TsugiCinema × NVIDIA
12-Slide Master Pitch Deck · HumanX 2026, Moscone Center
Software is the wedge. Trinity is the cash-flow moat.
Infinity is the venture-scale optionality.
Hoskins · Behnam · Hereth · Barnett · Hart
Slide 1 · Proof of Demand
100,000 Enthusiasts Downloaded Our Barebones Build.
The Signal
A Reddit guide for bit-for-bit Dolby Vision Profile 7 FEL 12-bit 4:2:0 playback on Amlogic S922X generated 100,000+ organic downloads with zero marketing spend and zero customer acquisition cost.
Why It Matters
This is pure, uncoerced demand from a demographic the industry has written off as too complex to serve. These 100,000 users are the top of the funnel for a hardware company that has not yet existed. The cohort is ready; the product is the only missing piece.
Slide 2 · The Market Failure
QD-OLED Panels Can Render 12-Bit Color. Streaming Gives Them 4-Bit Equivalent.
The gap between panel capability and content delivery is staggering. Modern QD-OLED displays are engineered to receive and render extraordinary color fidelity, yet the dominant delivery mechanism systematically discards that capability.
Streaming Reality
4K HDR streams compress to 15 to 25 Mbps, discarding approximately 80% of the color data the panel can physically display.
Hardware Vacuum
NVIDIA SHIELD Pro has not been refreshed in seven-plus years. SoC vendors are reallocating silicon from video decode to NPUs, abandoning the category.
The Workaround Economy
Over 100 million Americans resort to piracy or NAS hacks. There is no legitimate, high-fidelity local-playback platform at any price point.
Even the best consumer quality file avaiable in the format of 4K UHD Blu-Ray serving the dual-layer Dolby Vision Profile 7 FEL, is chroma subsampled delivering 100% luminance data, but only 25% of the original file's color data.
Slide 3 · The Wedge
TsugiOS: A Localized, High-Fidelity
Spotify for Premium Cinema
TsugiOS is a four-layer system.
  • Custom Android-based OS. Pre-downloaded 50 to 100 GB master files. Sustained 70 to 130 Mbps local playback.
  • Existing Amlogic S992X owners can download and install.
  • Phase 1 ships on white-labeled Ugoos hardware (zero CapEx OEM relationship) pre-flashed with our Android ROM to lower barrier of entry for consumer
  • Captures the 100k Reddit cohort as the day-one user base.
Slide 4 · Trinity Hardware
Trinity: The Hardware Moat (USPTO 63/987,139)
Three commodity SoCs synchronized to sub-scanline alignment without genlock. A patented architecture that produces image quality four times richer than Blu-ray.
3× RK3588 Decoders
Each Rockchip RK3588 runs free on its own oscillator (plesiochronous). No shared clock. No genlock. No clock distribution network required.
AMD Xilinx Artix UltraScale+ FPGA
FPGA compositor with elastic FIFOs 16 scanlines deep. Processes and composites all three decode layers pixel by pixel at 594M pixels/sec.
Hysteretic Sideband GPIO
1.8V LVCMOS GPIO drives V-Blank modulation. One pin per decoder. Watermark comparator eliminates clock distribution entirely.
12-bit 4:4:4 HDMI 2.1 FRL Output
Uncompressed 12-bit 4:4:4 output. Four times the color data of Blu-ray. Full panel capability finally reached.
Trinity: The Hardware Moat (USPTO 63/987,139)
The patented Trinity architecture synchronizes three commodity SoCs to achieve sub-scanline alignment without traditional genlock, delivering unparalleled image fidelity.
Slide 5 · How Trinity Works
V-Blank Modulation as a Clock-Correction Sideband. 30 LUTs Per Decoder.
Trinity's synchronization mechanism is elegant precisely because of its simplicity. Three free-running decoders, each on its own oscillator, are kept in lock-step by a single GPIO pin per decoder, driven by a watermark comparator watching the depth of an elastic FIFO buffer.
  • Each FIFO read at 148.5 MHz fabric clock; watermark comparator at WM_HIGH=14, WM_LOW=4 scanlines.
  • When fill exceeds WM_HIGH, the FPGA asserts a single GPIO pin. The decoder responds by extending its next vertical blanking interval.
  • Hysteresis between watermarks prevents chatter. MTBF of the synchronizer exceeds 100 years.
  • No shared clock. No genlock. No clock distribution network.
Why This Is Unprecedented
Prior art requires genlock or shared clock distribution networks, adding cost and complexity. Trinity achieves sub-scanline alignment across three plesiochronous oscillators using 30 LUTs and one GPIO pin per decoder. No shared clock. No genlock. No clock distribution network. The synchronizer MTBF exceeds 100 years of continuous operation.
Bandwidth multiplier: ~150 Mbps in becomes 18+ Gbps out. Over 120×.
V-Blank Modulation as a Clock-Correction Sideband. 30 LUTs Per Decoder.
V-Blank modulation as a clock-correction sideband. 30 LUTs per decoder.
Slide 6 · Unit Economics
Commodity Silicon. 8-Layer PCB.
Margins Like a Software Company.
BOM total: $517 to $584 at volume. All commodity components. No proprietary silicon required.
Preferred embodiment in the patent used expensive components to provide for more overhead. For example, the Decoder SoC's can be $10 phone video decoders.
It is entirely possible to lower the BOM total to achieve a $100 to $200 Retail price to compete with streaming boxes such as Apple TV 4K or Nvidia Shield Pro.
Retail ASP
$899 to $999 · 40%+ gross margin
Concierge ASP
$5,000 to $15,000 bundled · 70%+ gross margin
Manufacturing
White-label Ugoos OEM · Zero factory CapEx
Slide 7 · Infinity
Infinity: The Venture-Scale Optionality
The Trinity sync fabric scales from 3 video decoders to 64 AI training nodes. Same engineering DNA; orders-of-magnitude larger addressable market.
Same Core Primitives
Elastic FIFO, hysteretic watermark, sideband backpressure. The identical mechanisms that sync three video decoders now coordinate 64 distributed AI training nodes.
White Rabbit Sub-ns Precision
White Rabbit (WR) protocol disciplines per-node TCXO/OCXO oscillators to sub-nanosecond precision across the entire 64-node fabric. No GPS. No PTP degradation.
CXL.mem with Regime Versioning
CXL.mem with regime versioning eliminates nondeterministic cache eviction during gradient consensus. Deterministic memory behavior enables hardware-level quorum.
1,361 ns Worst-Case Consensus
Worst-case round-trip consensus is 1,361 ns. Roughly an order of magnitude under best-case NCCL software all-reduce on equivalent tensor sizes.
Infinity: The Venture-Scale Optionality
Slide 8 · Why Infinity Matters to NVIDIA
A PHY/MAC Primitive That Sits Beneath NCCL, Not Against It
Infinity is not a competitor to NCCL. It is a hardware substrate layer beneath the software stack that makes NCCL's own performance more deterministic and lower-latency. The positioning is deliberate: designed to be integrated, not licensed standalone.
NCCL Software All-Reduce
10 to 100 microseconds per round trip on 1 to 10 GB tensors. Non-deterministic under contention.
Infinity Hardware Quorum
Approximately 1,361 ns worst case. Two to three orders of magnitude of headroom versus software baseline.
Hardware Byzantine Defense
Native SignSGD majority voting, weighted averages, and Byzantine quorum support. Hardware defense against catastrophic forgetting in LLMs.
Latency Budget Breakdown (1,361 ns)
Slide 9 · The Team
Hyperscale Systems. Physical AV Operations. Municipal-Grade Finance.
Tong Liu · CEO / CTO
Five years on Google Searchbox; eliminated 1.5B daily RPCs from an 8.5B-query/day pipeline. Inventor on both the Trinity and Infinity provisional patents. Architect of the full technical stack.
Dylan Lindeberg · COO
Audio engineer and commercial AV builder for national flagship installations. Owns the concierge channel relationship and the Shenzhen supply chain. Zero-CapEx OEM execution.
Wenling Ma · CFO
Chief of Staff at Boston Water and Sewer Commission, a $100M+ public utility. Three decades of governance, compliance, and institutional financial management experience.
Jason So · Marketing Consultant
Owns top-of-funnel strategy and the Reddit enthusiast community that generated 100,000 organic downloads. Owns custom UI for CoreElec developed over the course of 6 months of iteration.
Slide 10 · Roadmap
Software Ecosystem First. Hardware Silicon Second. AI Infrastructure as the Long Arc.
Three disciplined phases. Ring-fenced capital. Each phase funds the next without cross-contaminating the focus of the team on delivery.
Phase 1 · Months 0–14
TsugiOS MVP on Ugoos AM9/AM6B+. Concierge channel launch. Cash-flow positive at approximately 150 installs. $500K ring-fenced strictly here.
Phase 2 · Months 12–24
Trinity reference hardware tape-out. Retail SKU at $899 to $999. Concierge bundles scaling to $5,000 to $15,000. Supply chain already in place via Ugoos OEM.
Phase 3 · Months 18+
Infinity feasibility prototype; 8-node bench. Public benchmark versus NCCL software baseline. Zero CapEx in this raise. Documentation and IP track only until Trinity is profitable.

Ring-fencing is the single most important structural commitment in this deck. $500K goes to Phase 1 and Phase 1 only.
Infinity carries zero CapEx until Trinity is cash-flow positive.
Slide 11 · Risk and Mitigation
What We Know Is Hard, and
What We Are Doing About It
One honest risk slide buys more credibility than ten additional upside slides. Here is what we have identified and how we have mitigated each exposure.
DRM and Studio Licensing
Strict TEE isolation throughout. Phase 1 libmpv is kept architecturally separate from Phase 3 Media3 and Widevine L1 integration. Open-HDR fallback available via expired Demos and Demografx patents. Licensing pathway is documented and staged.
Dual-Track Defocus Risk
$500K is ring-fenced for Phase 1. Infinity is a documentation and IP-only track until Trinity is cash-flow positive. The team's operational bandwidth is not divided until the revenue base can support it structurally.
Hardware Capital Intensity
White-label Ugoos OEM relationship requires zero factory investment from TsugiCinema. No tooling, no minimum run commitments beyond initial pilot volume. This is the same capital-light model that funds Phase 1 entirely.
Slide 12 · The Ask
Three Things from NVIDIA, In Priority Order
We are not necessarily asking for a check. We are asking for three specific, named commitments that unlock disproportionate value for both sides.
1
NVIDIA Inception Onboarding
Formal program admission for developer credits, technical resources, go-to-market support, and ecosystem visibility. This is the foundation that legitimizes TsugiCinema's position within the NVIDIA partner network and accelerates every subsequent milestone.
2
DGX Cloud / OCI Credits
Compute credits to power the TsugiAI public-domain film restoration pipeline. This directly monetizes the content library side of the business while showcasing NVIDIA compute in a consumer-facing, high-visibility application.
3
A Named Technical Sponsor Inside NVIDIA Research
One engineer who can sanity-check the Infinity feasibility track on a 30-minute call.
A Level 11 Senior Engineering Manager at ASML has provided preliminary sign-off on the project.
Thank You
Tong Liu · CEO/CTO · tong@tsugicinema.com
TsugiCinema · HumanX 2026 · Moscone Center
Come Visit our Demo Room in San Francisco! Open House demo with 11 Speaker Active Noise Cancellation on Sunday 4/12!
Software
110,000-user wedge. Zero CAC. Organic demand already proven.
Trinity
Patented hardware moat. 40 to 70%+ margins. USPTO 63/987,139.
Infinity
Venture-scale optionality. A PHY/MAC primitive beneath NCCL. NVIDIA-shaped acquisition path.