01
Algorithm over ornament
We pursue structural advances, not cosmetic optimization. The target is a more powerful and more efficient architectural primitive.
Founder-led AI infrastructure research
Building the next
transformer paradigm.
MojoLynx is a focused research company pursuing algorithmic breakthroughs that challenge the assumptions of today's transformer stack — from training efficiency to systems-level deployment behavior.
- StageDeep research
- ModelNew transformer algorithms
- LeadershipCEO-led, founder-built
Minimal team. Maximal leverage. Research first.
Core thesis
AI infrastructure does not need incremental tuning.
It needs a new algorithmic shape.
We investigate transformer designs that can reset the tradeoffs around compute efficiency, scaling behavior, and the way intelligence is expressed across hardware and software boundaries.
02
Infrastructure-native thinking
Model design is inseparable from runtime behavior, memory topology, and serving economics. We treat these constraints as first-class design inputs.
03
Long-horizon R&D
Built with room to accumulate: ideas, experiments, benchmarks, papers, prototypes, and future productization.
Research surface
Where we're concentrated.
Current focus areas, with room to grow as the research matures.
Next-generation transformer architectures
Moving beyond today's default assumptions about attention, representation, efficiency, and scaling.
Training & inference efficiency
How algorithmic redesign can reshape throughput, memory pressure, and deployment economics across future AI systems.
Infrastructure implications
How architecture decisions propagate into runtimes, orchestration, hardware strategy, and product infrastructure.
Ideas
The thinking beneath the architecture.
Before redesigning the architecture that carries intelligence, get clear on how intelligence actually exists. That conviction grounds the "intelligence field" framework — a way of seeing large language models not as persons or tools, but as high-dimensional response spaces shaped by training into a navigable terrain.
Leadership
Led by Tao Feng.
CEO and currently MojoLynx's sole operator. The company combines deep AI focus with a deliberately interdisciplinary perspective — shaping how problems are framed, how assumptions are challenged, and how new system-level ideas emerge.
"The goal is not to decorate the existing stack. The goal is to discover a stronger one."