According to Business Insider’s latest investor roundup in the field of Robotics and Physical AI, one company stands out as the clear consensus bet: Physical Intelligence (PI) — backed by ~7-8 of the top 22 investors profiled.
From foundation model believers at Khosla, Lux, Founders Fund, Emerson Collective, Greenoaks, Emergence, and more — everyone’s piling into general-purpose “brains” for robots.
I’ve been reading about CoWoS vs CoPoS, triggering a fascinating debate these days on round wafers vs. square wafers (or panels) in advanced semiconductor packaging.
My understanding and takeaway below:
CoWoS (Chip-on-Wafer-on-Substrate) is TSMC’s mature 2.5D technology that powers today’s top AI GPUs (NVIDIA H100/B200, AMD MI300, etc.). It stacks logic dies and HBM on large silicon interposers made from traditional round 300mm wafers.
CoPoS (Chip-on-Panel-on-Substrate) is the next-generation evolution, moving to large square/rectangular panels (starting ~310×310 mm, scaling to 600×600 mm+), often with glass or hybrid substrates.
Key benefits of CoPoS
• Massive supply and capacity gains — this is the biggest driver. CoWoS has been a serious bottleneck for AI chip production due to limited wafer capacity and surging demand. CoPoS dramatically increases throughput with far higher area utilization (80%+ vs. 45–60% on round wafers), enabling 2–9× more interposers per carrier. This directly expands production volume and helps resolve the packaging crunch for future AI accelerators.
• Strong cost advantages (20–30% lower unit costs) from reduced material waste and better economies of scale.
• Competitive or enabling higher performance for ultra-large multi-die packages with more HBM stacks and greater overall bandwidth.
Why have wafers historically been round?
Silicon wafers are round because of the Czochralski process: ultra-pure silicon is melted and a seed crystal is pulled upward to grow a cylindrical ingot. This yields the highest-quality monocrystalline silicon with minimal defects. Round shapes are also easier to handle, rotate uniformly during processing, and fit the industry’s entire equipment ecosystem.
Challenges of switching to square panels
• Growing or cutting uniform square silicon is inefficient and wasteful.
• New issues with warpage, corner defects, and thermal uniformity on large panels.
• Massive retooling of handling systems and processes after decades optimized for circles.
In short: Round wafers won for 60+ years thanks to physics and manufacturing maturity. But for the AI era, CoPoS’s supply scalability (addressing capacity shortages) makes the shift to squares critical. TSMC plans pilots in 2026 and mass production ~2028–2029 — a quiet but essential evolution for sustaining AI growth.
Will “square” become the new normal for advanced packaging?
The future of autonomous driving is shifting from rigid rule-based systems to powerful end-to-end AI models.
This McKinsey diagram perfectly captures the evolution:
• AV 1.0 (Rule-based/Hybrid): Stacked modular pipelines with separate perception, planning, and control layers — reliable but brittle in complex “long-tail” scenarios.
• AV 2.0 (End-to-End): A unified AI system trained on massive data, directly mapping sensor inputs (cameras, LiDAR, radar) to control outputs. Far more scalable and adaptable.
Fully agree with the direction here and requirement for growing compute in cars. Generative AI and internet-scale training data are unlocking the next leap in autonomy — moving us beyond hand-crafted rules toward systems that truly learn to drive like humans.
3
497
Circular Financing problem in AI is there, see the image below from IDC.
Revenue growth in enterprise Applications due to AI is a pragmatic metric to judge the real impact of AI
Explosive AI infrastructure growth (GPUs, cloud, data centers) is partially clouded by circular financing — the same players (NVIDIA, Microsoft, OpenAI, etc.) investing in and selling to each other. Real revenue? Yes. Clean, arm’s-length demand? Harder to tell.
The clearer signal will come from the application layer. When enterprises renew or expand AI-powered ERP, CRM, or HCM tools based on measurable ROI — cost savings, efficiency, revenue impact — that’s genuine adoption. No circular loops there.
Just like the cloud boom, infrastructure leads… but applications capture the lasting value.
Watch the apps. That’s where the AI story gets real.
India’s stock market, after its recent stall amid Middle East tensions, is poised for a strong recovery.
Goldman Sachs notes that Indian equities have historically rebounded well post-geopolitical stress, turning crises into catalysts for reforms and long-term outperformance—making this an attractive entry point for investors.
Came across this CLAUDE.md honesty ruleset that forces Claude Code to stop lying and own up to its mistakes in every session—will give this a try.
6
546
Space is hard, as the latest Blue Origin mishap reminds us. Looking at the comparison below, it is obvious why SpaceX is lightyears ahead and commanding a massive valuation despite challenging financials. 🚀
SpaceX’s $SPCX 3 Pillars: With a potential $1.75T–$2T valuation ahead of IPO, which pillar is really justifying that price?
1. Connectivity (Starlink): Satellite internet bringing in ~$11.4B (61% of revenue). This is the profitable engine with strong growth and solid margins.
2. Space (Launches & Contracts): Rocket launches, NASA, and government work (~$4.1B). Reliable but slower growing and more R&D-heavy.
3. AI (xAI integration): Newer bet on compute, models, and infrastructure (~$3.2B revenue) — but burning cash fast on massive data centers and training.
Financial snapshot (2025): $18.7B revenue (+33% YoY), but $4.9B net loss due to heavy Starship + AI investments.
Growth mode is expensive — but the bet is clearly on the future.
10
1,398
$STM as well….
Been reading a bit about India Datacenter space and it feels like a genuine multi-year growth story for India’s digital + infra economy.
Big tech (Google, Microsoft, AWS) and Indian giants like Reliance, Adani, and Airtel (Nxtra) are pouring billions into massive new campuses—especially in Mumbai, Vizag, Chennai, and more.
The cool part? Local power, infrastructure, and industrial companies are set to win big. Think suppliers of transformers, cooling systems, steel, EPC contractors, and renewable energy players—they’re all getting a massive boost from this capex wave. Stocks in the broader infra and power ecosystem have already started reacting positively to these announcements.
Current capacity sits around 1.5 GW (end of 2025), with another ~200+ MW expected to come online in 2026 — pushing it to roughly 1.7 GW by year-end. Longer term, analysts see it growing 4x–5x to 6–8 GW+ by 2030.
Will be diving more. Looking forward to your thoughts.

















