Why is the next main focus the NVIDIA GTC Conference in March? A tenfold increase in three years all because of GTC!

Why is the next core theme the NVIDIA GTC Conference in March? [Taogu Ba]
And this was already reminded last week, with further reminders yesterday. This is the main opportunity for this week!
Today’s surge is mainly due to: 1. Easing Middle East tensions. 2. Global tech rebound!
So we anticipated this in advance, and at the end of February, we also used the same approach: integrated computing and electricity! It continued to explode last week!

1. Why is the NVIDIA GTC Conference in March the next core theme?

The NVIDIA GTC (GPU Technology Conference) has become the “core mainline” and “weather vane” of AI computing market mainly based on three key reasons:

1. Industry turning point: shifting from “scale of computing power” to “efficiency + integration”
   In the past two years: AI computing was dominated by GPU training, focusing on total computing power, HBM capacity, rack scale.
   By 2026: inference demand will explode (OpenClaw, AI Agent, real-time interaction), with the core issue shifting from “whether there is enough computing power” to “whether the speed is fast enough, energy consumption low enough, and cost efficient.”
   GTC 2026 positioning: NVIDIA will officially release inference-specific chip LPU + next-generation general architecture Feynman, marking a new stage of efficiency revolution in AI computing.
2. Technological catalysis: two major black technologies + three hardware upgrades, resonating across the entire industry chain
   LPU (dedicated inference): solves high latency and high energy consumption issues of GPU inference, directly driving SRAM, high multi-layer PCB, liquid cooling, optical interconnects.
   Feynman (next-gen general): dual breakthroughs in computing power and heat dissipation, promoting MLCP liquid cooling, HBM4, 78-layer PCB, CPO mass production.
   Supporting upgrades: RubinUltra, NVL576, CPO, liquid cooling fully implemented, creating global resonance with Broadcom, Marvell ASIC demands.

3. Time window: March is the “starting gun” for the year-round AI hardware market
   Historical pattern: GTC (March) → new product launches → order fulfillment → performance realization → mainline for the year.
   2026 special point: simultaneous release of LPU + Feynman, combined with OpenClaw inference demand, driven by both technology and demand.
   Capital expectations: institutions have already pre-positioned in liquid cooling, PCB, CPO, with March conference as the node for expectations and market acceleration.

2. Which major stocks stood out in the GTC conferences over the past two years?

1. GTC 2024 (March): Blackwell + GR00T robots ignite three major lines
   Key releases: Blackwell architecture (B100), GR00T robot platform, NVL72 copper interconnect.
   Major stocks (A-shares): Liquid cooling: InnoVek, Shenling Environment, SiQuan New Materials (50%-100% rise from March to June). PCB / copper-clad boards: Shenghong Technology, Hudian Co., Shengyi Technology (explosive demand for high multi-layer PCBs). Optical modules / CPO: Tianfu Communication, Zhongji Xuchuang, Xinyi Sheng (NVL72 copper cables + optical modules logic). Robots: Green Harmonics, Sanhua Intelligent Control, Mingzhi Electric (catalyzed by GR00T).
   US stocks: NVIDIA (up over 150% from March to October 2024), Broadcom, Marvell (ASIC demand explosion).
2. GTC 2025 (October): Cosmos + Blackwell orders strengthen the hardware mainline
   Key releases: Cosmos model, Blackwell/Rubin $500 billion orders, NVQLink quantum interconnect.
   Major stocks (A-shares): Liquid cooling: JieBang Technology (MLCP), InnoVek (standard cold plates), Gaolan Co. (secondary surge). PCB materials: FeliHua (M9Q fabric), LongYang Electronics (PTFE copper-clad), Shenghong Technology (78-layer PCB). CPO / optical interconnects: Changguang Huaxin (laser sources), Zhishang Technology (optical engines), Taichen Optics (fiber). Storage: GigaDevice (SRAM), Lankuo Technology (HBM).
   US stocks: NVIDIA (market cap over $5 trillion after October GTC), Lumentum, Coherent (NVIDIA’s $2 billion investment).
3. Summary of historical patterns (key!)
   Mainline remains unchanged: upstream of computing hardware (cooling, PCB, optical modules / CPO, storage) is the eternal mainline of GTC.
   Logic remains unchanged: NVIDIA new products → upstream material/equipment demand explosion → performance realization → stock price surge.
   2026 upgrade: shifting from “GPU training” to “LPU inference + Feynman general,” with liquid cooling (MLCP), high multi-layer PCB (M9/PTFE), CPO as the biggest growth directions.

3. Two core black technologies: technical breakdown + core advantages + implementation pace

（一）LPU (LPX): a new benchmark for inference-specific computing power, filling NVIDIA’s inference shortcoming

1. Core background
   Positioning: benchmarked against ASIC, specially designed for AI inference, solving GPU inference issues like high latency, low energy efficiency, high cost
   Technology origin: NVIDIA obtained non-exclusive license from Groq in December 2025, absorbed 90% of its core staff, integrating low-latency inference tech
   Implementation pace: officially launched at GTC 2026, mass production in late 2026, targeting rack-scale cluster solutions
2. Core technological breakthroughs (compared to traditional GPUs)
   Memory architecture: built-in 230MB on-chip SRAM, memory bandwidth up to 80TB/s, no need for frequent external memory calls, greatly reducing latency
   Cluster integration: 256 LPUs per rack, using Realscale chip direct connection tech, distributed inference across thousands of cards, each responsible for local model computation and aggregation
   

Cost and energy: energy efficiency improved by over 30%, inference cost per unit reduced by 25%, capable of replacing some GPU inference scenarios

3. Cost breakdown of a single LPU (total $600, with 65% external purchase)

（二）Feynman architecture: next-generation general-purpose AI chip, breaking through compute-power and power consumption limits

1. Core positioning
   Beyond Rubin series, targeting large-scale compute scenarios like physical simulation and synthetic data
   Core dilemma: massive increase in compute density but power consumption exceeding 2000W, forcing comprehensive upgrades in cooling, power supply, and storage
   Implementation pace: technology roadmap disclosed at GTC 2026, mass production in 2027, with Rubin Ultra rack support
2. Core technological breakthroughs
   Compute upgrade: advanced process technology, over 50% increase in single-chip compute power over Rubin
   Cooling revolution: power exceeds 2000W, replacing traditional cold plates, upgrading to MLCP microchannel water cooling plates, improving cooling efficiency by over 50%
   Power supply upgrade: supporting 800V HVDC high-voltage DC solutions, modular vertical power delivery, improving efficiency
   Storage upgrade: fully equipped with HBM4, replacing HBM3e, meeting ultra-high bandwidth needs
   （三）Supporting core upgrades: Rubin racks + CPO mass production
   Rubin four racks: NVL72 / NVL144 / NVL576 / Vera CPU, with NVL576 as flagship
   Highlight of NVL576: PTFE + M9Q fabric composite substrate, CPO optical interconnect, marking large-scale commercial use of CPO
   Performance boost: VR200 NVL72 with HBM4, inference/training performance increased by 5x/3.5x respectively

4. Market incremental estimates (2026-2027, in billion USD)

（一）LPU rack market (2026, 100,000 units shipped, $380,000 per rack)

（二）Feynman rack market (2027, 20,000 units shipped, $800,000 per rack)

（三）Summary of core field growth over two years
Liquid cooling: total $12.4 billion (LPU $7 billion + Feynman $4.8 billion), MLCP expected to surpass $15 billion by 2028

PCB and materials: total $6.98 billion, with M9Q fabric + PTFE materials adding $3 billion growth

CPO field: $6 billion in 2027, penetration rising from 5% to 8%

**
5. Related companies:**

Summary: GTC 2026 is not just about new product launches but also a turning point for AI computing business models.
In the past: winning by buying the shovel (GPU).
In the future: winning by buying “cooling materials,” “high-frequency substrates,” and “dedicated interconnects.”
Core conclusion: focus on the two most promising and highest barrier growth areas—MLCP liquid cooling and PTFE/M9Q materials—and target key stocks for the AI hardware efficiency revolution in 2026, playing the Davy double.
If you don’t understand, please like + share + comment: Why is the March NVIDIA GTC Conference the next core theme? Tenfold in three years just because of GTC!

The above is just my personal trading review and reflection. Investment involves risks; trade cautiously! Plans are always outpaced by market changes. All content reflects personal thoughts and records, serving as a record of my understanding of the market, for personal sharing only, not investment advice. Trade at your own risk!
(Research and compilation are not easy; your likes, shares, and comments are our motivation. Thank you!)

Overall market: Today’s market rebounded as expected, with 4,302 stocks up and only 790 down, but volume shrank to 2.39 trillion, below 2.5 trillion. No breakthrough of the 4130 resistance; only a volume breakout tomorrow could lead to a better surge. Otherwise, the overall trend may continue to fluctuate, with a breakout to the upside inevitable! Please be aware of risks! The key is to watch for a breakout within the range! The overall view from the 9th to the 16th remains bullish!

Sentiment: Still at a low point, 55 limit-ups, 4 limit-downs, 71% limit-up rate, 5 stocks with 5 consecutive limit-ups, total 6 stocks with continuous boards.
Sector focus: currently no dominant theme, but strongest sectors are: computing power engineering + AI applications + optical communications! Continuous rotation, affected by ongoing conflicts and sustainability!
Assist: Commercial aerospace + robotics + electricity.
Optical communications: leader is HuiLv Ecology, with 2 limit-ups in 4 days, followed by RuiSi and QiangHuo; core is ChangFeng Optical Fiber.
Computing power engineering: leader is XingXing Toys with 2 limit-ups, followed by ZhongDian XinLong and others, but overall sustainability is uncertain, as rotation is heavily quantized!
AI applications: leader is WangLi Security with 5 limit-ups, followed by GuoAn and DeCai!
Success has no shortcuts—only discipline and persistence! Wishing everyone on the path of relentless effort, the more you work, the luckier you get!
Today’s highlights:

1. Feynman: GTC is coming soon; whether the disruptive Feynman architecture will be released is a market focus.

2. OpenClaw: igniting the tech scene, with giants like Google, Tencent, Xiaomi rushing to market, pushing it from “geek toy” to “mass tool.”

3. PCB: MGC announced that from April 1, the prices of copper-clad boards, prepregs, and backglues will be uniformly increased by 30%.

4. Optical communication: AOI announced receiving $200 million in the first batch of 1.6T mass production orders, marking the start of large-scale commercial deployment of next-gen high-speed optical interconnects.

5. Saudi Arabia, UAE, Iraq, and Kuwait reportedly cut oil production by up to 6.7 million barrels/day.

6. Live pigs: 10.52 yuan (-0.85%), pig costs 12, Muyuan 11.6 yuan, feed costs over 50%, labor 10-15%.

7. Battery lithium carbonate: 155,000 yuan (1.318%), salt lake costs 30,000-40,000, mica 60,000-90,000, lithium hydroxide 60,000-80,000.

8. Praseodymium-neodymium oxide: 847,500 yuan, NdFeB N35: 1,795,000 (0%), light cost 350,000-400,000. MP79, Australia 450,000-500,000. Mining: 8,000–10,000 yuan/ton, heavy rare earths 1.3-1.5 million yuan/ton.
   Like and watch, aiming for a million monthly income! Thank you all for your support!