Lunar Think Trade

THE BILLION DOLLAR QUESTION: DO EFFICIENT MODELS KILL MEMORY DEMAND? $SNDK $MU TurboQuant IS NOT NOVEL. FREE MEMORY STOCK DISCOUNT! The Google blog post dropped March 24, 2026, but the paper had been on arXiv since April 2025, and was accepted at ICLR 2026. GOOGLE IS PUBLISHING MARKETING AROUND EXISTING ACADEMIC WORK. TurboQuant is actually two papers bundled together into a "unified system." 1. PolarQuant, converts vectors from Cartesian to polar coordinates to eliminate memory overhead from normalization constants. 2. QJL (Quantized Johnson-Lindenstrauss), is a 1-bit error correction step. What it actually does technologically: the KV cache size scales with both model dimensions and context length, creating a significant bottleneck for long-context inference. TurboQuant achieves absolute quality neutrality at 3.5 bits per channel, maintains 100% recall on needle-in-a-haystack tests up to 104k tokens, and compresses the KV cache by over 5x. NOW TO ANSWER THE BILLION DOLLAR QUESTION in my humble opinion: The empirical answer is almost certainly no for me. My main hypothesis: EFFICIENT MODELS DO NOT REDUCE MEMORY ORDERS. THEY REDIRECT FREED MEMORY INTO MORE SCALE, LONGER CONTEXT, AND LARGER BATCHES. Let's take a look at the Jevons Paradox. In January of 2026, a paper on the Structural Jevons Paradox stated that the unit price of intelligence falls, downstream firms do not simply run the same workloads more cheapy, they redesign their agent architectures to consume dramatically more compute. Falling API prices induce developers to adopt deeper reasoning loops, larger context windows, tool-augmented multi-agent workflows, and chain-of-thought pipelines that multiply token consumption per task. This is why I do not think TurboQuant does not affect demand. It just simply changes what HBMs are used for. When 3-bit KV frees up HBM headroom, operators immediately run larger batches or context windows to even more tokens. These longer context windows ACTUALLY INCREASE total DRAM demand because the KV cache still grows linearly with sequence length, just compressed. To actually see this hypothesis applied in the real-world, we can take a look at Kimi K2.5. It was trained on 15T tokens and the trend of using massive open-weight models as fine-tuning bases means inference infrastructure stays heavy even when training is "efficient." Bottom Line: The risks of memory being cyclical are gone as per my last post. The risk of demand is gone as per my hypothesis on this post. The only risk in my opinion is capex pullback on hyperscalers, which we have not seen yet.

You are always one decision away from a different life

@FirstSquawk Signs that rising energy prices will lead to rise in living prices across all boards. One of my first posts talk about it. The US is not immune to spiking oil prices.

AMAZON'S FULFILLMENT SERVICE TO LEVY FUEL SURCHARGE FROM APRIL 17 AMAZON SLAPS 3.5% FUEL SURCHARGE ON ONLINE SELLERS

@FirstSquawk The irony.

US Treasury calls in regulators for talks on private credit risks-FT

What does the bond market say? The 10-year Treasury yield is currently around 4.31–4.34% as of March 31, 2026. Over the past month, the yield has climbed roughly 0.32 percentage points and sits about 0.11 points higher than a year ago. The 10-year yield spiked as high as 4.45%, up half a percentage point from the 3.96% level just before the US and Israeli attack on Iran. That's a massive move in bond market terms. Notably, only about a fifth of the rise in the 10-year yield is explained by investors anticipating higher inflation. The rest is explained by the "term premium" (the extra compensation investors demand for tying up money for years amid elevated uncertainty, expected higher federal borrowing to fund the war, and technical dislocations like hedge funds being forced to sell Treasuries) Before the war, markets expected the Fed to cut rates twice in 2026. Money markets are now overwhelmingly pricing in zero rate cuts from the Fed for the rest of the year. Prior to the start of the war, financial markets were anticipating two 0.25-point cuts in 2026. Today, they foresee just one, with roughly 39% odds that it happens in September. The yield curve dynamic: Bear flattening The narrative is now shifting: Recession vs. inflation Most recently, yields have started coming back down as Wall Street fears recession more than inflation. Before the war, yields eased on expectations for Fed rate cuts. Then the war drove yields up on inflation fears. Fed Chair Powell said Monday that he sees inflation expectations as grounded despite rising energy prices, suggesting the central bank doesn't need to respond with higher interest rates.

@jukan05 😂😂😂😂

Anthropic should rename itself to RealOpenAI. By generously handing Claude Code over to its competitors, it has done something truly worthy of the name “OpenAI” — something even OpenAI itself has never managed.

@KobeissiLetter What about the third carrier group?

BREAKING: President Trump says the US will leave Iran in “2 to 3 weeks.”

It's been a succession of California governors, state legislators, and unelected regulatory bodies like CARB (largely dominated by one party for decades) who have pursued aggressive climate goals while simultaneously making it more expensive and difficult for domestic oil and refining to operate. The trade-off is that California has cleaner air standards, but everyday drivers bear the financial burden of a policy most of them never directly voted on. Bottom line: Cost of clean energy is high.

@LunarThinkTrade @ENERGY True according to your numbers. Your numbers also indicate that 90% of the oil suppliers to Calif. are foreign, so only 10% comes from our country. Is this one of the reasons that Californians pay the highest gas taxes, and price for our gas? And who’s decided this for us?

Thanks to President Trump, more than 50,000 barrels of American oil per day will flow again in California. Previously, California imported more than 60 percent of their oil from the Middle East due to their Anti-American energy policies. bloomberg.com/news/articles/…

Will it help California gas prices? The honest answer is: probably only marginally, at least in the short term. The Sable pipeline ramping up to 50,000 barrels a day would provide around 3% of the state's daily oil needs. The addition of new supply could lower costs for refineries. California's governor argued that every $10-per-barrel increase in crude translates to roughly 24 cents more per gallon, and that what would actually stabilize global oil markets is resolving the broader geopolitical situation. The legal fight: California sued the Trump administration to block the order, arguing the state invented an energy emergency to justify forcing the restart of the shuttered offshore operation. So in short, 50,000 barrels is a real and meaningful boost to local supply that could provide some modest downward pressure on refinery costs, but it's far too small to meaningfully offset the global price shock driving California's current near-$7/gallon gas prices. There are many problems with the insane energy prices of California. It is the foundation of all prices, but Californians seem ok with it.

Update: Allegedly algorithmic trading from the President of Iran's comments went through. Oil tanked 5% in 3 minutes. 1 trillion in market cap moved.

Oil just tanked in the last 2 minutes. What happened? I saw the President of Iran and Trump looking to end the war but nothing in the past couple minutes.

@BitcoinMagazine Shouldn't crypto be STAYING ahead. If it lags behind even a second, all confidence is lost.

Satoshi Nakamoto in 2010 on quantum computers: "If it happens gradually, we can still transition to something stronger."

This is mostly true. The window is closing for crypto to stay ahead of quantum, but yes it is possible. It's always easier to encrypt than decrypt though. Quantum computing specifically attacks this asymmetry using Shor's algorithm (breaks ECC, RSA) and Grover's (halves effective strength of hash functions). The principle holds again once PQC replaces current schemes.

Saw some people panicking or asking about quantum computing's impact on crypto. At a high level, all crypto has to do is to upgrade to Quantum-Resistant (Post-Quantum) Algorithms. So, no need to panic. 😂 In practice, there are some execution considerations. It's hard to organize upgrades in a decentralized world. There will likely be many debates on which algorithm(s) to use, resulting in some forks. And some dead project may not upgrade at all. Might be a good to cleanse out those projects anyway. New code may introduce other bugs or security issues in the short term. People who self custody will have to migrate their coins to new wallets. This brings to the question of Satoshi's bitcoins. If those coins move, then it means he/she is still around, which is interesting to know. If they don't move (in a certain period of time), it might be better to lock (or effectively burn) those addresses so that they don't go to the first hacker who cracks it. There is also the difficulty of identifying all his addresses, and not confuse with some old hodlers. Anyway, it's a different topic for later. Fundamentally: It's always easier to encrypt than decrypt. More computing power is always good. Crypto will stay, post quantum.

JP Morgan downgrades Global Equities and US Treasuries because of these same problems. Dual mandate is getting sandwiched.

Rising oil prices are delivering exactly this double bind to the Federal Reserve, leaving policymakers caught between a weakening economy and stubbornly elevated inflation. Back in January, markets had priced in two rate cuts for 2026, but that has since been pared down to one, edging close to zero in terms of market expectations. The economic consequences of this prolonged tightening posture are significant: > for consumers: a delayed rate cut means higher borrowing costs that remain in place longer than expected on everything from mortgages to auto loans and credit cards > further squeezing household budgets already eroded by higher energy and food prices. The Fed faces a genuine dilemma: cut rates to support a softening labor market and risk inflation moving even higher, or hold rates steady and risk further weakness in employment This is a no-win scenario that underscores just how much an oil shock can paralyze monetary policy at the worst possible moment.

@jukan05 Rumors are it is mostly TPUs and Trainium? Cannot confirm though

AI lab folks, when the hell is the Blackwell-trained model finally dropping? Doesn’t look like it’s Gemini, and people are going crazy saying Claude Mythos is performing ridiculously well. Was that trained on Blackwell?

$MU Why would Trump and Witkoff meet with Micron’s CEO? - he’s Indian - has no public political views - his company is not a 1T+ megacap Witkoff has held some amount of Micron in the past and certainly knows the company, so it can be an investment angle. It’s certainly possible that Trump is eyeing an investment in Micron. It’s the only American DRAM manufacturer and pretty strategic for national interests. whitehouse.gov/wp-content/upl…

@KobeissiLetter With Gulf states backing with full support, I do not see a de-escalation plan.

BREAKING: The US is preparing for a potential ground invasion into Iran that would last for up to 2 months, per the Washington Post. Details include: 1. Thousands of American soldiers are arriving in the Middle East for what could become a "dangerous new phase" of the war 2. Any invasion would involve raids by a mixture of Special Operations forces and conventional infantry troops, US officials said 3. President Trump has wavered between declaring that the war is winding down and threatening to amplify it 4. Discussions over the past month have touched upon the possible seizure of Kharg Island and raids into other coastal areas near the Strait of Hormuz The Iran War appears to be entering a new era.

@BRICSinfo Part 1532 of Russia and US wanting to deepen relations. Fool me once shame on stories. Fool me 1532 times, shame on me x1531.

JUST IN: 🇷🇺🇺🇸 Russia says it's ready to deepen relations with the US as quickly as Washington is willing.