Hook
Over the past seven days, the AI-focused portfolio of prominent crypto and tech investor ‘Serenity’ lost 49.4% of its value. The account, which returned 4502% in 2026, is now bleeding capital. But the raw numbers hide a more critical story: the three holdings—AXTI, SIVE, and AAOI—are not random AI plays. They are precise bets on the physical bottlenecks limiting AI scaling. When the market repriced these bottlenecks from ‘promising’ to ‘uncertain’, the drawdown became a systemic signal. The data shows that even a 49.4% drawdown after a 4502% gain leaves a net multiple of 23x initial capital. Optimism remains, but the code of the portfolio—its concentration, its time dependence, and its vulnerability to technological disruption—needs to be audited.
Context
The portfolio’s thesis is rooted in the physical layer of AI infrastructure: optical interconnects and compound semiconductor substrates. AXTI (AXT Inc.) is a leading supplier of gallium arsenide and indium phosphide substrates used in laser chips for 800G/1.6T optical modules. SIVE (Sivers Semiconductors) focuses on photonic chips for next-generation AI cluster interconnects. AAOI (Applied Optoelectronics) manufactures the optical modules themselves, with vertical integration into laser chip fabrication. Together, they represent the supply chain for the datacenter’s bandwidth bottleneck. As AI models grow, the cost and power of optical communication become the limiting factor. Serenity’s bet was that 2027 H2 would be the inflection point for these suppliers, driven by the transition from 800G to 1.6T optics and the maturation of silicon photonics. The drawdown suggests the market is now pricing in execution risk.
Core
We need to dissect each holding at the code level—where the code is not Solidity but the financial and operational contracts of these companies.
AXTI: The indium phosphide (InP) substrate market is a classic duopoly (Sumitomo and AXT). AXT’s competitive advantage is its lower-cost manufacturing base in China, but this exposes it to geopolitical tail risk. The revenue inflection point of 2027 H2 assumes that 1.6T optical modules will enter volume production by then. Industry estimates from LightCounting place 1.6T module volume ramp in mid-2027. However, the adoption curve is non-linear: hyperscalers may skip 1.6T and jump to co-packaged optics (CPO). If CPO displaces traditional pluggables, demand for InP substrates could soften. The market’s 49.4% drawdown on AXT includes a 1.8% increase on the drawdown day—a potential dead cat bounce or insider buying? Without a detailed audit of AXT’s order book, the drawdown appears to be a correction of over-optimism rather than a fundamental collapse.
SIVE: This is the highest-risk, highest-reward bet. Sivers’ technology—silicon photonics for short-reach interconnects—promises lower power and latency than traditional optics. But the technology is pre-revenue for AI datacenters. The company’s revenue today comes from military and space contracts, which have low correlation to AI. The 2027 H2 inflection point hinges on winning a design win with a major optics vendor like Cisco or Lumentum. The probability of a design win by mid-2027 is moderate at best, given that Intel, Coherent, and Lumentum are all investing heavily in silicon photonics. The drawdown of 49.4% in the portfolio includes SIVE’s drop of approximately 60% in the same period (estimated from the aggregate). This suggests the market is pricing in a high probability of failure.
AAOI: The strongest of the three from a fundamentals perspective. AAOI has vertically integrated laser chip manufacturing, giving it a cost advantage over Chinese competitors like Zhongji Innolight. The 2027 H2 inflection point for AAOI is most plausible: 1.6T modules will be in early volume by then, and AAOI’s existing relationships with Microsoft and Amazon provide a path. However, the gross margin pressure from fierce competition is real. AAOI’s Q4 2026 gross margin was 32%, down from 38% a year prior. The portfolio’s drawdown likely reflects a repricing of AAOI’s margin compression risk.
Prescriptive Risk Mitigation: For any portfolio builder, concentration in three high-beta names is a structural risk. Serenity’s cost basis is low enough to withstand the drawdown, but a 50% drawdown on a concentrated portfolio implies a 90% drawdown on the original investment if leverage is involved. No leverage disclosure exists. The drawdown also reveals a time vulnerability: if the 2027 H2 inflection point slips by one quarter, the portfolio could face another 30% decline. The code of the portfolio—its liquidity, its correlation to macro, and its dependency on a specific technological roadmap—must be tested against adverse scenarios.
Contrarian Angle
The conventional narrative is that the drawdown is noise in a bull market for AI infrastructure. But the data suggests a deeper blind spot: the market is pricing in a technology substitution risk that the thesis ignores.
Co-packaged optics (CPO) is the silent threat. Traditional pluggable optical modules (800G, 1.6T) are reaching power and density limits. CPO integrates optics directly with switch ASICs, eliminating the need for pluggable modules. Major vendors like Broadcom, Cisco, and Intel are investing heavily. If CPO reaches commercial viability by 2028, the entire demand curve for traditional optical modules (and thus for AXTI and AAOI) shifts left. The 2027 H2 inflection point that Serenity expects may be the peak of the traditional module cycle, not the beginning.
Furthermore, the assumption that ‘bottleneck’ means 'high demand’ is flawed. Bottlenecks attract competition. For example, in the optical interconnect space, Chinese suppliers like Zhongji Innolight and Eoptolink are aggressively expanding capacity, potentially creating oversupply by 2028. AAOI’s vertical integration advantage may not be sufficient to maintain pricing power.
Another blind spot: regulatory risks. AXTI’s Chinese manufacturing base is subject to US export control. If the DOC imposes new restrictions on indium phosphide substrates deemed critical for national security, AXTI could lose access to customers like NVIDIA. SIVE’s military revenue provides a floor but also ties its fortunes to government contracts, which are unpredictable.
The portfolio’s optimism is also a behavioral risk. Public statements of confidence can mask hidden sell orders or position reductions. Serenity’s statement, “I can accept higher volatility,” may be a tactic to reassure followers while quietly hedging. The market is indifferent to narratives.
Takeaway
The Serenity drawdown is a case study in the risks of investing in deep tech bottlenecks. The portfolio’s thesis is sound at the physics level, but execution risk, technological displacement, and concentrated leverage create a fragile structure. The 2027 H2 inflection point is a make-or-break moment. If the industry transitions to CPO or if hyperscaler CapEx slows, the portfolio could face another 50% drawdown. The ledger does not forgive optimism without verification.
For investors tracking similar plays, the key metric is not the drawdown size but the velocity of technological displacement. Track the orders for 1.6T modules, monitor CPO prototype announcements, and audit each company’s customer concentration. Trust nothing. Verify everything. Complexity is the enemy of security. The data does not care about your narrative.