Last wk, S&P/Nas/Mag7 were +1.2%/+1.7%/+4.0% despite oil +4% & bond ylds +7bps on Iran flare-up. 2 days remain during this seasonally favorable time from the last 3 trading days of June through first 9 days of July. Earning season starts w/ big banks on Tuesday. As I first posted about on June 28th and reiterated on July 5th, the seasonality is historically very favorable from the last three trading days of June through the first nine trading days of July during which the S&P/Nasdaq have advanced 1.6%/2.5% and been up 78% of the time since 1985. The S&P/Nasdaq is up 3.0%/3.6% during the first 10 trading days of this time period in 2026 already. But for the heart of the AI trade, the Semiconductor (SOX) Index is down 7.0% during these last 10 trading days on fears of a “speed bump” that I have been posting about since 6/28. For the AI trade going forward, two separate thoughts with two different implications are beginning to coalesce in my mind. The first thought is that AI revenue growth for the leading LLM models is likely to hit a “speed bump” and slow in the September quarter. Last week, $SPCX and $META released new LLM models that both closed much of the gap with the leading edge models from OpenAI and Anthropic. But Meta plans to price their model at roughly 1/4th the cost of the two leaders with SpaceX aggressive as well. This is particularly relevant as companies switched from token maxing in March to token minimization in June. As an example, the CEO of Coinbase posted on June 26th, “How to keep AI spend flat while token usage grows exponentially... Putting this into practice has cut our AI spend nearly in half, while our token usage continues to grow.” Sam Altman on Thursday: “we have heard enterprises on their concerns about AI costs, and 5.6 sol is a huge step forward for dollars-per-task, as are terra and luna”. But on a positive note, two more public cloud providers have been recently added to the four that already existed of $AMZN, $GOOGL, $MSFT and $ORCL. SpaceX in the past month raised over $100B in equity plus debt and plans to spend hundreds of billions to attack the $28.5 trillion TAM they talked about in their IPO prospectus. News surrounding Meta last week seems to imply that they are launching a cloud offering not because they have excess compute but so 1) they can double their compute spend from 7 MegaWatts in 2026 to 14 MegaWatts in 2027 and 2) have capacity to sell their Muse Spark LLM. So in summary, with this seasonally favorable period of time for the overall market coming soon to a close, I believe cutting back excess exposure is prudent. Reaction by investors on Tuesday to bank earnings which should be strong, may give us a good tell as to the risk vs reward at current levels over the shorter-term. As for the AI trade, I believe selectivity is key as we work our way through the “speed bump” and the positive implications of 4 aggressive public cloud vendors now becoming six versus the negatives of token maxing in March turning to token minimization in June. All the best in the week ahead.

HERE IS HOW THE MARKET CLOSED TODAY Five of eleven sectors finished green. Technology was the worst at -1.1%, Basic Materials -1.8% and Real Estate -1.1% behind it. Chips led the selling: - Lam Research $LRCX -5.6% - Intel $INTC -5.6% - Micron $MU -4.7% - Advanced Micro Devices $AMD -3.4% - Nvidia $NVDA -2.4% Software and hardware went with them. Oracle $ORCL -5.2%, Dell $DELL -5.4%, Hewlett Packard Enterprise $HPE -6.2%, Palantir $PLTR -2.2%. Apple $AAPL +3.6% was the largest megacap gainer, the day after its iPhone event. Google $GOOGL +0.6% and Microsoft $MSFT +0.2% held up. Meta $META -1.4% and Tesla $TSLA -1.2% did not. Defensives caught the rotation. Philip Morris $PM +2.2%, AbbVie $ABBV +1.6%, Uber $UBER +2.1%, Intuitive Surgical $ISRG +2.0%. Consumer Defensive was the best sector at +0.3%.

Excited to team up with @radixark to make large-scale RL data movement faster 🚀 Miles is built for high-performance, large-scale post-training, and Mooncake is now integrated as a rollout data-transfer backend for the fragmented, heterogeneous data moving between rollout and training in disaggregated RL. On rollout data captured from Miles: ⚡ 10–14× faster remote GET ⚡ 1.2–1.6× faster PUT By turning fragmented rollout objects into efficient bulk I/O while preserving their original structure, Mooncake helps reduce rollout-to-training handoff latency without changing the RL programming model. Read more:

New top speed unlocked: Mach 1.6. Since supersonic, the team hasn’t stopped pushing to expand the envelope to higher speeds. Last week, Quarterhorse Mk 2.1 broke its own record — reaching Mach 1.6 on just its fourth flight, with under 90 minutes of total flight time logged. The fastest way to build fast aircraft is to keep flying them.

🆕 A new open-source 1.6 TRILLION-parameter MoE that scores 0.1 points behind Claude Opus 5 on an agent benchmark has been released. @NexEcosystem released Nex-N2.5-Max. It beats DeepSeek V4 Pro 0813 on all but one comparable benchmark. This thing is enormous! Stats ... 🧠 1.6 TRILLION parameters 🔀 Mixture-of-Experts 🤖 Built for agents + long-horizon workflows 💻 Coding + tool use 📚 262K context 📝 Text-only ⚖️ Apache 2.0 📦 Weights are already on Hugging Face Nex's reported results 👇 AutomationBench 🟣 Claude Opus 5 → 50.3 🔥 Nex-N2.5-Max → 50.2 🔵 GPT-5.6 Sol → 45.8 🟢 Qwen3.8-Max → 39.8 And on BrowseComp: 🔥 Nex-N2.5-Max → 92.6 🟣 Claude Opus 5 → 90.8 Terminal-Bench 2.1 → 86.1 Toolathlon Verified → 74.7 😁 You're probably NOT running this one under your desk. The official model repository is about 1.65TB, and Nex's recommended deployment uses this monster hardware .... 🎮 16× H200 🖥️ 2 nodes ⚡ TP16 + expert parallelism But here's why I'm posting the monster first. It has a 35B multimodal little brother. And THAT one may actually belong in your home AI lab. 👀🔥

a=(m,d=mag(k=2*cos(i*342),e=sin(i*271)*2)/1.6)=>point(k*(p=5+2*sin(d*8-t*3+m))+9/d*sin(k*2)+89*sin(c=d*d/9-t/8+m)+200,79*sin(c*2)+9/d*sin(e*2)+e*p+200) t=0,draw=$=>{t||createCanvas(w=400,w);background(9).stroke(w,116);for(t+=PI/60,i=1e4;i--;)a(i%4*5)}//#つぶやきProcessing#

India’s coal demand is projected to jump from about 1.2 billion tonnes today to 1.6 billion tonnes by 2030 as electricity generation and industrial activity expand. Coal still supplies roughly 60% of the country’s power. #India# #Coal# #Energy#

This map should worry everyone The replacement fertility rate is roughly 2.1 births per woman, the level needed for a population to replace itself over generations Most of the developed world is now way below that line • US: around 1.5–1.6 births per woman • UK: around 1.5–1.6 • France: around 1.5–1.6 • China: around 1.0 • South Korea: around 0.8 Meanwhile, countries like Chad and Somalia are still around 5.9 births per woman This is one of the biggest demographic divides on Earth The richest, most industrialized, most technologically advanced countries are no longer having enough children to sustain their populations naturally That means aging societies, shrinking workforces, fewer young people, more pressure on pensions, healthcare, housing, taxation, and national productivity