Profit Optimism • #NVDA: AI GPU king, pricing power intact. • #INTC: Foundry surge + Tesla/SpaceX Terafab deals. • #TSLA: Custom chips & full-stack hardware edge. They sell the shovels → strong margins & volume. #AIHardware Real Pain: Server & Streaming Companies #Meta (#META), #Oracle (#ORCL), streaming giants (#NFLX, #DIS): Massive CapEx ($600B+ hyperscaler total) for GPUs/storage. Cash flow pressure, margin squeeze as costs soar before AI payback. #StreamingCosts #AIServers Bottom line: Hardware stocks optimistic. Server/streaming-dependent names threatened. Bullish #NVDA #INTC #TSLA | Cautious #META #ORCL #NFLX Thoughts? #AIStocks #ChipShortage (DYOR) 🚀

@TechRoo @TexasTSLA U must a Amd+hw3?

@ludydev @TexasTSLA Wdym? I have that and it is full screen FSD interface

FSD 14 lite.....is it good enough for one more wave to be sent out? 👀👀👀