Lyken.AI Announces Non-Binding MOU for 128-Node B300 AI Compute Cluster

Alset AI’s 90.1%-owned Lyken.AI entered a non-binding memorandum of understanding with Singapore-based SWARMNET to support a proposed five-year commercial arrangement for large-scale AI inference workloads and token optimization. The agreement is subject to definitive arrangements. The update is early-stage (MOU, not binding), but it is a positive potential pathway for scaling Lyken’s AI cloud compute business.

Analysis

This is more of a credibility/optionality event than a revenue event. A non-binding MOU around AI inference only matters if it turns into a priced, capacity-defined contract; until then, the market is effectively assigning value to a call option on utilization that may never monetize. For a microcap like GPUS, the immediate impact is more likely a sentiment-driven rerating than a change in earnings power, which raises the odds of a sharp giveback once traders realize there is no enforceable demand commitment.

The second-order read is that the real winners, if anything materializes, are the infrastructure layers with scarce power, GPUs, or managed inference distribution rather than a small-cap venture wrapper. If this deal is genuine, it suggests demand is shifting from headline training narratives toward lower-cost inference optimization, which favors operators that can squeeze utilization and sell repeatable compute with better gross margin density. But that is a months-to-years thesis, and it only becomes investable if disclosed economics show contracted capacity, minimum spend, or take-or-pay terms.

The contrarian view is that the market may be overestimating the quality of the signal: MOUs are easy to announce and hard to collect cash against. The main falsifier for a bullish GPUS view is the absence of a definitive agreement with explicit pricing, term, and customer concentration disclosure within 1-3 months; conversely, a material upward move in disclosed backlog or monthly recurring compute revenue would validate the story. Absent that, any spike in GPUS should be treated as a liquidity event, not a fundamental inflection.

Key Decisions for Investors

• Fade strength in GPUS on any post-announcement gap move of >15-20% unless a definitive agreement is signed; set a tight time stop of 1-2 weeks because the catalyst decays quickly without contract detail.
• Do not underwrite valuation uplift until GPUS discloses contracted compute revenue, minimum usage, or customer economics; treat this as a watch item for the next 30-90 days rather than a buy.
• If expressing the broader AI infrastructure theme, prefer higher-quality GPU/inference operators with real backlog and balance-sheet strength (e.g., IREN, HUT, NBIS) over GPUS; the spread trades better reflect actual capacity scarcity.
• Use a pairs lens: long profitable AI infrastructure exposure, short speculative microcap AI wrappers like GPUS if borrow/liquidity permit; the edge is in separating contracted economics from announcement optionality.
• Falsifier for the bearish read: a definitive agreement that includes committed spend, term length, and utilization floors; if that lands, reassess immediately for a tactical long despite the current low-conviction setup.