OctaiPipe
19/2/2024
By:
Donald Harmitt
Snapshot
Founders
Eric Topham
Ivan Scattergood
George Hancock
Founded
2016
Stage
pre-Series A
No. of employees
~20 (as of Dec. 2023)
Headquarters
Europe (London)
Total Funding
~$5.8M (as of Feb. 2024)
Thesis (TL;DR)
Internet of Things (IoT) is all around us and is growing at an unfathomable rate. By 2035, there is expected to be at least 1 trillion IoT devices globally. However, all these devices mean massive volumes of data are being transferred along the network by the second. The untapped potential for all this data, especially private data, is extremely far from being realised. Why? Data security threats, high cloud storage costs, and network constraints are some of the many problems tainting the traditional cloud-based method of extracting value using artificial intelligence (AI) processes such as machine learning.
The critical infrastructure sector is a vital industry which has significant untapped potential within AI, to increase productivity, monitor asset health and performance, while driving sustainable efforts. Being such a crucial part of our ecosytem, critical infrastructure sectors have to be under strict data security processes, which acts as a strong deterrent for implementing more vulnerable centralised AI methods at scale. The current centralised cloud method of using AI applications is a limiting barrier for sectors plagued by data security and privacy regulations, such as healthcare and critical infrastructure, which have the potential to garner attractive return on investment from adopting AI into business operations. An IBM study shows in the UK alone, data breaches cost tech and financial services companies on average £5.3m and £4.9m per year, respectively. Just by using AI these businesses were able to save £1.6m in costs by reducing the breach lifecycle. Then comes federated learning, an innovative machine learning technique which trains AI models by bringing the training models (algorithms) to the data at the Edge (devices), instead of sending the data to the model in the cloud server. Hence, the data never leaves the source device, maximising system resilience and allows companies to align with data compliance and regulations.
OctaPipe is a secure end-to-end artificial intelligence company using decentralised machine learning to transform IoT-dependent critical infrastructure sectors. The OctaiPipe platform enables data scientists and AI engineers to build or purchase pre-packaged solutions and deploy to the Edge, creating a trustworthy network of smart connected IoT devices. OctaiPipe gives their customers the capabilites to produce and deploy models faster, reduce cloud costs and dependency, and increased data security and privacy. Critical insfrastructure companies can scale their IoT network as desired as they can easily build, deploy, and track deployments/experiments within the platform which boasts extra security features via a one-time passkey code. Alongside this, OctaiPipe's platform is compatible with all machine learning frameworks and major cloud platforms and covers the entire ML-Ops lifecycle from data preparation to monitoring.
Company Overview
OctaiPipe:
OctaiPipe (previously The Data Analysis Bureau (T-DAB)) is on a mission to enable trustworthy AI for Internet-of-Things (IoT)-dependent processes to protect critical infrastructures globally, using edge computing and federated learning.
OctaiPipe is an end-to-end edge AI-enabling technology company and the UK's 1st federated learning IoT platform with the goal of transforming the critical infrastructure (e.g. energy, water etc.) sector using decentralised machine learning, to increase security and privacy while reducing cost and the need for network availability.
Problem
IoT refers to physical devices containing sensors which communicate with the outside world (computer systems) usually wirelessly. Typically, these devices work by transferring tons of data from the connected source devices to the cloud (central storage), where they are trained to become more “intelligent” through machine learning. Critical infrastructure is a vital industry to keep our cities and countries up and running, and need to be resilient, secure, and high-performing to ensure this. However, it is a very data-rich sector with automated systems and IoT devices transferring enormous volumes of data by the second, and as a result is exposed to cybersecurity risks and higher costs as the number of connected devices proliferate considerably. Furthermore, machine learning operations are prone to issues such as adversarial attacks which incorporate malicious data into models, data drifting, and hardware failures.
The movement of large volumes of data comes with its drawbacks:
The data becomes prone to security and privacy threats
Centralised storage and transfer of large amounts of data to the cloud is very costly at scale and inefficient
Reliable network connectivity is required to allow efficient transfer of the data which is problematic in remote areas with low latency or reduced network bandwith
Homogenous data is not useful in certain applications and requires pre-processing to meet increasing heterogeneity requirements
Founders
Core Products
Preamble:
Before diving into the technicalities of the OctaiPipe platform, it is useful to understand the AI-related jargons that have been mentioned in the memo so far. Internet of Things (IoT) devices communicate with each other and the external environment, usually via the Cloud ( e.g. Amazon Web Services (AWS)). The exorbitant amounts of data captured from these billions of devices globally is essential to train models (algorithms) to create the buzz word of recent times, artificial intelligence (AI).
How does this work? Traditionally, the data is moved from these IoT devices to the cloud where they are collated and used to train a "global model" which becomes more “intelligent” as it is exposed to data (over and over again) from source devices, by learning from new parameters. Now comes along a new way to do things, a relatively new machine-learning technique called Federated Learning (FL). With FL, instead of moving the raw data to the model in the cloud, the model is moved to the source devices where the data is initially generated, also known as the Edge. Ultimately, the devices independently train a model “locally” using its own data. The updates, improvements, or new learnings are the only thing sent to the cloud, not the raw data. This occurs across a network of intelligent devices for continuous learning. As you may have figured out by now, this decentralised process comes with a multitude of benefits including data privacy and security, reduced costs, higher scalability, time efficiency (related to cloud computing and storage) and reduction of network bandwidth issues during data transfer. OctaiPipe's platform allows learning to occur right at the Edge ensuring trustworthy and resilient systems, and scalable solutions.
The Solution:
OctaiPipe Edge AI Platform
OctaiPipe is a decentralised AI platform for IoT-dependent industrial sectors such as critical infrastructure. OctaiPipe enables the scaling of AI within IoT through tackling machine learning (ML) related issues, and enables the collaborative learning of IoT devices within a trustworthy system. The platform provides clients with the ability to build or purchase pre-packaged ML solutions trained at the Edge. This means critical infrastructure AI engineers and data scientists can deploy models faster, securely and more cost-effectively than the traditional centralised ML method. The models are deployed directly to the Edge reducing reliance on cloud services, while increasing performance and cost efficiency.
The platform combines three components:
Advanced and Private Federated Learning – enhanced security and data privacy whilst reducing cloud computing/storage costs
Distributed and Automated Edge ML-Ops systems – maximises Edge AI productivity and performance
Accelerated Deployment Architecture - pre-built applications and models to accelerate development and deployment of solutions
Traditional Centralised Method
Federated Learning Method
Source: Venture Views
Source: Venture Views
User Journey (Development Environment)
Source: OctaiPipe Demo Day
Key Features:
Cloud computing platform agnostic and portable
Automated and scalable model deployment to the cloud or edge
Federated learning capabilities
Accelerated automated deployment with infrastructure as code
Reliable edge-cloud connectivity
Experimentation and model management capabilities
User Interface
Source: OctaiPipe
Customer Benefits
End-to-end platform covering entire MLOps lifecycle - reduces costly cloud data storage, network, and cloud compute expenses typically involved in cloud infrastructure and data transfer, all in one platform
Drives new revenue streams with secure “on-device intelligence” - allows critical infrastructure companies to enhance existing products and service offerings using AI, opening new and adjacent market opportunities
Enhances security and privacy of customer data - drives greater productivity of IoT devices at the Edge as data is not removed from the source device and is trained locally (one-time passkey code for extra security), saving time that would have otherwise been spent on data transfer
Use-case and platform agnostic, fast, and scalable deployment - applicable to heterogenous data and a wide range of use cases and reduces time-to-value for customers through rapid automated deployment (pre-packaged solutions) on multiple frameworks and cloud platforms.
Reduces reliance on network connectivity and cloud access - decreases the frequency which data needs to be continuously sent back and forth between the device and the cloud (only to update training models)
Improves environmental impact and brand image - reduces carbon footprint due to reduction in data transfer and storage, lowering energy requirement and costs
Business Model
OctaiPipe leverages a B2B Platform-as-a-Service (PaaS) business model which is cloud agnostic and available on major cloud infrastructures such as AWS, Microsoft Azure and Google Cloud.
The PaaS model is an enhanced cloud computing model providing developers with a playground to run their applications and infrastructure without the need to worry about servers, operating systems, and vast overheads. The user only needs to handle the data and application. This model has several pricing models including pay-as-you-go (usage-based) where the client only pays for the amount used, which is prevalent on cloud platforms like Amazon Web Services (AWS). Other pricing models include a subscription-based model where the customer pays on a monthly basis to use the platform and its features. The PaaS model allows scalability as clients can easily-deploy and increase their application requirements as necessary. Similar to the SaaS model this offers personalised package solutions reducing time-to-value for B2B clients and typically shortens the sales cycle for the OctaiPipe team. Cashflow is less predictable for the usage-based model than the subscription based model, however, the strong gross margins of PaaS/SaaS companies allow either model to provide healthy cash-flow with adequate user adoption. The as-a-Service nature of the platform means there is a reduced barrier in the uptake of the service due to less upfront capital requirements, generally leading to increased customer demand and adoption, especially for SMEs who tend to have less cash lying around. Also, the B2B nature of the clients means longer and larger contract values than B2C, and lower price sensitivity compared to their B2C counterparts.
Customers have the option to either purchase pre-packaged solutions or develop their own to deploy and scale on OctaiPipe. This personalisation, alongside the cloud infrastructure and machine-learning framework agnostic nature of the platform, helps to propel user adoption and mitigate future churn due to the tailored nature of the product to customer needs and pain points. The OctaiPipe platform is use case agnostic and is applicable across a multitude of industrial verticals, including but not limited to, energy, utilities, telecoms, automotive, and manufacturing. However, the targeted end-users of the platform are data scientists and AI developers, and critical infrastructure operators/OEMs.
Pricing
OctaiPipe’s platform is available B2B and via a variety of cloud infrastructure platform marketplaces. OctaiPipe's B2B model likely adopts a custom subscription based pricing model, while the cloud marketplace is likely to adopt a pay-as-you-go (similar to AWS) or a similar subscription based model to the B2B channel.
Machine learning Ops competitors such as Edge Delta use a tiered-based custom pricing model. The "professional team" tier allows teams to pay per GB used up to a 10TB limit per month, hence, customers only pay for what they use. EdgeDelta's enterprise package which would be the most similar to OctaiPipe’s highest tiered pricing model package and similar to other competitors such as Weights & Bases, offers flexible pricing based on the specific needs and scale of the client. This pricing model enables flexibility within negotiations to fit budget constraints and technological requirements of the client. However, pricing may vary significantly between clients and more time and resources are required to negotiate and manage contracts.
Competitor Custom Pricing Models
Source: Venture Views Analysis
Strategic Partnerships
Market Snapshot
Forecast Period
2023-30
Market Size (Base Year)
~40B
Market Size (2030)
~340B
Compound Annual Growth Rate (CAGR)
~36%
Estimated Total Addressable Market (TAM)
$20B
Source: Venture Views Analysis (Market size calculations are based on average 3rd party market research reports). The TAM is calculated using both a top-down and bottom-up approach is based on the European and US market only.
The machine learning market is a fast-growing market which is expected to increase at a 36% CAGR until 2030. Propelled by the increased adoption of IoT devices and availability of big data, and the wide number of applicable use cases across a range of industries, machine learning continues to extract value from the enormous amount of data in a fast and efficient manner. This not only increases the competitive advantage of companies using machine learning, but generates costs savings through automated processes.
Within the machine learning market, the federated learning domain is poised to grow as businesses look to implement AI into their operations in a “trustworthy” manner. Federated learning has been on the rise providing the data privacy, security, and cost-efficiency companies vie for over the traditional centralised machine learning method. Sectors such as healthcare, banking & financial services, and industrial IoT already have innovative use cases to tackle key challenges within their respective sectors. This, coupled with various government bodies across the globe (especially Europe and the US) making considerable efforts to safeguard the use of AI, is a key driver for growth within this market. There are a number of stakeholders within the federated learning market including cloud service providers (Amazon, Google etc.), device OEMs, the companies who own the data, and technology providers.
Drivers:
Increased data privacy and security concerns within AI use cases
Accelerated growth in IoT devices, AI, and cloud computing
Growing need for better brand positioning using “trustworthy AI”
Increasing government policies such as the EU AI Act which aims to regulate the development and use of AI
Rising demand for cost-efficient cloud computing from consumers
Untapped potential of high ROI from AI and ML in industrial applications
Competition
The federated learning competitive landscape is highly fragmented with no clear incumbents in the space, likely owing to the nascency of the market. The large majority of emerging players are in the US with some competitors on the other side of the Atlantic in Europe. There are several mature players such as Google and Microsoft who offer open-source platforms to faciliate federated learning and have been pioneers in the space since 2016. What we are starting to see with emerging players is a focus and specialisation within specific industry verticals which are already witnessing vast value from the application of FL, such as healthcare and finance, primarily drug discovery and financial fraud detection, respectively. Majority of early-stage startups in this space covers these two industries as a minimum which serves as no surprise since these are currently two of the most mature industries for the application in terms of uses cases, alongside Industrial IoT. I believe we will continue to see a rising number of new market entrants which will hone in on other verticals such as retail, as customers within these sectors begin to realise the enormous potential to gain competitive advantage with this technology, driving their growth.
What is quite unique about the federated learning landscape is that startups benefit vastly from network effects, i.e. the more customers that use their product, the more valuable it becomes as the accuracy and learnings of the global model increases. Not only this, but economies of scale is also at play as due to the reduced reliance on cloud computing, there is a reduced cost of leveraging AI in IoT devices using FL which is immense and is likely to drive growth within the market. In the medium term, as the potential of FL becomes realised across the broader market players, larger tech corporations will likely acquire some of the early players who have a substantial customer base and unique product offering across a range of industries and use cases. Additionally, larger companies may acquire early-stage companies in a bid to bring FL capabilties in-house through M&A and leverage the additional revenue stream to feed the top line. Companies like OctaiPipe are differentiating through capitalising on the emerging field of TinyML, which optimises ML models for smaller, less powerful devices such as battery powered devices, a capability which was further enhanced in the release of version 2.1 of the OctaiPipe platform. Other players such as FedML are catering to the sigificant rise in large language models (LLMs) to improve the speed of custom AI development and is compatible with prominent LLM libraries such as Hugging Face.
Source: Venture Views Analysis. The list is not exhaustive.
Selected Competitor Highlights:
FedML
FedML is an open source AI platform and foundation model for production allowing companies to commercialise AI models quickly and collaborate with other companies on machine learning tasks, leveraging federated learning for data privacy. FedML has secured a range of contracts across financial services, retail, healthcare, mobility and more. Alongside this, FedML partners with incumbent players such as Amazon and Vmware and is cementing its place in the federated learning market.
Value Propositions:
Open platform on the public cloud allowing simplified collaboration across a broad range of clients anywhere: multinational, cross-city, multi-tenant
Zero code modification required for seamless migration between simulated experiments and real deployment
Unified cross-platform design, supporting smartphones and IoT devices
Apheris
Apheris is a network security provider on a mission to create “collaborative data ecosystems” for consumers to share data and extract insights from other companies’ data without either party worrying about data privacy and security. Focused primarily on the healthcare, pharmaceutical, and manufacturing industry where there are strong regulations regarding sharing consumer data, Apheris has a range of innovative use cases from drug discovery and medical imaging to anti-money laundering and risk modelling. Apheris' customers include the likes of BMW Group, Boston Consulting Group (BCG), and Johnson & Johnson.
Value Propositions:
Seamless integration through cloud and on-premise capabilties, complimented with ML library and code agnosticism, catering to all deployment and customer tech stacks
End-to-end ML-Ops tooling from pre-processing to aggregration and monitoring
Ready-made model implementations reducing building and deployment processes
Detailed audit logs and entreprise-grade authentication enhancing data protection and security
Edge Delta
Edge Delta is an AI observability platform. Edge Delta aims to help their customers spot issues before they happen through monitoring services. The platform integrates with over 50 technologies and this is continuously expanding. Their target end-users are DevOps, security teams, and site-reliability engineers. Edge Delta leverages federated learning to analyse data at the local level without sending it to the central server, producing more personalised and faster results. Edge Delta has a range of big-name customers across a range of industries, a few examples are Panasonic, T-Mobile, and Super League Gaming.
Value Propositions:
Faster mean-time-to detect (MTTD) than traditional tools, allowing customers to address issues before it impacts their end-users
Fast deployment and greater management ease, reducing customers time-to-value and reduces need for technical observability team and costs
Enhanced visibility into logging and anamoly events providing deeper context into abnormal events and reducing time to triage
Key Risks
Adversarial threats to data authenticity
Data authenticity within machine learning processes can be threatened as attackers can tamper with and skew the model parameters affecting the reliability and accuracy of the global model. The data at the edge is susceptible to inaccuracies and bias as attackers can use "stealth" generative adversial network (GAN) to manipulate parameters by generating “realistic” datasets through accessing local models. Poisoning attacks are a key threat to FL models whereby attackers introduce malicious data points, often undetectable from normal data points, which are incorporated into the global model to degrade its task performance. There is also the added risk of collaborators intentionally sabotaging the data model for their competitors. This is both a technological risk which threatens the uptake of the model due to this vulnerability, and a brand risk for OctaiPipe if their customer experiences this type of attack, potentially tarnishing their customer experience and reputation. OctaiPipe continuously improves intrusion resistance through automatic security and penetration testing during development processes. Additionally, version 2.1 of their platform introduced a one-time passkey for the device ensuring only approved devices can interact with the OctaiPipe portal experiments and receive the output trained models, reducing the susceptibility to poisoning attacks at the Edge.
Implementation complexity and lack of technical expertise
With its relative nascency and complex nature, federated learning can be be a barrier for many organisations both large and small who are not equipped or lack the technical expertise for implementation. Companies will require data scientists and AI developers with the relevant knowledge to embed and maintain FL-Ops into their current workflow to get the most value out of the integration. Lack of skilled expertise within data science has been prominent across the UK and other countries which can dampen the uptake of OctaiPipe’s core platform. However, OctaiPipe's pre-packaged solution makes it easy for customers to integrate the platform regardless of skillset and coverage of the end-to-end ML-Ops lifecycle ensures support and seamless integration throughout.
Increasingly crowded competitive landscape globally
Startups are constantly emerging within the federated learning market as the trend of AI and security concerns continue to escalate. The increasingly fragmented competitor landscape will be a barrier for startups aiming to capture ample market share and may even drive down product prices across the market. Market players will need to continuously differentiate their offerings by addressing key pain points and building traction with early customers in sectors such as heathcare, industry, and finance, to leverage early use cases in a bid to attract larger contracts. Educating target customers on the benefits of this complex technology will be vital in order to shorten sales cycle which are often longer in a highly competitive landscape due to plentiful options for customers to ponder on. OctaiPipe has the first to market advantage within the UK landscape and can leverage their pioneering expertise to capture market share as other competitors begin to emerge. With over 20 customers already using their tool, OctaiPipe is well positioned to continue building traction and ward off the effect of emerging competitors.
Customer reluctance to cross-collaboration
Federated learning relies on reliable data from multiple sources to generate a comprehensive model which can be leveraged by each collaborator for their individual use cases and needs. However, as competitors have to put their competitive differences aside to allow their data to be used to train the global model, this will create reluctance amongst competitors who may feel they will lose their competitive advantage. AI companies will need to target sectors primarily affected by data privacy and security regulations such as healthcare and finance, where the value of creating shared models will create more value for each individual company than training their individual data in silos.
Latest Investors
Latest Funding Round Amount: £3.5 million - £3 million (equity), £500k (non-equity)
Lead:
SuperSeed Ventures
Other latest round participants:
Equity:
Forward Partners
Atlas Ventures
D2
Atlas Venture
Martlet Capital
Gelecek Etki VC
Deeptech Labs
Non-Equity:
Innovate UK
Estimated Post-Money Valuations ($USD)
Below are similar competitors who have raised seed funding and their respective post-money valuations. This is used to estimate the expected valuation range of OctaiPipe in the form of a comparable analysis following the most recent funding round (January 2024). Based on these assumptions OctaiPipe's estimated post-money valuation is between $9-25 million.
*Dealroom's estimated valuation range of Bitfount. **New investor equity ownership observed within competitors is used to calculate the estimated post-money valuation range based on latest funding round investment amount.
Source: Venture Views Analysis
Traction/Key Customers
Element Six
Imperial College London
Amazon
Recent News
January 2024 – OctaiPipe raised a £3.5 million pre-series A round, with the aim of further developing their propreitary FL platform and scale its avaliability for critical infrastructure industries