# Market context and strategic case, November 2026 Research checked on 17 November 2026. These external indicators describe the problem space, not NodeLift revenue, eligible device inventory, a measured TAM, validated acceleration or environmental savings. Values are rounded only as explicitly noted. No annual growth curve is invented from isolated endpoints. ## Four market signals | Signal | Value and period | Evidence class | Meaning and boundary | | --- | --- | --- | --- | | Global corporate AI investment | USD 581.69 billion in 2025; displayed as USD 581.7B | Historical source-reported investment | Broad AI investment, including multiple financing activities. Not inference spend or NodeLift's addressable revenue. | | Unique mobile subscribers | 5.8 billion in The Mobile Economy 2026 executive summary | Industry estimate reported in 2026 | People/subscribers, not a count of phones, available nodes or compatible workers. | | Data-centre electricity | 485 TWh in 2025; central projection 950 TWh in 2030 | IEA estimate and projection | All data centres, not AI alone. No interpolation or claim that distributed phones reduce this demand. | | Global e-waste | 62 million tonnes generated in 2022; 22.3% documented formal collection/recycling | Historical global estimate, 2024 report | All e-waste, not a stock of reusable phones. Reuse still requires safety, useful work and lifecycle measurement. | Sources: [Stanford HAI, 2026 AI Index, Economy](https://hai.stanford.edu/ai-index/2026-ai-index-report/economy), [GSMA, The Mobile Economy 2026](https://www.gsmaintelligence.com/research/the-mobile-economy-2026), [IEA, Key Questions on Energy and AI, executive summary](https://www.iea.org/reports/key-questions-on-energy-and-ai/executive-summary), [ITU/UNITAR, Global E-waste Monitor 2024](https://www.itu.int/en/ITU-D/Environment/Pages/Publications/The-Global-E-Waste-Monitor-2024.aspx). The mobile report's exact baseline year should not be inferred from its title. The public page therefore identifies the 2026 report rather than assigning the 5.8 billion subscribers to an unverified observation year. The e-waste reference remains explicitly dated 2022. Dollar amounts are current source-reported USD, not inflation-adjusted comparisons or an infrastructure purchasing budget. ## The opportunity to test Our strategic hypothesis is that existing hardware can contribute useful, authorized AI work when a controller matches a task to measured resources, connection cost and operating policy. Personal clusters, office equipment and field deployments are candidate use cases. Broad infrastructure pressure makes this worth exploring; it does not establish product demand or a working runtime. A defensible market assessment needs discovery interviews, reachable customer segments, actual compatible hardware, supported workloads, willingness to pay and delivery cost. An addressable-capacity estimate must discount incompatible hardware, memory reserved by the OS, owner limits, connectivity, thermal budgets and scheduling overhead. Multiplying billions of subscribers by presumed RAM would not produce valid capacity or market size. ## Prior work: achievements and the remaining gap Calling all existing projects failures would contradict their published work. The following are source-reported mechanisms, not independently reproduced results or universal limitations across all versions. - [Apple MLX distributed inference](https://machinelearning.apple.com/research/mlx-distributed-inference), 15 October 2026: demonstrates model partitioning and scaling on Mac Studios with low-latency RDMA over Thunderbolt. This is a concrete achievement in its hardware/link configuration, not evidence about heterogeneous Wi-Fi phones. - [exo](https://github.com/exo-explore/exo): publishes device discovery, topology-aware partitioning and MLX-based inference. Its stated fast-link results should be understood in their recorded setup, not dismissed or projected onto NodeLift's mixed consumer hardware. - [llama.cpp RPC](https://github.com/ggml-org/llama.cpp/blob/master/tools/rpc/README.md): exposes remote GGML devices for distributed execution. Its documentation calls the backend a proof of concept and warns against exposing its RPC server on open networks. Public Hive therefore cannot simply proxy raw RPC to the world. - [DistAI](https://github.com/alex8819/DistAi): demonstrates an Android/PC partitioning prototype and documents practical limits. Our [commit-bound assessment](https://github.com/sev7enITA/NodeLift/blob/main/research/distai/LESSONS_LEARNED.md) preserves its contribution and the distinction between capacity expansion and chat speed. The unsolved product combination we target is broad hardware participation, local and remote topology, streamlined consent/onboarding, sustained useful performance, correctness and measured resource/energy contribution. This is a research direction, not a claim that NodeLift has solved that combination or surpassed the listed projects. ### Five barriers and falsifiable responses | Barrier | Why adding devices can disappoint | Evidence NodeLift must produce | | --- | --- | --- | | Model placement | Installed RAM is not an automatic shared address space; weights, KV cache and execution state must fit a supported partition. | A model that exceeds one node's actual budget completes correctly across the assigned nodes. | | Communication | Activation/KV transfer, serialization, queueing and synchronization can exceed saved compute time. | End-to-end task time and TTFT against the best applicable single-node/local baseline on the actual link. | | Heterogeneity | Backends, quantization, kernels, OS budgets and instruction sets differ. | Hardware/runtime probing plus compatible, repeatable workload qualification. | | Sustained operation | Thermal throttling, battery limits, background suspension and network changes affect availability. | Long-running results with energy coverage, pauses, failed attempts, fallback and uncertainty retained. | | Trust and correctness | Nodes and tenants need scoped authority; output and cached state must remain correct. | Isolation, consent, integrity checks, cancellation/recovery and appropriate reference-output tests. | ## PESTEL: strategic interpretation The following is our interpretation of public sources. It is not legal advice, a statement of compliance, an endorsement by a source organization or a claim that every response is implemented. | Factor | External pressure or opportunity | NodeLift design response / question | | --- | --- | --- | | Political | Compute access and digital sovereignty feature in the EU AI Continent agenda. | Allow explicit resource/location policies and auditable hub decisions; do not label four planned regions as active facilities. | | Economic | Large AI investment and infrastructure cost make resource efficiency worth testing. | Measure delivered useful work, total operating cost and network overhead against a credible alternative; validate willingness to pay separately. | | Social | Broad mobile access coexists with unequal AI access and varying digital skills. | Owner-controlled participation, accessible onboarding and clear contribution/consumption reporting; subscribers are not automatically eligible nodes. | | Technological | Distributed execution exists, but link quality and hardware/software diversity constrain it. | Probe, qualify and schedule compatible workloads; preserve a baseline and reject unproven performance promotion. | | Environmental | E-waste and growing data-centre demand motivate hardware reuse. | Measure energy per useful completed task with missing sensor coverage and lifecycle boundaries; account for network and battery wear before making savings claims. | | Legal | Data protection, model licensing, processing authority and use-dependent AI obligations remain relevant. | Scope access and data flows by tenant/workload; assess jurisdiction and applicable obligations. NDA access is not GDPR/AI Act certification. | Policy sources: [European Commission, AI Continent Action Plan](https://digital-strategy.ec.europa.eu/en/factpages/ai-continent-action-plan), [European Commission, AI Act](https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai), [GDPR text on EUR-Lex](https://eur-lex.europa.eu/eli/reg/2016/679/oj/eng). The page deliberately does not prescribe application dates, legal risk classes or a blanket compliance conclusion for unspecified deployments. ## Authorship and media provenance The "Who I am" section identifies Fabrizio Degni as NodeLift's creator and links to his own work. The portrait is reused unchanged from the owner's public PolicyWatcher asset, with no AI restoration, face alteration or invented professional qualifications. The reference author section describes an independent builder maintaining public-interest tools with traceable sources. - [PolicyWatcher About](https://policywatcher.online/about) - [Source author panel](https://github.com/sev7enITA/policywatcher/blob/main/src/app/about/page.tsx) - [Original public portrait](https://raw.githubusercontent.com/sev7enITA/policywatcher/main/public/fabrizio-degni.png) - Portrait SHA-256: `8ce325476b5ad588c2f4b6ce45c8676ca964fd1e26aab767e9c3f349595593e6` - [Personal site](https://www.fabriziodegni.com/) - [Canonical LinkedIn profile](https://www.linkedin.com/in/fdegni/) The topology illustration depicts the target global Hive, four planned regional zones and local/remote peers. Its animated paths are schematic, not live traffic, measurements, deployed capacity or an animation of actual model partitioning.