MindMap Gallery 2023-22 Artificial Intelligence Industry Research
2023-22 Artificial Intelligence Industry Research, investment logic includes AI chips, CPO, large models and other knowledge, such as AIGC: simplifying the two major links of R
Edited at 2023-07-28 16:35:36This is a mind map about bacteria, and its main contents include: overview, morphology, types, structure, reproduction, distribution, application, and expansion. The summary is comprehensive and meticulous, suitable as review materials.
This is a mind map about plant asexual reproduction, and its main contents include: concept, spore reproduction, vegetative reproduction, tissue culture, and buds. The summary is comprehensive and meticulous, suitable as review materials.
This is a mind map about the reproductive development of animals, and its main contents include: insects, frogs, birds, sexual reproduction, and asexual reproduction. The summary is comprehensive and meticulous, suitable as review materials.
This is a mind map about bacteria, and its main contents include: overview, morphology, types, structure, reproduction, distribution, application, and expansion. The summary is comprehensive and meticulous, suitable as review materials.
This is a mind map about plant asexual reproduction, and its main contents include: concept, spore reproduction, vegetative reproduction, tissue culture, and buds. The summary is comprehensive and meticulous, suitable as review materials.
This is a mind map about the reproductive development of animals, and its main contents include: insects, frogs, birds, sexual reproduction, and asexual reproduction. The summary is comprehensive and meticulous, suitable as review materials.
AI
basic situation
what is
development stage
calculate
Perception
cognition
current stage
consciousness
small model
Specific scenarios, universality check, switching scenarios requires retraining
large model
Universal
Model implementation
Data is the foundation
The data element market realizes data sharing, and data feeds back to accelerate the commercialization flywheel
Data acquisition, storage, transmission and management
Computing power is the support
Talent is the key
Business value closed loop
Values, ethics, political risks
Possessing the ability to emerge
AIGC
Why watch AI
Steam engine in the digital economy era
Bringing cost reduction and efficiency improvement to the industry
gap narrows
Domestic AI technology is only 1-3 years behind the United States, with huge development potential
Reasons for the widening gap
There is a lack of original models in China, and the introduction of algorithms needs to be improved.
After August 2022, the United States will restrict the export of some high-end GPUs to China
At present, some calculation example constraints will be replaced by domestic GPUs or ASICs.
Market size and growth rate
2025
worldwide
5 trillion yuan
30%
China
1 trillion yuan
40%
What to see
Industrial chain
upstream
cloud computing
wave
Ali
Baidu
Tencent
Huawei
chip
Jing Jiawei
Baidu
Nvidia
Alibaba
Cambrian
NavInfo
SMIC
IDC
Aofei data
Zhongke Shuguang
GDS
Optical module
Zhongji InnoLight
Xinyilong
Optical Information Technology
Server liquid cooling
Inspur information
Wangsu Technology
Tsinghua Unigroup
data provider
Haitian Ruisheng
midstream
multimodal
Strategy generation
NLP
3D generation
code generation
virtual person
food generation
downstream
E-commerce
worth to buy
media
marketing
educate
humiliation
game
government affairs
C-side applications
Investment direction
Upstream chip
Computing power support
Upstream optical module
The core device of the optical communication system, among which CPO technology is the only way to achieve high speed, large bandwidth and low power consumption.
Midstream algorithm layer, large model
Downstream landing scene
Downstream applications achieve cost reduction and efficiency increase through AI empowerment
C-side companies with high-quality application scenarios and B-side companies with data accumulation are expected to be the first to benefit.
AI chips: Global annual compound growth rate of 50%. US$70 billion in 2025
Classification
training and inference
The training chip has a large computing scale, mainly GPU
Inference is to use the trained model to infer using new data, which requires less calculation.
CPU\GPU can run, so can FPGA and ASIC
Edge inference
Cloud inference
The current market mainly comes from training, but the demand for inference has begun to increase explosively.
Subcategories
Category 4: Versatility gradually decreases (flexibility), but efficiency gradually increases
CPU/GPU/FPGA/ASIC
FPGA
What is FPGA
field programmable gate array
Widely used in 5G and AI fields with programmable flexibility
Market size
Global revenue of US$12.5 billion in 25 years, with an annual compound rate of 11%
China
17% compounded annually
competitive landscape
Overseas Xilinx and Intel dominate, while domestic manufacturers emerge.
In 2021, the combined market share of Ziguang Guowei, Fudan Microelectronics and Anlu Technology will exceed 15%
Process manufacturing and gate-level scale are important indicators for evaluating product performance.
ASIC
what is
Integrated circuits customized for specific purposes, high performance and low energy consumption
Due to cost and R&D barriers, penetration in the AI industry is low. As the industry matures in the future, downstream applications will continue to optimize upstream R&D costs. ASIC chips rely on their excellent performance and low power consumption and will be the first choice for low-cost AI technology.
Market size
In 20217, the world is 170 billion yuan, with an annual compound rate of 9%
competitive landscape
There is no obvious leading manufacturer yet
Nvidia, Intel, Google
domestic
HiSilicon
Cambrian
Suiyuan Technology
CPO
What is CPO
Optical module
The cycle for planning, infrastructure, and equipment to be equipped with optical modules is 2 years. In 2024, there will be a concentrated explosion of demand for optical modules brought by AI.
CPO
Optoelectronic co-packaging technology assembles the switching chip and optical engine together in one slot
Traditional: pluggable optical module
Advantages of CPO
Low power consumption, large bandwidth and high speed
As the scale of data centers continues to expand, the demand for optical communication bandwidth and transmission rate increases, and the traditional model is not cost-effective.
market
scale
USD 5.4 billion in 2027 (global)
It will be gradually commercialized in 2024
Manufacturer
overseas
Cisco, Broadcom, and Intel have made forward-looking plans and are expected to go on sale next year.
domestic
There is no high-speed switch chip or a complete set of CPO switches. It mainly provides optical engines for overseas equipment leaders.
Tianfu Technology, Liant Technology, Bochuang Technology
Tianfu Technology is the fastest
Raised funds to build a light engine (2021)
Downstream cooperation with Cisco, etc., the product has passed verification, is undergoing reliability testing, and will be mass-produced
Liant Technology
Complete product design for high-density optoelectronic connections and build customization capabilities
There is cooperation with Cisco, but the progress is slow
Bochuang Technology
Sample delivery stage, not mass production yet
Talking about cooperation with Intel
large model
Classification
Train on large-scale unlabeled data and learn features and rules, with strong generalization ability
NLP: natural language processing
At present, large models are mainly oriented to NLP, such as GPT-3
Learn a common language so that the model has language understanding and generation capabilities
CV: Visual Common
development challenges
Valid data for training
study method
Different vision applications rely on different models
Training image sizes are getting larger and larger
It is difficult to have a major outbreak in the short term
Microsoft's swin-transformer
Google's VIT series
META’s SAM model
domestic
SenseTime
Ririxin large model
multimodal
Text-image fusion, GPT-4
Domestic large model
Overall situation
Mostly small models
Those with advanced progress include
Baidu Wenxinyiyan, Tencent Hunyuan, Huawei Pangu
Specific analysis
Tencent and Baidu have layouts in three types, and Huawei has one more large scientific computing model
ecological difference
Tencent and Baidu have C-side business
Baidu is making the fastest progress. It has open web pages for public testing, which is equivalent to the level of GPT2.5 and is more than 3 years behind.
Tencent established the project in 2019 and may announce it in the near future
Huawei has a deeper understanding of the industry, to the B-side
Competitive trends
next 2-3 years
Tencent
NLP\CV
Baidu
multimodal
Huawei
Balanced, industry deepening
Pay attention to the upstream manufacturers that cooperate with these three companies
Application areas
The first two directions of profit
C-side enterprises with high-quality application scenarios
For consumer needs: games, office, film and television, search
B-side companies with data accumulation
AI games
Application link
The four stages of game launch: project establishment - research and development - distribution - commercialization
Impossible triangle: cost-efficiency-quality
AIGC: Simplify the two major links of R&D and distribution, shorten the R&D cycle, and reduce production costs.
R&D end
text generation
Prop and character production
sound effects synthesis
scene synthesis
Issuing side
localized translation
Material creation
Community operations
Quantitative Analysis
Currently saving more than 50% of manpower and time
ROE will increase by 20% in the future
layout manufacturer
Dachang
AI technology: Tencent, NetEase, MiHoYo
AI application level: NetEase, Perfect World, Giant Network
Most models are embedded, using GPT, etc.
Mainly test related functions in side tasks
AI Others
C-terminal
Office: Kingsoft
WPS is connected to Baidu large model
B side
Bloomberg
Flush, Oriental Fortune
other
Financial tax reporting software, marketing management software that masters corporate data, service providers that master market commodity data, etc.