FortifAI NOL8 vs Google RE2: 118x Data Cost Edge [2026]

A small Australian hardware company just published a benchmark claim that, if it holds up under scrutiny, reframes how enterprises should think about the cost of filtering and classifying data at AI scale. On October 12, 2026, FortifAI Limited (ASX: FTI) released results from an independently reviewed benchmark showing its NOL8 data plane processed between 4.1 times and 118 times more data per infrastructure dollar than Google’s widely used RE2 regular-expression engine, according to a report carried by TradingView News and corroborated by TipRanks. The gap, FortifAI says, widens dramatically as the complexity of the filtering rules applied to the data increases.

The announcement lands at a moment when AI infrastructure spending is under intense cost scrutiny across the industry, from the AI pricing war triggered by Claude Haiku 5.5’s price cut to ongoing debates over container orchestration costs on AWS. If NOL8’s reported numbers survive independent replication outside the company’s own testing environment, it would mark one of the more significant data-plane cost claims of 2026. This article breaks down what was actually reported, what remains unverified, how the figures compare with the rest of the AI infrastructure market, and what it could mean for enterprises currently budgeting for data classification and filtering workloads.

Google · Preferred Sources

Don't miss new tech stories on Google

Add TrendinTech once in the Google app and our stories appear in your news suggestions.

Add Now

What FortifAI Actually Announced on October 12, 2026

According to the report published by TradingView News, FortifAI Limited disclosed benchmark results comparing its NOL8 product, also referred to in company materials as the Nol8 AI Data Plane, against Google’s RE2 engine. RE2 is a regular-expression library originally built at Google and widely deployed across the industry for high-throughput pattern matching, data filtering, and text classification tasks. It is not a consumer-facing Google product but rather an open-source engine embedded inside countless data pipelines, security tools, and content filters.

The headline figure from the benchmark is a data-processed-per-infrastructure-dollar comparison. Across six separate test runs, NOL8 reportedly delivered between 4.1 times and 118 times more data processed per dollar of infrastructure spend than RE2, according to the TradingView report. Both TradingView and TipRanks described the testing as independently reviewed, though the full methodology, the hardware configurations used in every run, and a publicly accessible primary report were not available in the coverage reviewed for this article.

What makes the claim notable is not the existence of a performance gap between a specialized data-processing appliance and a general-purpose regex engine, that would hardly be surprising, but the scale of the reported gap and how it behaves as rule complexity increases. That relationship, more than the raw multiplier, is the part of this story worth examining closely.

The Numbers: Cost Per Gigabyte-Second Across Rule Complexity

The benchmark tested performance across a range of filtering-rule complexity, from 128 rules up to 5,000 rules, according to the TradingView report. This matters because rule complexity is a reasonable proxy for how sophisticated a real-world data classification or content-filtering workload is. A system enforcing 128 simple rules is a fundamentally different computational problem than one evaluating 5,000 rules against every byte of incoming data, and that difference is exactly where the reported cost curve diverges.

At the low end of the complexity range, 128 rules, the reported advantage for NOL8 was 4.1 times the data processed per dollar compared with RE2. At the high end, 5,000 rules, the reported advantage grew to 118 times, according to the same report. In dollar terms, NOL8 maintained an approximate cost of US$2 per gigabyte per second of throughput per hour across the reported trials. RE2’s reported cost, by contrast, climbed from roughly US$8 per gigabyte per second of throughput per hour at the simplest rule sets to approximately US$240 at the most complex tier tested, based on figures cited in both the TradingView and TipRanks coverage.

That is a cost curve that bends sharply upward for RE2 as complexity rises, while NOL8’s reported cost stayed comparatively flat. If accurate, that behavior suggests RE2’s architecture struggles to scale efficiently once rule sets become dense, a known characteristic of general-purpose regex engines handling large pattern libraries, while NOL8’s approach appears designed specifically to absorb that complexity without a proportional cost penalty.

Rule Complexity Tier	NOL8 Reported Cost (per GB/s/hr)	Google RE2 Reported Cost (per GB/s/hr)	Reported Data-Per-Dollar Advantage
128 rules (low complexity)	~US$2	~US$8	4.1x
Mid-range complexity	~US$2	Rising progressively	Between 4.1x and 118x
5,000 rules (high complexity)	~US$2	~US$240	118x

Additional performance figures cited in the TipRanks report described NOL8 sustaining approximately 950 MB/s of throughput, with latency of 3 to 5 milliseconds for up to 98% of workloads, in tests described as running on AWS hardware. Under the same heavy rule-set conditions, RE2’s throughput reportedly dropped to around 2 MB/s. That throughput gap, roughly 475 times in raw MB/s terms under those specific conditions, is separate from the data-per-dollar cost figures above and should not be conflated with the 118x cost-efficiency number, which was derived from the six-test cost comparison rather than this single throughput measurement.

Who Is FortifAI and What Is NOL8

FortifAI Limited trades on the Australian Securities Exchange under the ticker FTI. The company’s flagship product, NOL8, is marketed as an AI data plane, infrastructure designed to sit between raw incoming data and the AI systems or compliance pipelines that need to filter, classify, or govern that data before it is used. The company has previously described NOL8 as using FPGA acceleration, hardware that is reconfigured at the circuit level for a specific workload rather than running as general-purpose software on a CPU.

That FPGA-based design is the most plausible technical explanation for why NOL8’s reported costs stay flat as rule complexity rises while RE2’s reported costs climb steeply. General-purpose regex engines like RE2 run as software on CPU cores, where every additional pattern rule adds computational overhead that scales with the size and complexity of the rule set. A purpose-built hardware data plane, if implemented well, can parallelize that pattern-matching work in silicon rather than in sequential software execution, which would explain a cost curve that resists the kind of complexity penalty RE2 reportedly experienced.

What Remains Unverified in This Benchmark

It is worth being direct about the limits of what has actually been confirmed. The coverage reviewed for this story does not establish the complete benchmark methodology, the exact hardware configurations used across every one of the six test runs, the test code itself, the full identity and credentials of whoever conducted the independent review, or a publicly accessible primary report that other researchers or engineers could use to reproduce the results. “Independently reviewed” and “independently verified” are the terms used in the coverage, but without a public methodology document, outside engineers cannot currently check FortifAI’s work against their own infrastructure.

There is also a history of escalating performance claims from FortifAI that predate this specific cost benchmark and should not be confused with it. Earlier 2026 announcements from the company, including coverage from Bellwire News in August and FinerMarketPoints in May, cited separate figures such as a 10x data-throughput advantage in multi-continent testing, a claimed throughput advantage exceeding 200,000 times under AI-grade workloads against RE2, sustained throughput of 1,500 MB/s, and the suggestion that a single NOL8 FPGA appliance could replace the equivalent compute capacity of up to 60,000 CPUs. None of those figures are confirmed as part of the October 12, 2026 cost-per-dollar benchmark described above. They come from earlier, separate test rounds and should be treated as a distinct, prior chapter of FortifAI’s public claims rather than evidence supporting the newest numbers.

That pattern, a string of escalating benchmark claims released over several months without a stable, consistently cited figure, is itself useful context for readers evaluating how much weight to put on any single announcement from the company. A buyer evaluating NOL8 for production use should ask FortifAI directly for the raw methodology behind the October 12 figures before treating 118x as a number that will hold in their own environment.

Why RE2 Is the Benchmark of Choice

Google’s RE2 is a sensible, if slightly unusual, choice of comparison target. It is not Google’s flagship AI infrastructure, nor is it a product Google actively markets for AI data pipelines specifically. It is a general-purpose, linear-time regular-expression engine that avoids the catastrophic backtracking problems found in some other regex implementations, and it has become a de facto industry standard embedded inside security tools, log processors, data loss prevention systems, and content filters precisely because it is fast, predictable, and free to use.

Choosing RE2 as the comparison point lets FortifAI frame NOL8 against something enterprise engineers already understand and likely already run somewhere in their stack, rather than against an obscure or proprietary competitor. It also sets a relatively low bar in one sense: RE2 was never designed as dedicated AI-scale data-plane hardware, it is CPU-bound software built for general pattern matching. A more rigorous comparison for evaluating NOL8’s real-world value would include dedicated hardware-accelerated alternatives and purpose-built data governance platforms, not just a general software library running on commodity CPUs.

Competitive Landscape: Where NOL8 Sits Among Data Infrastructure Options

Enterprises filtering, classifying, or governing large volumes of data for AI workloads today generally choose among a handful of approaches: general-purpose CPU-based regex and pattern-matching libraries like RE2, GPU-accelerated data pipelines, cloud-native managed services, and increasingly, specialized FPGA or ASIC-based appliances like NOL8. Each category trades off differently on cost, latency, flexibility, and operational complexity.

Approach	Typical Strength	Typical Weakness	Cost Behavior at Scale
CPU-based regex (e.g. Google RE2)	Free, mature, widely integrated	Throughput collapses under complex rule sets	Cost rises sharply with rule complexity, per the Oct. 12 benchmark
FPGA-accelerated data plane (NOL8)	Reported flat cost curve regardless of rule complexity	Newer, less independently documented, requires dedicated hardware	Reported ~US$2 per GB/s/hr across all tiers tested
GPU-accelerated pipelines	Highly parallel, flexible for broader AI workloads	Expensive GPU capacity, often shared with model training/inference demand	Variable; tied to GPU market pricing and availability
Cloud-native managed filtering services	No hardware to manage, elastic scaling	Per-request or per-GB pricing can compound at high volume	Scales with usage-based cloud billing

This positions NOL8 less as a direct head-to-head rival to Google RE2 in the traditional sense, since RE2 is free open-source software rather than a commercial product Google sells, and more as a challenger to the broader status quo of running data filtering on general-purpose CPU infrastructure at all. The real competitive question FortifAI is raising is whether dedicated hardware acceleration for data governance and filtering, a category that has historically been a niche compared with GPU acceleration for AI training and inference, deserves a bigger seat at the infrastructure table as AI-grade data volumes grow.

Historical Context: The Rising Cost of Data Governance at AI Scale

Data filtering and classification have historically been treated as a relatively low-cost, background function, something handled by a regex library or a basic rules engine bolted onto a larger pipeline. That assumption made sense when data volumes were modest and rule sets were simple. The rise of AI-grade workloads, where every byte flowing into a training set, a retrieval-augmented generation pipeline, or a compliance system may need to be filtered, classified, or governed against increasingly complex and evolving rule sets, has changed that calculus.

This is part of a broader pattern across the infrastructure market in 2026, where cost efficiency at scale has become a central competitive battleground rather than a secondary concern. The same dynamic is visible in Kubernetes cost-monitoring tooling gaining adoption as container spend scrutiny increases, in serverless pricing wars among AWS, Azure, and Google Cloud, and in the AI model layer itself, where frontier model pricing gaps of 3x or more are now a routine part of procurement decisions. FortifAI’s benchmark, whatever its ultimate accuracy, is tapping into a real and growing enterprise anxiety: AI-scale data operations that once ran on commodity infrastructure can quietly become one of the largest line items in a cloud bill once rule complexity and data volume both climb.

Market Impact: What This Means for FortifAI and the Sector

As a small-cap ASX-listed company, FortifAI’s benchmark announcements function partly as technical marketing and partly as investor communication, a dual purpose that is common for small technology companies seeking to build credibility with both customers and capital markets simultaneously. The TipRanks report characterized the announcement as part of FortifAI’s ongoing effort to demonstrate commercial validation of NOL8’s data economics against an industry-standard benchmark.

For the broader data infrastructure sector, the more interesting signal is not whether FortifAI specifically succeeds commercially, but whether the underlying premise, that FPGA-accelerated data planes can meaningfully undercut CPU-based filtering costs at high rule complexity, gains traction with larger infrastructure vendors. If true even directionally, it would suggest an underexplored opportunity for hardware-accelerated data governance that larger players in cloud infrastructure and AI tooling have not yet aggressively pursued, since most hardware acceleration investment to date has concentrated on model training and inference rather than data filtering and classification.

How Enterprises Should Read a Vendor-Commissioned Benchmark

Benchmark announcements from a vendor about their own product, even when described as independently reviewed, warrant a specific kind of skepticism that differs from skepticism toward a neutral third-party study. The questions worth asking before acting on a number like 118x include: what workload was actually tested, and does it resemble the buyer’s own data; what hardware baseline was RE2 run on, since an under-provisioned CPU comparison would inflate the apparent gap; was the “independent review” conducted by a party with no commercial relationship to FortifAI; and is there a public, reproducible methodology that a buyer’s own engineering team could run against their own data before committing budget.

None of this means the underlying claim is false. FPGA acceleration genuinely can outperform CPU-based pattern matching dramatically under the right conditions, and the physics of parallel hardware versus sequential software execution make a large gap at high rule complexity plausible in principle. But plausible in principle and verified in an enterprise’s own production environment are two different standards, and only the second one should drive a procurement decision.

Technical Breakdown: Why Rule Complexity Breaks CPU-Based Regex Engines

To understand why the reported cost gap widens so sharply between 128 and 5,000 rules, it helps to understand how regex engines like RE2 actually process pattern matching. RE2 compiles patterns into finite automata and evaluates input data against them in linear time relative to the input length, which is RE2’s main design advantage over regex engines that can suffer catastrophic backtracking. But linear time relative to input length does not mean the cost is independent of rule-set size. As the number of distinct patterns a system must check grows into the thousands, the combined automaton or the overhead of evaluating many separate patterns against the same data stream grows the computational burden on every CPU cycle, which is reflected in the climbing cost-per-gigabyte-second figures reported in the benchmark.

A hardware-based approach, by contrast, can in principle evaluate many patterns simultaneously across parallel circuit paths rather than sequentially stepping through software logic. That is the structural reason an FPGA-based system’s cost curve would plausibly stay flatter as rule complexity increases, assuming the hardware is designed and configured well for that specific task. Whether NOL8’s implementation achieves that in practice at the scale and consistency FortifAI describes is exactly the open question that independent, reproducible testing outside the company’s own benchmark would need to settle.

Predictions: Where This Story Goes Next

Expect scrutiny from independent analysts and engineers over the coming weeks, given the absence of a fully public methodology; a lack of third-party replication would likely dampen enthusiasm even if FortifAI’s underlying numbers hold up internally.
FortifAI is likely to continue issuing incremental benchmark updates tied to investor communications, following the pattern of its 2026 announcement cadence; expect further data points before year-end rather than a single definitive report.
If the cost-efficiency claims attract attention from larger cloud or AI infrastructure vendors, watch for competitive responses framed around hardware-accelerated data governance, an area that has received far less public investment than GPU acceleration for model training and inference.
Enterprises with AI-grade data filtering needs at high rule complexity are likely to request direct benchmarking against their own workloads before adopting NOL8, which would be the real test of whether the 118x figure generalizes beyond FortifAI’s own test conditions.
Expect continued conflation online between this October 12 cost-per-dollar benchmark and FortifAI’s earlier, separate throughput claims from mid-2026, since the headline multipliers (4.1x to 118x versus the unconfirmed 200,000x figure) are easy to mix up without careful reading of what each number actually measures.

What Buyers Evaluating Data Plane Infrastructure Should Do Now

For technical teams currently budgeting AI-scale data filtering, classification, or governance infrastructure, this benchmark is a useful prompt to re-examine assumptions rather than a signal to switch vendors immediately. The first step is auditing current rule-set complexity against actual data volume to understand whether a CPU-based system like RE2 is already showing the kind of cost escalation described in the FortifAI benchmark, since that is a measurable, internal data point that does not depend on trusting anyone else’s numbers. The second step is requesting a proof-of-concept benchmark from any vendor, FortifAI included, run against representative production data rather than a vendor-selected test case. The third is factoring in operational overhead: FPGA-based appliances require different deployment, monitoring, and maintenance skill sets than software running on standard cloud compute, and that transition cost belongs in any total-cost-of-ownership comparison, not just the per-gigabyte-second figure.

The Broader Signal: Hardware Acceleration Returns to the Data Layer

Much of the AI hardware conversation in 2026 has centered on GPUs, training clusters, and inference chips, a focus reflected in ongoing memory pricing pressure between AMD and Nvidia and in continued debate over GPU allocation strategy. FortifAI’s benchmark is a reminder that the data layer feeding those AI systems, the filtering, classification, and governance work that happens before data ever reaches a model, is its own infrastructure category with its own cost dynamics, and one that has received comparatively little specialized hardware investment to date.

Whether NOL8 specifically becomes a meaningful commercial product or remains a niche ASX small-cap story is almost beside the point for the broader industry. The pattern it is pointing at, data governance costs scaling non-linearly with rule complexity on general-purpose CPU infrastructure, is a real and under-discussed problem as AI data volumes keep growing. Expect more vendors, large and small, to start making similar cost-efficiency arguments over the next year, which makes the skepticism outlined above about methodology and independent verification more relevant, not less, as these claims multiply.

Frequently Asked Questions

What is FortifAI NOL8?
NOL8, also referred to as the Nol8 AI Data Plane, is a data-processing product from FortifAI Limited (ASX: FTI), an Australian technology company. It is described in company materials as using FPGA hardware acceleration to filter and classify large volumes of data for AI and compliance workloads.

What did the October 12, 2026 benchmark actually show?
According to a report from TradingView News, corroborated by TipRanks, NOL8 processed between 4.1 times and 118 times more data per infrastructure dollar than Google’s RE2 engine across six tests, with the advantage increasing as filtering-rule complexity rose from 128 to 5,000 rules.

Is Google RE2 a Google product people use directly?
No. RE2 is an open-source, general-purpose regular-expression engine originally developed at Google. It is not a consumer product but is widely embedded in data pipelines, security tools, and content-filtering systems across the industry, which is why it functions as a common benchmark reference point.

Is this benchmark independently verified?
The coverage reviewed describes the testing as independently reviewed, but the full methodology, complete hardware configurations, test code, and a publicly accessible primary report were not available in that coverage. The identity and credentials of the independent reviewer were also not fully established in the reports examined for this article.

How does this relate to FortifAI’s earlier claims of a 200,000x advantage?
Those figures come from separate, earlier 2026 announcements and are not confirmed as part of the October 12, 2026 cost-per-dollar benchmark. They should be treated as a distinct and earlier set of claims rather than evidence supporting the newest reported multipliers.

What is FortifAI’s stock ticker?
FortifAI Limited trades on the Australian Securities Exchange under the ticker FTI.

Should enterprises switch from RE2-based pipelines to NOL8 based on this benchmark alone?
Not without independent, reproducible testing against their own data and rule sets. Vendor-reported benchmarks, even when described as independently reviewed, should be validated with a proof-of-concept against representative production workloads before any procurement decision.

Why does the cost gap widen so much at higher rule complexity?
CPU-based regex engines like RE2 process pattern matching largely in sequence, so computational overhead grows as the number of distinct rules grows. Hardware-based approaches like FPGA acceleration can, in principle, evaluate many patterns in parallel, which would explain a flatter reported cost curve as complexity increases, though this has not been independently confirmed outside FortifAI’s own testing.

FortifAI Claims 118x Data Cost Edge Over Google RE2 [2026]