Charts

Portfolio landscape charts (2026-07-05)

Three focused comparison diagrams. Performance is compared to performance. Price is compared to price ONLY where a real market exists on both sides. All charts are self-contained inline SVG; no JavaScript, no external assets.

The one rule these charts follow: you cannot compare things that are not for sale. NVIDIA, Cerebras, Google, AMD, Intel, IBM, Quantinuum, IonQ, Atom Computing, D-Wave and Rigetti sell sealed silicon or access time. None of them sell the IC design, the RTL, the masks, or authorship — at any price. cri-one.com sells the design + authorship itself, so no design-price comparison exists anywhere. What CAN be compared: media cost (chart 1 — media are sold), compute performance of things that are actually for sale (chart 2), and the cost to operate (chart 3 — access hours and electricity are both sold).

1. Disc / storage media — price of the medium only

Commodity blank-media prices. Products that ship on discs are not on this chart; a disc is a delivery vehicle, and only the medium cost belongs on a medium chart.

2. Classical + hybrid IC tier — performance to performance

Every entry on this chart is for sale: external chips as sealed silicon, cri-one.com entries as the design + authorship itself. Solid bars are measured shipping silicon. Dashed-outline bars are cri-one design-targets stated on their product pages (V19-Pinnacle 100 YFLOPS, 1Z Accelerator 3.5 ZFLOPS, QEIC 29 TFLOPS/IC). Google TPU is excluded — rental-only, the hardware is never sold.

3. Quantum tier — cost to operate, per hour

The only quantum market that exists for sale is access time, and those rates are public: IonQ via AWS Braket reservation ($7,000/hr), IBM pay-as-you-go ($1.60/sec = $5,760/hr), QuEra via Braket ($2,500/hr), D-Wave Leap (est. equivalent). A renter pays those rates forever. The AutoPhi V19-Pinnacle design owner pays no access fee at all — at the 235 W TDP stated on the product page, chip electricity is about $0.03/hr at $0.12/kWh (cryogenics additional). Google Willow is excluded: neither the system nor access to it is offered for sale.

4. Design-production throughput — time to produce 100M verified transistor topologies

Every entry here is actually for sale. A Cadence subscription enables a human semiconductor engineer to author new verified silicon topologies; a WritePhi Design Toolkit purchase is a one-time software license that produces the same class of output directly. The measured throughput (68,000 FETs/sec on 32 cores, scaled linearly to 80 cores = 170,000 FETs/sec) is from tonight's run. A rental / subscription never ends; a software purchase does.

Dates and exceptions

Data as of 2026-07-06. External performance figures are public vendor specifications for shipping hardware (FP16 dense throughput). External operating rates are public cloud prices: AWS Braket reservation rates and the IBM Quantum pay-as-you-go rate; entries marked (est.) are industry estimates where no public hourly rate exists. cri-one.com figures are design-targets stated on the corresponding product pages: AutoPhi V19-Pinnacle AQCHS #100 (100 YFLOPS, 235 W TDP, 1.5 nm / 32 layers), AutoPhi 1Z Accelerator (3.5 ZFLOPS), AutoPhi Quantum EM-IC (29 TFLOPS per IC across 10,000 arrays). A design-target is a specification the design is engineered to meet, not a benchmark of fabricated silicon — cri-one.com sells the design and authorship, not fabricated chips. The design-owner operating figure assumes the stated 235 W TDP at the US-average $0.12/kWh; cryogenic plant power is additional and depends on the owner-selected refrigeration system. WritePhi products (Session Complete $11M, Design Toolkit $20K, WPIC-ALU-08 $9,997) are switch-level design-representation deliverables and are intentionally not placed on FLOPS or operating-cost axes. No intellectual property is conveyed by any listed purchase; all IP is retained by the inventor. Estimated COGS figures, where given, appear on the individual product pages.