Serenity's AI Thesis: Photonics, Memory & Nebius Set for Re-Rating in New Infrastructure Cycle

Serenity's AI Thesis: Photonics, Memory & Nebius Set for Re-Rating in New Infrastructure Cycle

2026/06/20 11:11:00
Custom ImageSerenity’s latest AI thesis signals a major shift in the AI market. Investors are no longer looking only at chatbots, software platforms, GPU leaders, and large language models. The next focus is moving toward the infrastructure that allows AI to scale: AI-native cloud capacity, high-speed optical networking, high-bandwidth memory, data center power, storage, and cooling. Serenity argues that three themes are now leading this new cycle: neoclouds, photonics, and memory. Nebius stands out in the neocloud story because it is building AI cloud capacity for training, inference, and production workloads. Applied Optoelectronics, or AAOI, is gaining attention in photonics as AI clusters require faster optical networking. Micron, SK Hynix, and Samsung are central to the memory theme because HBM is becoming essential for advanced AI accelerators. The bigger message is clear: the AI trade is becoming more selective, and the next re-rating may favor companies that control the real bottlenecks behind AI growth.
 

Why Serenity’s AI Thesis Signals a New Infrastructure Re-Rating Cycle

Serenity’s thesis matters because it explains how the artificial intelligence market is moving from broad excitement into a more mature infrastructure-driven phase. The early AI rally was supported by visible breakthroughs in generative AI, enterprise automation, coding assistants, and productivity tools. Those products made AI easy to understand for investors, but they also created a more important question: what infrastructure is required to support AI at global scale? The answer goes far beyond software. AI needs compute capacity, high-speed networking, memory bandwidth, storage, electricity, cooling, data center construction, and specialized hardware supply chains. This is why the market is beginning to look deeper into the companies that support AI behind the scenes.
 

1. AI Investing Is Moving From Software Narratives to Infrastructure Bottlenecks

The first stage of AI investing was dominated by software stories because applications were the most visible part of the trend. Investors could easily understand AI chatbots, copilots, coding tools, and enterprise automation platforms. However, as usage expands, the market is realizing that AI software cannot grow without major infrastructure investment. Large models require expensive training clusters, while production AI systems require continuous inference capacity. This makes infrastructure a long-term demand driver rather than a temporary support layer. The market is no longer only asking which company has the most impressive AI product. It is also asking which companies own the infrastructure that allows those products to run at scale.
 
This shift also fits the broader AI and crypto infrastructure landscape, where investors are paying closer attention to compute, automation, data networks, and physical infrastructure. The important change in Serenity’s thesis is that investors are beginning to value bottlenecks. If AI demand keeps growing, companies that provide scarce infrastructure may gain stronger pricing power and higher strategic value. This includes companies that supply cloud capacity, optical networking, high-bandwidth memory, data center power access, and specialized systems.
 
Key AI infrastructure layers now gaining attention include:
  • AI cloud capacity for model training, inference, and enterprise deployment
  • Neocloud platforms designed specifically for high-performance AI workloads
  • Photonics and optical networking for faster data movement inside AI data centers
  • High-bandwidth memory, or HBM, for GPUs and AI accelerators
  • Power, cooling, and storage systems needed to support large AI clusters
  • Specialized chips, servers, and interconnects that improve AI system performance
 
This is why Serenity’s thesis signals a re-rating cycle. Infrastructure companies that were once treated as secondary suppliers may now be valued as core AI beneficiaries if they control important parts of the AI stack.
 

2. The Three Themes Behind Serenity’s AI Infrastructure Thesis

Serenity’s framework is built around three connected themes: neoclouds, photonics, and memory. Neoclouds represent the compute layer because AI developers and enterprises need access to specialized cloud infrastructure. Photonics represents the networking layer because AI clusters need faster data transfer between GPUs, servers, and storage systems. Memory represents the performance layer because AI accelerators require high memory bandwidth and capacity to process large workloads efficiently. These themes are important because they describe the full AI infrastructure chain rather than only one part of it.
 
A model cannot run without compute, compute cannot scale without efficient networking, and advanced accelerators cannot deliver full performance without HBM. Data centers also cannot expand without power and cooling. This creates a connected investment story where each layer supports the next. Instead of treating AI as a single software trend, Serenity’s thesis treats AI as a physical infrastructure buildout similar to earlier technology cycles, where long-term winners were often found in platforms, suppliers, and bottleneck assets beneath the application layer.
 

3. Why Stock Selection Is Becoming More Important in the AI Trade

A key point in Serenity’s view is that the AI trade is becoming more selective. In the early stage of a major technology theme, many related stocks can rise together because investors are buying the broad narrative. Over time, the market usually becomes more disciplined. Companies with real orders, strong margins, customer demand, and supply-chain advantages continue to attract attention, while weaker names may lag even if they are connected to the same theme.
 
This is why Serenity’s mention of IREN is important. Some AI infrastructure names may underperform if they face fundraising pressure, dilution risk, heavy selling pressure, or weak execution visibility. The market may like the broader theme but still punish companies that need too much capital or lack clear demand support. This means the next stage of AI investing may not be about buying every company with an AI label. It may be about identifying companies with strong positions in real infrastructure bottlenecks.
 
The most important selection factors include:
  • Strong demand visibility from major customers
  • Real exposure to AI infrastructure bottlenecks
  • Product leadership in compute, memory, networking, or data centers
  • Clear revenue growth supported by orders or long-term agreements
  • Disciplined financing and manageable dilution risk
  • Ability to scale capacity without damaging margins
 
This selective approach is central to Serenity’s thesis because the market is moving from broad AI enthusiasm toward more focused thematic exposure.
 

Nebius, Neoclouds and Photonics: The Next AI Data Center Growth Story

Nebius, neoclouds, and photonics are important because they sit inside the next phase of AI data center expansion. As AI models grow larger and inference demand increases, companies need more than GPUs. They need cloud platforms that can provide reliable capacity, and they need networking systems that can move data quickly across large clusters. This is where neoclouds and photonics become connected. Neoclouds provide AI-ready compute, while photonics supports the data movement required to make that compute efficient. Together, they form one of the clearest examples of how the AI market is shifting from software applications toward infrastructure ownership.
 

1. Why Neoclouds Are Becoming Core AI Compute Infrastructure

Neoclouds are cloud infrastructure companies built specifically for AI workloads. Traditional cloud platforms were created for general computing, storage, web services, and enterprise software, but AI workloads demand a more specialized environment. Training large models and running high-volume inference require dense GPU clusters, fast interconnects, advanced cooling, high utilization, and infrastructure software designed for machine learning operations. This is why neoclouds are gaining attention as a new category within the cloud market.
 
The rise of neoclouds is also linked to scarcity. AI compute capacity is expensive and difficult to build quickly because it depends on chip supply, power access, data center construction, cooling systems, and technical expertise. When demand exceeds available capacity, customers may be willing to sign longer-term agreements to secure access. This makes AI cloud capacity a strategic asset rather than a simple commodity service. For investors, the neocloud theme provides a way to gain exposure to the infrastructure demand behind model training, inference, enterprise AI, and AI agents.
 
Neocloud demand is supported by several factors:
  • AI startups need scalable compute without building their own data centers
  • Enterprises need reliable infrastructure to move AI into production
  • Large technology companies are securing future capacity in advance
  • Inference workloads may create recurring long-term compute demand
  • Specialized cloud platforms can optimize around AI performance and utilization
 
This is why Serenity places neoclouds at the center of the AI infrastructure cycle. Compute is no longer just a background input; it is one of the main constraints on AI growth.
 

2. Nebius as a Major AI Cloud Re-Rating Candidate

Nebius is one of the strongest names in Serenity’s neocloud thesis because it gives investors direct exposure to AI cloud infrastructure. The company is building a full-stack platform for AI developers and enterprises, supporting model training, inference, and production deployment rather than general-purpose cloud workloads. Its five-year AI infrastructure agreement with Meta has made the story more compelling, showing that major technology companies are securing future AI capacity early as compute becomes a strategic asset. Nebius has also reported strong revenue growth and is expanding its infrastructure footprint, including a large AI factory project in Pennsylvania with major power access. Still, the opportunity comes with risk because AI cloud infrastructure requires heavy capital spending, advanced hardware, data center construction, power supply, and high utilization. If financing pressure rises or execution slows, investors may become more cautious. Even so, Nebius remains one of the clearest examples of a company positioned around the AI compute bottleneck.
 
Important Nebius points include:
  • Nebius is focused on AI-native cloud infrastructure
  • Its platform supports training, inference, and production AI workloads
  • The Meta agreement improves long-term demand visibility
  • Strong revenue growth supports the AI cloud demand story
  • Major risks include capital intensity, execution, financing, and customer concentration
 

3. Why Photonics Is Becoming a Critical AI Data Center Layer

Photonics is becoming important because AI data centers need faster and more efficient ways to move data. Large AI clusters depend on thousands of GPUs and accelerators working together. These systems constantly exchange information between chips, servers, storage devices, and networking equipment. If networking is slow, the entire cluster becomes less efficient, even if the GPUs are powerful. This is why optical networking is becoming a major infrastructure theme.
 
Photonics uses light-based technology to transmit data at very high speeds. In AI data centers, this can improve bandwidth, reduce latency, and support larger clusters. As hyperscalers move toward faster infrastructure, demand is shifting from older optical systems toward 800G and 1.6T transceivers. These upgrades are not just technical improvements; they are part of the broader AI capacity buildout. The larger AI clusters become, the more important optical networking becomes.
 
Photonics matters because:
  • AI clusters need fast communication between GPUs and servers
  • Networking can become a bottleneck if it does not scale with compute
  • Optical transceivers help support higher bandwidth and lower latency
  • Hyperscalers are upgrading data center networks for AI workloads
  • Photonics may become one of the next AI supply-chain themes after GPUs and memory
 
This makes photonics one of the most important early-stage areas in Serenity’s thesis. The GPU trade has already received major attention, but optical networking may become more visible as investors study the full AI data center stack.
 

4. AAOI and the 1.6T Optical Transceiver Upgrade Cycle

Applied Optoelectronics, or AAOI, is one of the companies connected to the photonics part of Serenity’s thesis. The company supplies optical networking products used in data center infrastructure, and its first volume order for 1.6T data center transceivers from a major hyperscale customer shows that AI networking demand is moving into real commercial orders. This is important because 1.6T transceivers are designed to support the higher bandwidth requirements created by larger AI clusters.
 
AAOI’s story explains why photonics can become a re-rating theme. Investors first focused on the chips that power AI systems, but as cluster sizes grow, the surrounding infrastructure becomes more important. Optical transceivers are part of that surrounding infrastructure. If hyperscalers continue upgrading toward higher-speed networking, companies with exposure to 800G and 1.6T products may benefit from stronger demand. However, AAOI also shows the risks of the theme because optical suppliers can be sensitive to customer concentration, margin pressure, production execution, and order timing.
 
Key AAOI points include:
  • AAOI supplies optical products used in data center networks
  • The company received a volume order for 1.6T transceivers from a major hyperscale customer
  • 1.6T technology supports higher bandwidth for AI workloads
  • Demand may rise as AI clusters become larger and more network-intensive
  • Risks include customer concentration, production execution, margins, and valuation volatility
 
Nebius and AAOI represent different parts of the same AI data center story. Nebius is tied to compute capacity, while AAOI is tied to bandwidth and networking. Both show why AI infrastructure investing is expanding beyond the obvious chip leaders.
 

Memory Stocks, HBM Demand and the Next Phase of AI Infrastructure Investing

Memory is one of the most important parts of Serenity’s AI infrastructure thesis because AI systems depend heavily on bandwidth and capacity. For many years, memory companies were treated mainly as cyclical semiconductor businesses. Investors watched DRAM and NAND pricing, inventory levels, supply growth, and demand cycles. AI is changing this framework because high-bandwidth memory is becoming a strategic component in advanced AI accelerators. Without enough fast memory, powerful GPUs cannot perform efficiently. This is why Micron, SK Hynix, and Samsung are now viewed as core AI infrastructure names rather than only traditional memory suppliers.
 

1. Why HBM Demand Is Reshaping the Memory Stock Narrative

HBM demand is reshaping the memory sector because AI workloads require much more memory bandwidth than traditional computing. Large language models, multimodal AI systems, long-context inference, agentic AI, and enterprise deployment all need fast access to massive amounts of data. In AI servers, memory is not just a support component; it can directly affect performance. If accelerators cannot access data quickly, system efficiency declines and expensive compute capacity is wasted.
 
This is why investors are beginning to value memory companies differently. HBM is harder to manufacture than standard DRAM because it requires advanced stacking, packaging, testing, and customer qualification. Supply cannot be expanded instantly, which may support stronger pricing if demand remains high. Serenity’s thesis suggests that memory companies could receive a higher valuation if the market treats HBM as a structural AI infrastructure asset rather than a simple cyclical product.
 
HBM demand is supported by:
  • More memory content per AI server
  • Higher bandwidth requirements for advanced accelerators
  • Growth in inference, agentic AI, and long-context workloads
  • Complex manufacturing that limits quick supply expansion
  • Long-term customer agreements that may improve earnings visibility
 
This is why memory is central to the AI infrastructure cycle. Compute performance increasingly depends on how quickly data can be accessed and moved.
 

2. Micron’s Role in AI Memory and Storage Growth

Micron is a major part of the memory thesis because it has broad exposure to AI memory and storage. The company is positioning its portfolio around the full AI infrastructure hierarchy, from high-bandwidth memory and DRAM to data center SSDs and storage products. This matters because AI workloads require more than HBM alone. Training and inference systems also need dense storage, fast data movement, and reliable memory across the server stack.
 
Micron’s opportunity comes from rising memory content in AI servers and stronger demand for HBM products. If AI infrastructure spending continues, Micron may benefit from higher-value memory products, tighter supply, and growing customer demand from data center operators. At the same time, Micron remains exposed to the risks of the memory cycle. Pricing can weaken if supply expands too quickly, and competition from SK Hynix and Samsung remains intense. The key question is whether AI demand is strong enough to reduce the severity of traditional memory cycles.
 
Important Micron points include:
  • Micron is expanding its AI memory and storage portfolio
  • HBM is part of the current AI accelerator demand cycle
  • AI data centers need DRAM, HBM, NAND, and SSD products
  • Tight supply can support stronger pricing and customer commitments
  • Risks include competition, supply growth, pricing cycles, and high expectations
 
Micron’s re-rating depends on whether investors believe AI memory demand is durable rather than temporary.
 

3. SK Hynix and the HBM-Led Memory Supercycle

SK Hynix is one of the clearest beneficiaries of the AI memory cycle because it has a strong position in high-bandwidth memory. Serenity’s thesis includes SK Hynix because HBM is essential for AI accelerators, and SK Hynix remains closely tied to the leading edge of AI memory supply. The company has emphasized HBM3E and HBM4 as central products for the 2026 market, with HBM3E expected to remain important while HBM4 begins shaping the next phase of growth.
 
The SK Hynix story also explains why South Korean semiconductor exposure is relevant to the AI infrastructure thesis. Because South Korea is home to major memory leaders, investors sometimes look at broader vehicles such as EWY for exposure to the country’s semiconductor ecosystem. However, EWY is not a pure AI memory investment because it includes many sectors beyond semiconductors. It is better understood as a broader South Korea exposure tool that may benefit if memory leaders continue gaining market attention.
 
Key SK Hynix points include:
  • SK Hynix is a major leader in high-bandwidth memory
  • HBM3E remains important in the 2026 AI memory cycle
  • HBM4 supports the next generation of AI accelerator platforms
  • The company has strong exposure to AI data center demand
  • Risks include capacity expansion, customer concentration, competition, and valuation pressure
 
SK Hynix may remain central to the AI memory trade if HBM demand continues to exceed available supply.
 

4. Samsung’s HBM4 and HBM4E Push in the AI Memory Race

Samsung Electronics is another important name in the AI memory thesis because it combines scale, manufacturing depth, and a broad semiconductor ecosystem. The company is pushing HBM4 and HBM4E products for next-generation AI systems, where higher bandwidth, larger capacity, and better energy efficiency are becoming critical. Samsung’s strength comes from its ability to compete across memory, logic, foundry, packaging, and advanced manufacturing, giving it the resources to challenge rivals in the fast-growing HBM market. However, execution remains the key risk because AI customers require strict performance standards and product qualification. If Samsung gains stronger traction with HBM4 and HBM4E, investor confidence could improve and the company may become a bigger beneficiary of the AI memory re-rating cycle.
 
Samsung’s AI memory thesis includes:
  • HBM4 and HBM4E development for next-generation AI systems
  • Large-scale manufacturing across memory and semiconductor technologies
  • Potential to regain or expand share in advanced HBM supply chains
  • Exposure to broader AI data center and semiconductor demand
  • Execution risk if qualification or customer adoption trails competitors
 
Samsung is important because it could add more competitive supply to the AI memory market while also benefiting from the sector’s long-term growth.
 

5. Why Memory Stocks Could Receive Higher Valuation Multiples

Memory stocks could receive higher valuation multiples if the market believes HBM demand is structural. In past cycles, investors often discounted memory companies because the industry could quickly move from shortage to oversupply. AI does not remove that risk, but it may improve the quality of demand. HBM is technically complex, customer-specific, and essential for AI accelerators. If supply remains tight and customers sign longer-term agreements, investors may treat leading memory companies differently from traditional DRAM-cycle stocks.
 
The re-rating argument also depends on inference growth. Training created the first wave of AI infrastructure demand, but inference may become even more important as AI applications move into daily use. Enterprise copilots, AI agents, search tools, robotics, and multimodal systems could all increase memory requirements. If this happens, memory companies may benefit from higher content per server and more predictable demand. This is why Serenity places memory alongside neoclouds and photonics as a core infrastructure theme.
 
Reasons memory stocks could be re-rated include:
  • HBM is essential for AI accelerator performance
  • AI servers use more memory than traditional servers
  • HBM supply is difficult to expand quickly
  • Long-term customer commitments may support earnings visibility
  • Inference growth could extend demand beyond the first training wave
  • Investors may assign higher multiples if memory becomes less purely cyclical
 
The opportunity is significant, but it still requires careful stock selection because memory remains a competitive and capital-intensive industry.
 

Why AI Infrastructure Also Matters for Web3

Although Serenity’s thesis mainly focuses on AI infrastructure stocks, the theme also connects indirectly to crypto. As AI demand grows, crypto sectors such as decentralized compute, DePIN, blockchain-based data networks, and AI agents may become more relevant because they aim to support open infrastructure for compute, storage, and automation. This does not mean Nebius, photonics, or HBM are crypto projects, but the same infrastructure trend is important for Web3 because future AI applications may need cheaper compute, verifiable data, decentralized networks, and machine-to-machine payments.
 

Key Risks in Serenity’s AI Infrastructure Thesis

Serenity’s AI infrastructure thesis highlights strong long-term opportunities, but the theme is not without risk. Neoclouds, photonics, and memory are capital-heavy sectors where valuations, customer demand, supply cycles, and execution can change quickly. Investors should understand these risks before treating the AI infrastructure re-rating as a guaranteed trend.
  • Valuation risk: AI infrastructure stocks may already price in strong future growth, leaving less room for upside.
  • Capital intensity: Neoclouds, data centers, memory, and optical suppliers need heavy investment to scale.
  • Dilution risk: Companies may issue shares or raise debt to fund expansion, which can pressure shareholders.
  • Customer concentration: Many suppliers depend on a few large hyperscale buyers, creating order-delay risk.
  • Supply-cycle risk: HBM and optical markets could move from shortage to oversupply if capacity expands too fast.
  • Execution risk: Product ramps, data center builds, power access, and customer qualification can face delays.
  • AI spending risk: If hyperscalers slow AI capex, demand for compute, photonics, and memory could weaken.
 

Conclusion

Serenity’s AI thesis shows that the next phase of the AI market may be driven less by software hype and more by infrastructure demand. Neoclouds like Nebius, photonics players such as AAOI, and memory leaders including Micron, SK Hynix, and Samsung are gaining attention because they support the real bottlenecks behind AI growth: compute, data movement, and high-bandwidth memory. The opportunity is strong, but investors still need to watch valuation, dilution, customer concentration, and execution risk. Overall, the thesis suggests that AI infrastructure could become one of the most important re-rating stories in the next technology cycle.
 

FAQs

What is Serenity’s AI thesis?

Serenity’s AI thesis is that the next phase of the AI market may shift from software hype to infrastructure demand. The thesis focuses on neoclouds, photonics, and memory because these areas support the real backbone of AI growth: compute capacity, data movement, and high-bandwidth memory.

Why is AI infrastructure becoming more important?

AI infrastructure is becoming more important because advanced AI models need massive data centers, GPU clusters, optical networking, memory, storage, power, and cooling to operate at scale. As companies move from AI testing to real deployment, demand for these infrastructure layers continues to grow.

What are neoclouds in AI?

Neoclouds are specialized cloud infrastructure providers built mainly for AI workloads. They offer GPU capacity, high-performance computing, model training support, and inference infrastructure, making them different from traditional cloud platforms that focus on broader enterprise computing.

Why is Nebius important in Serenity’s AI thesis?

Nebius is important because it is positioned as an AI-native cloud infrastructure company. It gives investors exposure to the AI compute capacity theme, especially as major technology companies and enterprises look for reliable cloud infrastructure to support training, inference, and production AI workloads.

What is photonics in AI data centers?

Photonics refers to light-based technology used to move data at very high speeds. In AI data centers, photonics helps improve bandwidth, reduce latency, and support large GPU clusters where fast communication between chips, servers, and storage systems is essential.

Why is memory important for AI infrastructure?

Memory is important because AI accelerators need fast access to large amounts of data. High-bandwidth memory, or HBM, allows GPUs and AI chips to process workloads more efficiently. Without strong memory bandwidth, even powerful processors may not reach full performance.

Which companies are connected to Serenity’s AI infrastructure thesis?

The main companies connected to Serenity’s thesis include Nebius for AI cloud infrastructure, AAOI for photonics and optical networking, and Micron, SK Hynix, and Samsung for AI memory and HBM demand. Each company represents a different part of the AI infrastructure supply chain.

What are the biggest risks in the AI infrastructure trade?

The biggest risks include high valuations, heavy capital spending, dilution risk, customer concentration, supply expansion, and execution challenges. AI infrastructure is a strong long-term theme, but stocks in this sector can be volatile if growth expectations become too high or demand slows.
 
 

Disclaimer

The information provided on this page may originate from third-party sources and does not necessarily represent the views or opinions of KuCoin. This content is intended solely for general informational purposes and should not be considered financial, investment, or professional advice. KuCoin does not guarantee the accuracy, completeness, or reliability of the information, and is not responsible for any errors, omissions, or outcomes resulting from its use. Investing in digital assets carries inherent risks. Please carefully evaluate your risk tolerance and financial situation before making any investment decisions. For further details, please consult KuCoin’s Terms of Use and Risk Disclosure.