Xunce TokenONE aims to cut AI data prep time from 80% to 30% by 2026

Xunce (3317.HK) launched TokenONE on May 25, 2026, unveiling what it calls the world's first TokenOS operating system designed to transform dormant enterprise data into AI fuel. The product directly attacks the most persistent bottleneck in enterprise AI adoption: data scarcity. Today, companies spend 60-80% of AI project time on data preparation ??cleaning, labeling, and formatting raw information before any model training begins. Xunce targets slashing that to 20-30% by the end of 2026 through a tokenization layer that invokes enterprise data as programmable "data Tokens." The business model shifts from traditional subscriptions to Token-based metering, with vertical-domain Token pricing between $10 and $100 per million Tokens. Xunce plans to co-establish vertical Token Factories in healthcare, manufacturing, finance, and energy. The launch arrives as enterprise AI faces a dual crisis: production agents cause untracked failures, and model guardrails from Meta and Google can be stripped in minutes. TokenONE positions itself as the governance and data infrastructure layer that bridges these gaps. Why this matters now: if Xunce hits its data prep reduction target, it will compress the single largest cost in enterprise AI deployment by more than half, reshaping the economics of AI adoption across every industry.

TokenONE rewrites the data preparation playbook

TokenONE operates as a TokenOS, an operating system layer that sits between raw enterprise data and AI workloads. Instead of forcing data scientists to manually extract, transform, and load data from siloed databases, TokenONE tokenizes the data at source, making it directly invocable by AI agents and models. Each data Token represents a structured unit of enterprise information: a customer record, a manufacturing sensor reading, or a financial transaction that carries its own metadata, access controls, and usage rights. The system replaces the traditional ETL pipeline with a tokenized invocation model, where AI workloads call data Tokens on demand rather than copying entire datasets. This architectural shift attacks the root cause of data prep overhead: the friction of moving and transforming data across incompatible systems. Xunce's target of reducing prep time from 60-80% to 20-30% implies a 3x to 4x improvement in data pipeline velocity. The Token-based metering model also changes the economics: enterprises pay per million Tokens consumed, not per seat or per server hour. For vertical domains, pricing ranges from $10 to $100 per million Tokens, creating a direct link between data value and cost. This structure incentivizes enterprises to tokenize high-value data first, accelerating ROI on the most critical datasets. The TokenOS architecture also embeds governance at the token level, addressing the chaos engineering failures that AI agents currently introduce into production systems without proper tracking. For example, a manufacturing plant using TokenONE can invoke sensor data Tokens directly from its production line, bypassing the weeks of data cleaning that traditional ETL pipelines require.

The $570M revenue model hiding in plain sight

Xunce's shift from subscription to Token-based metering fundamentally rewrites its revenue mechanics. Under the old model, enterprise customers paid a flat annual fee regardless of data consumption. TokenONE introduces consumption-based pricing where revenue scales directly with data usage intensity. At $10-100 per million Tokens, the unit economics depend on volume: a large manufacturing client processing 500 million Tokens monthly would generate between $5,000 and $50,000 per month in Token revenue. Across Xunce's target verticals (healthcare, manufacturing, finance, and energy), the addressable Token volume is enormous. Healthcare alone generates petabytes of patient records, imaging data, and clinical notes daily. If Xunce captures even 1% of the healthcare data tokenization market, the revenue run rate exceeds $570 million annually within three years. That ceiling is realistic: global healthcare AI spending already surpasses $45 billion, and Xunce's tokenization layer lowers the cost of entry for every buyer in that market. The margin profile also improves because TokenONE's marginal cost per Token is near zero once the tokenization infrastructure is deployed. This creates a classic platform economics flywheel: more Tokens drive more AI workloads, which drive demand for more Tokens. The metering model also provides Xunce with real-time visibility into customer adoption and usage patterns, enabling dynamic pricing and upsell opportunities. For investors, the shift to consumption-based pricing transforms Xunce from a cyclical software vendor into a recurring revenue machine with expansion revenue built into every data Token consumed. The key risk is adoption velocity: enterprises must migrate existing data pipelines to the TokenOS, a process that requires trust in the platform's security and reliability.

Who wins and who loses in the tokenization land grab

The competitive landscape splits into three camps: incumbent data infrastructure vendors, AI platform providers, and pure-play tokenization startups. Incumbents like Cisco Systems and Splunk face the most direct threat because their business models depend on data being locked into proprietary formats and licensing schemes. TokenONE's open tokenization layer commoditizes data access, reducing the switching costs that keep customers locked into legacy platforms. Cisco and Splunk must either build their own tokenization layers or risk losing enterprise data workloads to Xunce. AI platform providers like Meta and Google face a different challenge. Their models rely on vast, centralized training datasets, but enterprise AI requires access to proprietary, real-time data that these platforms cannot easily reach. TokenONE creates a bridge between enterprise data and any AI model, reducing the moat that Meta and Google have built around their training data advantages. For startups like Cerebras Systems and Astera Labs, TokenONE is complementary: they sell hardware acceleration for AI workloads, and faster data preparation drives demand for their chips. The biggest losers are data integration middleware vendors that charge per-connection or per-record fees for ETL services. TokenONE's tokenized invocation model eliminates the need for most ETL middleware, directly threatening companies that have built businesses on data plumbing. The wildcard is whether hyperscalers like Alibaba or Amazon build competing tokenization layers, which would commoditize the market and compress Xunce's pricing power.

Second-order effects on hyperscalers, fabs, and enterprise buyers

TokenONE's compression of data preparation time from 80% to 30% ripples through the entire AI supply chain. Hyperscalers running AI training clusters currently allocate massive compute capacity to data preprocessing jobs, sometimes 40-50% of total GPU cycles. If TokenONE reduces prep time by 50 percentage points, hyperscalers can repurpose that compute capacity for inference or training, effectively increasing their AI revenue per GPU by 40-60%. This drives demand for more GPUs and AI accelerators from suppliers like Astera Labs and Arm Holdings, whose chips power the inference and training workloads that will fill the freed-up compute cycles. For fab operators, the net effect is ambiguous: more AI chips ordered, but lower demand for data-center networking gear because tokenized data moves more efficiently across networks. Enterprise buyers, particularly in healthcare and finance, benefit from faster AI deployment cycles and lower total cost of ownership. A hospital system that previously spent 18 months preparing data for a diagnostic AI model can now deploy in six months, accelerating ROI on AI investments. The downstream effect on regulators is significant: faster AI deployment means faster accumulation of production incidents, which will pressure agencies to establish governance frameworks for tokenized data. The Stanford Trustworthy AI Research Lab's finding that model-level guardrails are insufficient against fine-tuning attacks underscores the need for token-level governance, which TokenONE provides by embedding access controls and audit trails into each data Token.

What the TokenONE launch signals about AI market direction

Xunce's TokenONE launch is a strategic signal that the AI industry is pivoting from model-centric to data-centric architectures. The past two years have been dominated by model scaling: bigger LLMs, more parameters, and longer context windows. But the marginal returns to model scaling are diminishing, and the bottleneck has shifted to data availability and quality. TokenONE represents a bet that the next phase of AI value creation will come from unlocking dormant enterprise data rather than building larger models. This aligns with broader market trends: SpaceX's anticipated IPO at a $1.75 trillion valuation, followed by OpenAI and Anthropic, reflects investor appetite for AI infrastructure plays rather than pure model companies. The mixed performance of recent AI IPOs (Astera Labs up 700% since 2024, but Figma down 35% from its $33 IPO price due to generative AI commoditization fears) suggests that investors are rewarding companies that own proprietary data or data infrastructure rather than those that simply apply AI to existing products. Xunce's Token-based metering model also signals a shift toward usage-based pricing in enterprise software, moving away from seat-based licensing. This mirrors the broader trend in cloud computing where consumption models dominate. The policy implication is clear: regulators must prepare for a world where enterprise data becomes programmable and tradeable as Tokens, raising questions about data sovereignty, privacy, and cross-border data flows. Xunce's vertical Token Factories in healthcare and finance will test these regulatory boundaries directly.

The next 12 months will determine whether TokenONE becomes the standard for enterprise data tokenization or a footnote in AI history. Xunce must execute on its vertical Token Factory partnerships in healthcare, manufacturing, finance, and energy while convincing skeptical CIOs to migrate from proven ETL pipelines to an unproven tokenization layer. The company's ability to hit its 20-30% data prep time target by end of 2026 will be the single most important metric for investors. If successful, TokenONE will trigger a wave of enterprise AI deployments that were previously stalled by data scarcity, compressing AI project timelines from years to months. The tokenization model will also attract competition: expect Cisco, Splunk, and hyperscalers to launch competing offerings within 12-18 months. Xunce's first-mover advantage in vertical Token Factories gives it a narrow window to lock in customers and build switching costs through embedded tokenized workflows. The broader implication is that the AI industry's center of gravity is shifting from model builders to data infrastructure providers, and TokenONE is the most aggressive bet yet that data, not models, will determine who wins the next phase of the AI revolution.