The Underpriced Truth: Agentic AI Is a Paradigm Shift Centered on Memory 1/ The market will slowly realize: Agentic AI is memory-centric, not compute-centric. The new hardware stack is: ① Memory — HBM / DRAM / NAND ② Parallel compute — GPU / ASIC ③ Coordinator — CPU CPUs stopped doing the heavy lifting a long time ago. This isn't a cycle. It's a paradigm. 🧵👇 2/ First principles Humanity's ultimate pursuit of intelligence has always been two things: Infinite memory + infinite compute. When we say someone is smart, we mean two things: "good memory" + "fast thinking." Machine intelligence is walking the exact same path. 3/ The story the market already understands: HBM LLM inference's decode stage is a textbook memory-bound workload. Every token generated → drag the entire KV cache across memory. Bandwidth too low → expensive GPUs sit idle. That's why every new GPU generation ships with more HBM bandwidth and capacity. 4/ The story the market is missing The "1M context" you keep hearing about? It is not assembled inside the GPU inference cluster. So where is it actually built? 5/ It's built on the traditional servers running the agentic system Those CPU + huge-DRAM servers are quietly doing the heaviest lifting: • loading user long-term & short-term memory • loading the agent's system spec / prompt • loading skill / tool / subagent definitions • compressing the context once it overflows 1M tokens All of this lives in DRAM, not HBM. 6/ Compare this to the previous era In the web / mobile era, we barely stored any user context at all. Only search / recsys / ads kept a small user profile — maybe 1/20, even 1/100 of the data volume an agentic system needs today. That asymmetry is the real overlooked inflection point. 7/ The supply chain is already telling this story Server CPU : DRAM ratio is climbing fast: • Web / Mobile era: 1 core : 4 GB • Agentic AI today: 1 core : 16 GB • Deep agentic future: 1 core : 64 GB and beyond 8/ And it's NOT just "4x more memory" Under agentic workloads, a single CPU serves a fraction of the users it used to. When the entire IT stack migrates to agentic: • CPU count grows several-fold to ~10x • DRAM total grows tens-fold to ~100x That's the part nobody is pricing in. 9/ The conclusion Agentic AI is a paradigm shift centered on storage + parallel compute. The software paradigm changed. The hardware paradigm changed with it. Only those who deeply understand the technology will see it: This isn't a memory cycle. It's a memory paradigm. 10/ Time horizon Given how early we still are on: • user adoption rate • depth of usage per user We are at least 5 years away from the cyclical top of this memory wave. (Zoom out far enough and everything is a cycle — but this one is nowhere near peak.) $MU $DRAM $SNDK

We're hiring at Gradient. Building open-source environment infrastructure for our distributed RL training stack — reproducible, scalable to thousand-GPU runs Looking for 1–2 RL Environments engineers / tech leads: You've designed verifiers, built sandboxes for agentic RL rollouts, or shipped RL training data pipelines that survived contact with real training. Domain depth in math, code, agent, tool, or GUI is a plus. PhD not required. Also hiring research interns: PhD / Masters students with hands-on RLHF / RLVR / GRPO / DPO / agentic RL experience. Open-source footprint matters more than paper count. Most intern roles convert post-grad. No age cap. Founding-team-level equity for the right people. DMs open.

Proven Political Winner Andy Barr is running for the United States Senate in the Great Commonwealth of Kentucky, a place I love and WON BIG, six times, including Primaries, in 2016, 2020, and 2024! I know Andy well, and he is always a Vote we can count on because he knows what it takes to GET THINGS DONE and, MAKE AMERICA GREAT AGAIN! I Endorsed Andy years ago, in his first Race, and all others, for Congress, and he never let me down — He is a 100% solid American Patriot! Andy Barr is a Strong Supporter of TERMINATING THE FILIBUSTER, before the Democrats do it (which they will, on Day One, if they get control of the U.S. Senate!). He will do everything in his power to get it done. It is desperately needed by the Republican Party to pass the SAVE AMERICA ACT, and all other things necessary for a strong and brilliant Country! In the Senate, Andy will fight tirelessly to Grow our Economy, Cut Taxes and Regulations, Champion the Interests of our Amazing Farmers and Ranchers, Promote MADE IN THE U.S.A., Unleash American Energy DOMINANCE, Keep our Border SECURE, Stop Migrant Crime, Strengthen our Military/Veterans, Safeguard our Elections, and Defend our always under siege Second Amendment. Andy is the only Candidate who will easily defeat the Democrat in what will be one of the most important Elections in American History. He will help ensure Victory against these Radical Left, Country Destroying, THUGS. Andy Barr has my Complete and Total Endorsement to be the next United States Senator from Kentucky – HE WILL NEVER LET YOU DOWN! ( TS: May 1 2026, 6:59 PM ET )

MssCorp (TWE: 6830) Serenity High Conviction Bet（analysis fully grounded in TSMC’s latest public expansion plans and 2026 industry data）： – this is the functional monopoly in CPO/SiPh inspection (90%+ share targeted, pricing power is real). Long list of customers ($TSM, $NVDA, $AAPL, $AMAT , $LRCX, $ASML , $INTC ) all need to go through them for yields. NVIDIA dedicated AI Chip Zone in US facility = strategic lock-in. -GS CPO TAM still the anchor: $91B by 2028. -TSMC CoWoS doubling to 130k-140k wafers per month by end-2026 via AP7 Chiayi (world’s largest advanced packaging hub, CoPoS pilot 2026/volume ramp 2027-28) + AP8 Tainan (P1/P2) + AP5/AP6 upgrades. We are still in the frontrunning window. Pure TSMC fab expansion model only — no NVIDIA exclusive assumed. Factory-by-factory HG demand: HG is specialized QA/FA tool (not standard per-line gear). Base Assumption~60 units per major packaging “module group” (20 front / 20 mid / 20 back). Using latest TSMC 2026 data: AP6 (Longtan/Taichung): Operating + upgrade → ~30 units / MssCorp 25-30 units AP7 Chiayi: World’s largest, CoPoS pilot → ~45-120 units / MssCorp 40-100 units AP8 Tainan (P1-P2): Construction/ramp → ~120 units/ MssCorp 100-110 units AP8 later phases + AP9: 2028+ planning → ~240-260 units / MssCorp 200-220 units TSMC only total: ~435-530 units / MssCorp 365-460 units Updated HG model (reflecting monopoly + pricing power): Industry total demand 130-200 units 26-30. With 90%+ monopoly → MssCorp ships 120-180 units (spares + repeat buys). ASP NT$60M (pricing power) + GM 60-75%. HG contribution build (NT$ bn, cumulative 2026-2030): Equipment sales + Recurring services/consumables (25–45% of equipment value over 5 years) + IP licensing (20–35% of equipment rev, 80-90% margin). Total HG contribution NT$9.5bn–NT$13bn. Core MA/FA business growing 20-30% CAGR from NT$22B 2025 base on top. Revenue path (NT$ bn): 2026: 30–38 2027: 45–62 2028: 65–88 2029-30+: 180–240/yr (normalized) Mix shifts hard to high-margin equipment + recurring + IP as CPO goes volume Share count 51.78M. With the structural monopoly in a critical yield choke-point + TSMC/NVIDIA tailwinds, long-term normalized forward EPS can realistically reach NT$180–240 (US$5.6–7.5). 40–60x forward P/E (standard for AI/SiPh leaders with real moats) → target NT$5,000+. Current MC ~$1.2–1.4B.

Count of CLARITY Act amendments in Senate Banking filed by senator: Warren 44 Reed 18 Van Hollen 8 Hagerty 5 Lummis 5 Kim 4 Britt 3 Cortez Masto 3 McCormick 3 Smith 3 Scott 2 Warnock 2 Alsobrooks 1 Cramer 1 Gallego 1 Rounds 1 Warner 1

The non-English tax is real. Sutton's Bitter Lesson, translated across languages and normalized to OpenAI English token count: Hindi: OpenAI 1.37×, Anthropic 3.24× Arabic: OpenAI 1.31×, Anthropic 2.86× Chinese: OpenAI 1.15×, Anthropic 1.71× Claude’s tokenizer charges a much higher linguistic tax.