Local AI brings the power back to ppl ✊🏽

⚡ INSIGHT: China's GLM-5.2 is reportedly the first Chinese AI model to match and often beat leading public AI models from US labs.

Now decentralize and open-source it

SpaceX's D3 chip unlocks orbital data centers that shatter Earth's AI energy limits

Sun-synchronous orbits deliver 5-13x more solar energy per area than ground panels while deep space vacuum cooling eliminates water use, targeting one terawatt of capacity impossible on terrestrial grids.

Mars in 30 Years: Realistic Timeline Using Starship + Optimus + AI

Assuming Starship achieves rapid reusability and high-cadence launches:

> Years 1–5 (Uncrewed Phase): Cargo Starships deliver Optimus fleets, ISRU pilots, power systems, and initial habitats. Robotic construction of propellant plants, solar/nuclear infrastructure, and resource mapping.

> Years 5–10 (Initial Crewed Outpost): First human landings (~2029–2031 range, per evolving plans). Small crews supported by robots establish a base, validate life support, and expand ISRU. Focus on self-sufficiency milestones.

> Years 10–20: Fleet-scale deliveries enable city-scale growth (hundreds of residents). Advanced robotics and AI handle most construction and operations. Food, power, and manufacturing loops close. Governance and psychological support systems mature.

> Years 20–30: Self-sustaining city with thousands of inhabitants, local industry, and expansion capability. Multiplanetary redundancy achieved, reducing existential risks.

This timeline assumes technical success, regulatory support, and iterative learning. Optimus and AI significantly de-risk and accelerate development by performing hazardous tasks autonomously.

Pov: You took control of your life

Remember: sometimes you don't get what you want because you deserve better

Cry cry more. Open Source models gonna whoop your aszz 😂😂😂

“Everything in life is gamble, so you might as well bet on yourself”

When everyone tells you how bad it feels to be used, but you're too useless for anyone to use

AI isn’t slowing down because the tech hit a wall.

It’s slowing down because one very specific safety culture won the narrative.

Anthropic’s EA-flavored doomerism helped turn “frontier AI” from a tool into something governments now treat like enriched uranium. Once that framing wins, public access becomes a privilege, not the default.

Less access means slower diffusion, less competition, fewer builders, and more power for insiders.

The safest AI future isn’t a tiny priesthood deciding who gets intelligence.

It’s broad access, transparency, competition, and speed.

A post aging society would be one of the biggest psychological shocks in human history.

Imagine people no longer planning life around decline, disease and a slow countdown. You could start over at 60. Study something new at 90. Build a company at 120. Fall in love again without the shadow of “too late” hanging over everything.

Of course there would be problems. Inequality could get ugly. Old power structures might refuse to die, literally. People would need to rethink family, retirement, identity and what a “normal life” even means.

But honestly, the upside is insane.

More time with the people we love. More time to learn. More time to fix mistakes. More time to become who we actually want to be.

A society without aging would not be perfect.

But compared to the current system where everyone slowly breaks and we pretend that’s somehow noble?

I’ll take the weird future.

The creator of High Bandwidth Memory said something that reframes the entire AI investment thesis, AI equals memory (Save this).

Most people still think about AI hardware through a training lens.

During training, the bottleneck is raw compute, GPUs stay near 100% utilization crunching through billions of gradient updates.

Inference is a completely different problem.

When a model generates a response, it produces tokens one at a time and at every single step, the entire model has to be loaded from memory into the processor to generate just one token.

The GPU cores sit there, waiting for data to arrive.

This is what engineers mean when they say inference is memory bound, the bottleneck is not how many calculations you can do per second but rather how fast you can move data from memory to the chip.

Adding more GPUs does not fix a memory bandwidth problem, it just gives you more processors starving for the same data.

Modern LLMs use a KV cache, a data structure that stores the conversation's context so the model does not have to recompute it from scratch on each step.

The KV cache is what gives a model its memory of the conversation.

It grows with every token and for long documents or deep reasoning chains, it can dwarf the model weights themselves in memory consumption.

This means memory directly determines how long a context the model can hold, how many users you can serve simultaneously, how fast it responds and how cheaply you can run it.

A memory constrained model is not just slower but rather qualitatively worse, it forgets earlier parts of the conversation, truncates context and hallucinates more because it literally cannot hold the relevant information long enough to use it.

The world now spends more on inference than training, and every ChatGPT query, every Claude document analysis, every API call is an inference workload.

Inference economics, cost per token, latency, context length, concurrent users are memory problems first and compute problems second.

The companies that control memory bandwidth and supply are not suppliers to the AI trade but rather are the AI trade.

Jensen Huang is investing in every photonics company he can find and the reason why tells you everything about where AI is headed (Save this).

Lip-Bu Tan, the CEO of Intel says, when he looks for investment opportunities, he looks for the bottleneck and right now, the bottleneck is the interconnect, the pipes that move data between chips inside an AI data center.

That is why he backed Credo Semiconductor, Astera Labs and Celestial AI on the optical side.

Here is the simple version of what the interconnect bottleneck actually means.

Think of an AI data center like a city, the GPUs are the buildings where all the work happens but for those buildings to function, you need roads connecting them, fast roads that can carry enormous traffic without congestion.

And those roads are now the single biggest constraint on AI performance.

As clusters scale to hundreds of thousands of GPUs, traditional copper wiring is hitting its physical limits and that is where this entire sector comes in.

Credo Semiconductor ($CRDO) is the most direct pure play on this theme, Credo makes high speed cables and optical chips that connect GPUs inside data center racks.

Their revenue tripled in fiscal 2026 to $1.3 billion, growing 272% year over year at its peak and four of the world's largest hyperscalers each individually account for more than 10% of Credo's revenue.

Astera Labs ($ALAB) solves the connection problem between different chip types.

Astera makes the PCIe and connectivity chips that manage data flow between GPUs, CPUs, and memory without errors or slowdowns.

Their revenue grew 93% year over year to $308 million in Q1 2026 alone.

The optical companies are where the longer-term and potentially larger opportunity lives.

Copper has physical limits, you can only push electrical signals so far before the signal degrades, the heat spikes and power consumption explodes.

The solution is light, fiber optic connections that move data using photons instead of electrons which is faster, cooler and far more energy efficient.

Jensen Huang made this clear at Computex 2026 because copper works as long as physically possible but at greater distances and larger scale, optics takes over.

Coherent ($COHR) is the most established optical company in this space.

Coherent makes the lasers, transceivers, and optical components at the foundation of all fiber optic communications.

Nvidia signed a multibillion-dollar purchase commitment and invested $2 billion directly into the company and their customer order books are already extending out to 2028.

Marvell ($MRVL) is the most comprehensive bet across the entire connectivity stack.

Marvell makes chips for optical networking, PCIe switching and custom AI silicon.

Jensen Huang called Marvell the next trillion dollar company at Computex 2026 and backed it with a $2 billion Nvidia investment.

Marvell also acquired Celestial AI, the exact company Lip-Bu Tan backed for $3.25 billion, gaining photonic fabric technology delivering 16 terabits per second of bandwidth.

Lumentum ($LITE), Corning ($GLW), and Ciena ($CIEN) round out the major public names.

Lumentum received a $2 billion Nvidia investment for laser and photonics components.

Corning known mostly for phone glass received $500 million from Nvidia for optical connectivity work and is up over 100% year to date.

Ciena runs the optical networking systems between data centers and is seeing analyst price targets raised on the back of the AI optics boom.

Every time a hyperscaler spends a billion dollars on Nvidia GPUs, the surrounding infrastructure, cables, switches, transceivers, optical components has to be upgraded to match.

The smarter the GPU gets, the more the interconnect matters.

Nvidia has committed at least $6.5 billion to photonics companies in the past 4 months alone and the companies building the roads between the GPUs may end up being just as valuable as the companies building the GPUs themselves.

Bears are back at it again claiming Apple buying Chinese memory chips will crush the memory industry and here's why they're completely wrong (Save this).

CXMT makes the kind of memory that goes in your iPhone and MacBook, standard, commodity chips that any consumer device needs.

The memory that actually matters for AI is something completely different.

It's called High-Bandwidth Memory, a highly specialized chip that is custom-engineered for AI servers and physically stacked directly on top of Nvidia GPUs.

It is one of the most technically complex products in the semiconductor industry, and only three companies on earth can produce it at scale.

CXMT is not one of them.

CXMT's most ambitious goal right now is to begin certifying HBM3, the previous generation for mass production by end of year.

The Big 3 are already shipping HBM3E, the generation after that, to hyperscaler customers today.

That gap doesn't close quickly. because HBM development requires years of co-engineering directly with customers like Nvidia, Google and Microsoft, embedded in their chip design process 18 to 24 months before a product ships.

CXMT has never shipped a single HBM chip to a Western hyperscaler.

They have no supply relationships, lack co-development history, and still do not have a certified product in the generation that actually matters.

When CXMT potentially wins some MacBook memory business from Apple, (very low probability of US approving this) it is not taking anything from the part of the memory market that actually drives valuations.

It is like worrying that a budget airline winning a short-haul route will hurt Boeing's order book for long-haul widebody jets.

They are different product, customer and market entirely.

Everyone thinks Apple buying cheap Chinese memory chips is a threat to Micron and the memory industry and it's not even close (Save this).

Dylan of Semi-Analysis is absolutely right and once you understand why, the whole narrative falls apart.

CXMT is not sitting on a warehouse full of unsold chips waiting for Apple to show up.

Every gigabyte it produces is already committed to Chinese domestic demand, phones, laptops and consumer electronics made and sold inside China.

There is no flood of cheap supply hitting the global market and there is no price war being waged against Samsung, SK Hynix or Micron.

Now look at the chart I attached from Semi-Analysis, from 2022 through mid-2025, all four companies CXMT, Samsung, SK Hynix and Micron were priced at roughly the same depressed level, around $0.20–0.50 per gigabyte.

That was the commodity DRAM downcycle, everyone was getting hurt equally bu then in late 2025, three of those lines broke sharply upward.

Micron, SK Hynix, and Samsung are all projected to reach $2.70–$2.90 per gigabyte by end of 2027.

CXMT is also rising but its green line tops out around $2.10, a full 25–30% below where the Big 3 are headed.

That gap is not a sign that CXMT is catching up but rather a sign that the Big 3 are selling a completely different and far more valuable product, High Bandwidth Memory, the specialized chips that go inside every Nvidia AI GPU.

HBM generates 3–5x more revenue per wafer than standard commodity DRAM, CXMT cannot make it at any meaningful scale.

So while Apple potentially buys some cheap commodity DRAM from a Chinese supplier, Micron is busy repricing its entire product mix into the highest margin memory product ever made.

Those are two completely separate stories and the market is confusing them.

Apple buying CXMT chips does not touch Micron's HBM revenue and on top of all of that, there is an extremely low probability the US government actually approves this.

CXMT is on the Pentagon's Chinese Military Company blacklist for alleged ties to the People's Liberation Army.

The Trump administration has been tightening semiconductor export controls throughout 2025 and 2026, Apple is lobbying hard, but the regulatory environment is working directly against them.

The most likely outcome is that this never gets approved at all.

$OKLO $CCJ $SMR $CEG $LEU | American Fusion prepares a five megawatt (MW) Texatron preproduction engine for independent testing following final component assembly.

The upcoming campaign aims to gather verified scientific diagnostic data under Texas state regulatory oversight.

$MU $SNDK $WDC $INTC $ASML | Sources report that China's CXMT secures a multiyear server memory supply contract with Tencent worth around $3 billion.

The massive chip pipeline stabilizes essential component access for cloud networks amid a global DRAM shortage.

$SPCX $RKLB $VOYG $PL $ASTS | A new US think tank report warns that China now leads the United States in GPS style satellite navigation and remote sensing.

The strategic shift also places Beijing ahead in orbital warfare and advanced counterspace weapons.

$MP $UUUU $UAMY $USAR $LAC $CRML | The EU pitches Brazil an advanced critical minerals deal, promising domestic refining infrastructure and tech transfers.

The strategic framework counters extraction heavy models favored by global competitors like Washington and Beijing.

Anthropic's internal loop engineering playbook just got leaked.

And it's the most valuable AI guide I've read all year.

This guide is packed with info, and there are five things you need to know if you want to maximize your AI productivity with loops (save this):

1. You should structure every loop around these 5 principles:

• Discovery → Let the agent find its own work (CI fails, issues, commits)
• Handoff → Give every task its own isolated git worktree
• Verification → Never let the generator grade its own work
• Persistence → Always write state to disk (markdown or board)
• Scheduling → Run it on a timer so it works while you sleep

2. Separate Generator from Evaluator (most important rule)

Use two agents: one writes, the other is a skeptical judge who assumes the code is broken.

Make the evaluator act (run tests, click buttons, take screenshots) - this is what actually stops bad output.

3. Build with these 6 parts:

• Automations (the timer)
• Worktrees (safe parallelism)
• Skills (permanent project knowledge)
• Connectors (talk to GitHub, Linear, etc.)
• Sub-agents (generator + evaluator)
• Memory (state files that survive between runs)

4. Things you must watch out for with loop engineering:

• Verification debt (use verification agents)
• Losing understanding of your own codebase (start fresh if needed)
• Token costs exploding (solution below)
• Cognitive surrender (don't stop thinking because “the loop handles it”)

5. Solving token costs

Loop engineering can be crazy expensive.

I recommend you use an 80/20 "barbell" approach to loop engineering.

For your most complex tasks that require the best intelligence, use expensive models (Opus).

For the remaining 80% of your tasks (the gruntwork), use cheap, open-source models within the Claude Code harness (GLM-5.2 is great for code execution).

Save these 5 rules so you don't forget them.