@ChuteJerred @forallcurious IC 1101 (~98B solar masses) is the largest known. At Alpha Centauri distance, its event horizon would look noticeably bigger than TON 618's in this view.

▶ Ajinomoto Raises ABF Substrate Film Prices by 30% - Japan’s Ajinomoto has decided to raise prices for its core ABF build-up film by 30%, with the new prices set to take effect from Q3 2026. - Taiwanese package substrate makers said they have officially received notice of the price increase, at a time when cost pressures remain elevated across the IC substrate supply chain. - Specific new pricing is still under negotiation with suppliers, but in the near term, the hike is expected to increase production costs before being gradually passed through to product ASPs. - Despite the ABF price increase, the largest cost burden is still understood to come from repeated price hikes in upstream raw materials, particularly CCL. - With urgent demand from AI chip customers continuing, quarterly price increases for ABF and BT substrates are expected to persist through year-end. The magnitude of price hikes could widen further in the second half. - Ajinomoto holds more than 95% share of the global ABF market, giving it strong pricing power across the supply chain. IC substrate makers noted that the previous price increase took place around early 2025, roughly a year ago, and said the latest 30% increase is also a “reasonable level” given strong customer demand. - Unlike Nittobo, whose conservative capacity expansion stance has deepened the shortage of T-Glass fiberglass for IC substrates, Ajinomoto had already anticipated changes in ABF supply-demand dynamics three years ago and moved ahead with proactive capacity expansion. - As a result, despite its near-monopolistic market position, the supply bottleneck is viewed as relatively limited. Ajinomoto recently announced plans to build a third ABF plant in Gifu Prefecture, Japan, in order to address demand beyond 2030. - The company has secured land for the project with an investment of around JPY 12 billion, or approximately USD 76 million, and plans to begin construction in 2028, with mass production targeted for 2032. The new Gifu plant is expected to be much larger in scale than Ajinomoto’s existing production sites in Kanagawa and Gunma prefectures. - In addition, as AI chip packaging layer counts are expected to expand from the current 3+3 level to 11+11 layers, and potentially to 13+13 layers after 2030, ABF demand is likely to sustain structural growth. - According to the industry, as the AI CPU, GPU, and ASIC upgrade cycle accelerates, demand is rising for larger substrate area and higher layer counts. As a result, ABF substrates are understood to have re-entered a supply shortage phase from 1H 2026. - Shortages are also emerging simultaneously across the upstream value chain, including fiberglass, copper foil, and drill bits. Therefore, the ABF supply-demand imbalance in 2027–2028 is more likely to intensify than ease. - The industry expects the IC substrate sector to enter a so-called “Super Expansion Cycle” over the next two to three years, which should improve order visibility across the sector.

$HIMX Q1’26 EARNINGS HIGHLIGHTS 🔹 Revenue: $199.0M (Est. $195M) 🟢 🔹 EPS Per Diluted ADS: $0.046 (Est. $0.03) 🟢 🔹 Gross Margin: 30.4%, at high end of guide (Est. 30%) 🟢 Q2 2026 Guide: 🔹 Revenue: +10.0% to +13.0% QoQ (Est 5%) 🟢 🔹 Gross Margin: Around 32% (Est. 30.8%) 🟢 🔹 EPS Per Diluted ADS: $0.086-$0.103 Segment Performance: 🔹 Large Display Driver Revenue: $24.2M; +11.7% QoQ 🔹 Large Display Driver Revenue Mix: 12.2% of total sales 🔹 Small & Medium Display Driver Revenue: $135.8M; -2.4% QoQ 🔹 Small & Medium Display Driver Revenue Mix: 68.2% of total sales 🔹 Non-Driver Revenue: $39.0M; -7.7% QoQ 🔹 Non-Driver Revenue Mix: 19.6% of total sales Other Metrics: 🔹 Automotive Driver Sales: Declined double digits QoQ in Q1 🔹 Smartphone IC Sales: Increased QoQ, driven by new OLED solutions entering mass production for a leading smartphone brand’s mainstream model 🔹 Tablet IC Sales: Increased QoQ, driven by renewed mainstream demand and shipments for a new premium OLED tablet 🔹 Automotive Tcon: Hundreds of secured design wins across a broad customer base 🔹 WiseEye: Adopted by a leading brand for smart glasses, with mass production expected later this year 🔹 CPO Gen 1: Small quantity shipments expected in 2H26 🔹 CPO Gen 2: Nearing completion of customer product validation for AI data center applications 🔹 FOCI Stake: 5.36%, valued at NT$4.96B / $156M as of May 7 close 🔹 Patents: 2,564 granted, 331 pending as of March 31, 2026 Financials: 🔹 Operating Profit: $10.2M 🔹 After-Tax Profit: $8.0M 🔹 Operating Expenses: $50.3M; -8.4% QoQ, +9.9% YoY 🔹 Operating Margin: 5.1% 🔹 Cash, Cash Equivalents & Other Financial Assets: $287.6M 🔹 Long-Term Unsecured Loans: $27.0M, including $6.0M current portion 🔹 Inventory: $151.7M 🔹 Accounts Receivable: $190.9M 🔹 DSO: 86 days 🔹 CapEx: $2.9M Capital Return: 🔹 Annual Cash Dividend: $0.252 per ADS 🔹 Total Dividend Payout: $44M 🔹 Dividend Payout Ratio: 100% of previous year’s profit 🔹 Dividend Payable Date: July 10, 2026 Commentary: 🔸 “We expect upward momentum through the remainder of 2026, supported by a meaningful number of new automotive projects scheduled to enter mass production in the second half.” 🔸 “The positive outlook is also supported by the anticipated growth in our non-driver IC businesses, particularly Tcon and WiseEye AI.” 🔸 “Despite ongoing macro uncertainty, Himax continues to expand beyond its traditional display IC business, focusing on key growth areas including smart glasses, ultralow power AI and CPO.” 🔸 “These emerging technologies present significant growth opportunities that help diversify our revenue base into areas with attractive gross margin profiles and profitability while also strengthening our overall competitiveness.”

What if Your Neural Network Was Forced to Obey Physics? Physics-Informed Neural Networks (PINNs) are neural networks trained to satisfy a differential equation by building the PDE residual directly into the loss. They emerged from a very practical problem...classical PDE pipelines can be brilliant, but they often demand heavy discretization work (meshes, stencils, stability tuning), and the method you build is usually tied to one geometry and one solver setup. A PINN flips the workflow by representing the solution itself as a smooth function uᵩ(x,t) and enforcing the physics everywhere you choose to sample the domain. People often meet PINNs in the least helpful way...via a flashy solution plot, and almost no explanation of what was enforced to get it. In this series we keep the enforcement visible. We pick a differential equation, represent the unknown solution as a flexible function, measure how well that function satisfies the equation across the domain, and train it to reduce that mismatch everywhere we sample. A normal neural net learns from labels...you give it inputs and target outputs. A PINN learns from a differential equation...you give it inputs (x,t) and it gets punished whenever its output fails the PDE. By punish we mean that the loss increases when the mismatch is large we reward it if the loss decreases as the mismatch gets smaller. The network isn’t replacing physics, it’s becoming a flexible function that is forced to satisfy the same calculus you’d impose on any candidate solution. The math breakdown: We start with a PDE we want to solve on a domain Ω. Write it as uₜ(x,t) + N(u(x,t), uₓ(x,t), uₓₓ(x,t), …) = 0 for (x,t) in Ω A PINN replaces the unknown function u with a neural network output uᵩ(x,t) Now define the physics residual by plugging uᵩ into the PDE rᵩ(x,t) = ∂uᵩ/∂t + N(uᵩ, ∂uᵩ/∂x, ∂²uᵩ/∂x², …) If uᵩ were an exact solution, we would have rᵩ(x,t) = 0 everywhere. We may also have data points (xᵢ,tᵢ,uᵢ) from measurements or a known initial condition. The training objective is just a weighted sum of squared errors L(ᵩ) = L_data(ᵩ) + λ L_phys(ᵩ) + L_bc/ic(ᵩ) with L_data(ᵩ) = meanᵢ |uᵩ(xᵢ,tᵢ) − uᵢ|² L_phys(ᵩ) = meanⱼ |rᵩ(xⱼ,tⱼ)|² where (xⱼ,tⱼ) are the collocation points in Ω L_bc/ic(ᵩ) = penalties enforcing boundary conditions and initial conditions The key technical step is that the derivatives inside rᵩ are computed by automatic differentiation ∂uᵩ/∂t, ∂uᵩ/∂x, ∂²uᵩ/∂x², … So we can differentiate the total loss L(ᵩ) with respect to ᵩ and train with gradient descent. This is the whole idea behind PINNs. Learn a function, but make the PDE part of the loss, so the network is trained to be a solution, not just a curve-fitter. In the render, the main 3D surface is the network’s current guess uᵩ(x,t), drawn as a living sheet over the (x,t) plane. Hovering above is the neural scaffold...a visible graph of feature nodes and connections. The bright tension threads are the physics residual rᵩ(x,t): each thread tethers a collocation bead on the sheet up to the scaffold, and it thickens and brightens exactly where |rᵩ| is large (color encodes the sign). As training runs, those threads go slack across the domain not because we hid the error, but because the network has actually been pushed toward rᵩ(x,t) ≈ 0. #PINNs# #PhysicsInformedNeuralNetworks# #ScientificMachineLearning# #PDE# #DifferentialEquations# #Optimization# #MachineLearning# #AppliedMath# #ComputationalPhysics#

.@BillyStrings earns his first career entry on the #Hot100# this week, thanks to his featured appearance on @PostMalone's "M-E-X-I-C-O." The bluegrass star has already earned four No. 1s on Billboard's #BluegrassAlbums# chart, including the 25-week leader 'Home' in 2019. Details: