Tickwind
机会行情热门新闻对话
更新登录注册
Tickwind

读懂每一次跳动,看清市场风向。

产品

  • 行情
  • 主题专区
  • 全部股票
  • 热门股票
  • 机会
  • 新闻
  • 讨论
  • 自选
  • 更新

账户

  • 设置
  • 联系我们
  • 登录
  • 注册
行情主题专区全部股票热门股票机会新闻讨论自选更新设置联系我们登录注册

© 2026 Tickwind。非投资建议。

为好奇的投资者而做。

首页/主题专区

AI 产业链专区

黄仁勋「五层蛋糕」:供电 → 芯片 → 基础设施 → 大模型 → 应用

AI 产业链,一层一层看 —— 从供电、算力芯片,到大模型与应用价值捕获。每层的关键上市公司,并标出供应链「卡脖子」环节。层级为人工策展,价格为公开数据实时(延迟)行情。非投资建议。

5. 应用4. 大模型3. 基础设施2. 芯片1. 能源

价值自下而上 —— 供电/算力是地基,模型与应用捕获更高溢价 · 点击跳到对应层

1

能源 —— 供电与散热

硬性物理地基:发电、配电设备(变压器、开关柜、UPS),以及为 AI 工厂 50–130kW 机柜散热的液冷。电力供给决定一个地区能产出多少智能。

VRT
—
Vertiv Holdings

Cleanest public pure-play on AI data-center power management + liquid cooling.

ETN
—
Eaton

Broad electrical portfolio (power distribution, UPS/PDU) + liquid cooling; data-center orders surging.

GEV
—
GE Vernova

Supplies the electrons — gas turbines + grid electrification gear for AI campuses.

PWR
—
Quanta Services

Builds + connects the grid/substation power infrastructure feeding new AI data centers.

CEG
—
Constellation Energy

Largest US merchant/nuclear generator; signs power deals directly with hyperscalers.

VST
—
Vistra

Independent power producer (gas + nuclear) selling firm capacity to data-center load growth.

TLN
—
Talen Energy

Nuclear-heavy IPP; pioneered co-located data-center-at-the-reactor power deals.

另:无美股代码 — Schneider Electric — Major power/cooling systems vendor — French-listed (Euronext SU.PA), no clean US common ticker.
2

芯片 —— 算力硅

卡脖子

卖铲子的瓶颈:GPU 与定制 AI ASIC,加上谁都绕不开的上游 —— 台积电先进制程 + 约 90% 的 CoWoS 先进封装、ASML 的 EUV 光刻垄断、供不应求的 HBM 寡头。这一层由供给(而非需求)决定加速器产量。

NVDA
—
NVIDIA

Dominant merchant AI GPU + CUDA + NVLink; the reference platform for the whole stack.

AMD
—
Advanced Micro Devices

Instinct MI GPUs + EPYC CPUs; the credible second source in training/inference accelerators.

AVGO
—
Broadcom

Co-designs custom AI ASICs (XPUs) for the hyperscalers; ~70% of the custom-accelerator design market.

MRVL
—
Marvell Technology

Other custom-ASIC house (Trainium, Maia) + data-center networking/optical silicon.

TSM
—
Taiwan Semiconductor

THE chokepoint: sole leading-edge foundry + ~90% of CoWoS packaging that gates every accelerator.

ASML
—
ASML Holding

EUV-lithography monopoly — no advanced AI logic chip exists without its machines.

MU
—
Micron Technology

Only US-listed HBM supplier; HBM is a sold-out 3-player oligopoly gating GPU memory bandwidth.

ARM
—
Arm Holdings

CPU instruction-set IP under NVIDIA Grace + most data-center/edge CPUs — a royalty toll on compute.

AMAT
—
Applied Materials

Largest wafer-fab equipment maker; arms the foundries expanding leading-edge + packaging capacity.

LRCX
—
Lam Research

Etch/deposition leader critical to HBM stacking + advanced-node and packaging buildout.

另:无美股代码 — SK Hynix — HBM market leader (~62%) — Korea-listed (KRX 000660), no liquid US common ticker. · Samsung Electronics — Third HBM supplier + foundry — Korea-listed (KRX 005930), only foreign/GDR lines.
3

基础设施 —— AI 工厂

把几万颗芯片连成一台机器:GPU 服务器/机柜集成、机柜内互联芯片、突破机柜间「铜墙」的光互联、交换网络。一个互联端口降速就能拖住整个训练任务。

SMCI
—
Super Micro Computer

Leading GPU-server / liquid-cooled rack integrator assembling silicon into deployable AI-factory racks.

DELL
—
Dell Technologies

Tier-1 AI server + storage integrator shipping full GPU rack systems to hyperscalers + enterprises.

HPE
—
Hewlett Packard Enterprise

AI servers + Cray supercomputing systems and the orchestration to run them at scale.

ANET
—
Arista Networks

Ethernet switching leader for AI back-end fabrics — the merchant alternative to InfiniBand.

ALAB
—
Astera Labs

In-rack connectivity silicon pure-play (PCIe/CXL retimers + fabric controllers) wiring GPUs to CPUs/memory.

CRDO
—
Credo Technology

Active Electrical Cables + SerDes connecting GPUs within/between racks; rack-scale connectivity pure-play.

COHR
—
Coherent

Optical transceivers + silicon photonics connecting servers across the data center (the copper wall).

LITE
—
Lumentum Holdings

Optical components / lasers (EMLs) for 1.6T transceivers; named supplier in next-gen interconnect.

CIEN
—
Ciena

Coherent optical / DWDM systems for campus-to-campus data-center interconnect; record backlog.

4

大模型 —— 基础模型

语言、生物、物理、机器人领域的基础大模型。顶层突破把需求一路向下传导。前沿领跑者多为非上市公司 —— 公开敞口主要通过其算力与云合作方获得。

GOOGL
—
Alphabet (Google DeepMind)

Owns Gemini + DeepMind + the TPU stack — the cleanest public pure-frontier-model proxy.

META
—
Meta Platforms

Open-weight Llama family + massive in-house AI compute; the public open-model proxy.

MSFT
—
Microsoft

Deep OpenAI partnership + in-house models + Azure model hosting — public exposure to frontier models.

另:无美股代码 — OpenAI — Frontier-model leader, PRIVATE — no public ticker (reference editorially via MSFT). · Anthropic — Frontier lab (Claude), PRIVATE — no public ticker.
5

应用 —— 价值捕获

价值落地层:Copilot、智能体、自动驾驶,以及企业数据上的应用 AI。估值倍数高于下层 —— 价值取决于 AI 预算份额,而非物理瓶颈。

MSFT
—
Microsoft

Copilot across M365/Dynamics + GitHub Copilot — the broadest enterprise AI-app distribution.

CRM
—
Salesforce

Einstein / Agentforce — agentic AI embedded in the dominant CRM workflow.

NOW
—
ServiceNow

Now Assist — AI agents automating IT / enterprise service workflows.

PLTR
—
Palantir

AIP operationalizes models against enterprise/government data — an applied-AI deployment platform.

SNOW
—
Snowflake

Data + AI app platform (Cortex) where enterprises build/run AI on their own data.

TSLA
—
Tesla

Physical AI — FSD autonomy + Optimus humanoid robotics (Huang’s self-driving + robots example).

DDOG
—
Datadog

Observability for AI apps — the picks-and-shovels of running AI in production.

层级为人工策展 · 价格为公开数据实时(延迟)行情 · 仅供参考 · 非投资建议

更多专区

航天专区
AI 应用与软件专区
GLP-1 减肥药专区
医疗科技与手术机器人专区
基因编辑与基因组学专区
量子计算专区
全部专区 →