Ottawa Folds Shared Services Canada and the Canadian Digital Service Into One Agency and Puts a Former Google CFO in Charge
Prime Minister Mark Carney launched Digital Transformation Canada on September 3 and named Patrick Pichette as its chief executive. The new organization merges Shared Services Canada, the Canadian Digital Service from Employment and Social Development Canada, and selected functions from the Treasury Board of Canada Secretariat and Public Services and Procurement Canada. It reports to Joël Lightbound, Minister of Government Transformation, Public Works and Procurement.
The stated priorities are to scale shared digital solutions across government to cut cost and duplication, give public servants modern and secure technology tools, and use federal purchasing deliberately to support Canadian technology companies. The agency will run a fellowship model that brings private sector specialists into government for short assignments in AI and digital work. Pichette was Google's chief financial officer from 2008 to 2015, chaired Twitter's board until its sale, spent more than eight years as a partner at Montreal based Inovia Capital, and held senior roles at Bell and McKinsey. No budget or transition timeline was published with the announcement.
- Shared Services Canada, the Canadian Digital Service, and TBS and PSPC functions consolidated into one agency
- Mandate spans how the federal government develops, purchases, and uses AI and technology
- Fellowship model for short term private sector assignments; increased procurement from Canadian firms is an explicit goal
Enterprise Impact: For Canadian technology vendors, the buyer just changed shape. Procurement decisions that were spread across Shared Services, TBS policy, and departmental purchasing are being pulled toward a single organization led by someone who ran finance at a company that builds platforms rather than buys them, and the mandate says out loud that Canadian firms should win more of the work. Expect a shift from bespoke departmental systems toward shared platforms with a shorter list of larger contracts, which favours vendors who can show a product running at scale over those selling services by the hour. Companies with talent to lend should look at the fellowship route; it is a way to be inside the design conversation before the requirements are written.
Source: Prime Minister of CanadaOpenAI, Anthropic, and 21 Others Sign Canada's Data Centre Principles as Two Thirds of Canadians Say Not Near Me
Twenty three companies signed Canada's Responsible Data Centre Development Principles at the launch in Markham on September 3, among them OpenAI, Anthropic, Google, Microsoft, Meta, Amazon Web Services, Cohere, Bell, Telus, OVHcloud, Hypertec, ThinkOn, and Beacon Data Centres. Signatories commit to paying the electricity costs their projects create and contributing to new supply where needed, minimizing water use with approaches such as closed loop cooling, delivering local jobs, skills training, research partnerships, and compute access, respecting local approval processes with independently verifiable impact information, and contributing strategic value through investment, supply chain participation, and compute capacity.
The framework is voluntary and carries no penalties. It arrives as Angus Reid polling shows more than two thirds of Canadians oppose a data centre near their homes, and as the government pairs it with its sovereign compute strategy and its stated aim of reducing reliance on infrastructure outside the country. Minister Evan Solomon summarized the expectation as projects paying the costs they create, protecting local resources, and delivering lasting value to Canadians.
- Hyperscalers, Canadian carriers, and domestic operators all on the signatory list
- Commitments cover electricity cost, water, community benefit, transparency, and strategic value
- Voluntary framework aimed at unblocking municipal approvals amid public opposition
Enterprise Impact: Read the electricity principle as a pricing signal. Operators that must not shift power costs to ratepayers will recover them from customers, so Canadian hosted AI capacity is likely to carry a premium over regions that subsidize it, and the trade for that premium is jurisdiction, latency, and a defensible answer on sovereignty. Enterprises planning where to run AI workloads over the next three years should model Canadian capacity as available but not cheap, and should push providers for concrete commitments on when signed principles turn into operating megawatts near their users. The community and compute access commitments are also worth reading as a partnership channel: universities, municipalities, and regional businesses near approved sites will have access programs to apply to.
Source: BetaKitQuandela Puts the First Photonic Quantum Computer Into Canada's Quantum Computing Sandbox
Quandela has deployed a photonic quantum computer on the Quantum Computing Sandbox, a research and development program run by CMC Microsystems within the federally funded FABrIC network that supports Canada's semiconductor and quantum industries. It is the first and only photonic system on the sandbox, which already offers access to machines from IonQ, Quantinuum, QuEra, and Rigetti, all US based, and to IBM's superconducting system operated by PINQ in Sherbrooke, Quebec.
Participating small businesses and researchers gain access to the Quandela system through the sandbox, with the company providing technical and practical assistance on projects it judges to have high potential. Quandela emphasizes that hosting data on Canadian infrastructure addresses data sovereignty concerns for participants.
- Photonic architecture joins trapped ion, neutral atom, and superconducting options on one Canadian access program
- CMC Microsystems operates the sandbox under the FABrIC network
- Canada based hosting positioned as a sovereignty feature for research data
Enterprise Impact: The practical value of the sandbox is not a quantum advantage, it is cheap, supervised experimentation across four different hardware approaches without buying anything. Enterprises in logistics, materials, finance, and energy that expect quantum to touch their optimization or simulation workloads should nominate one team to run a bounded problem through the program and report back on which architecture behaved best, because the answer will inform vendor conversations years before a purchase. The sovereignty framing is also a reminder that quantum access will be evaluated on where the data sits, the same way cloud AI is now.
Source: BetaKitVancouver's Mundo AI Raises $24 Million USD to Build the Data Layer for Machines That See and Hear
Mundo AI announced a $20 million USD Series A led by GreatPoint Ventures, along with a previously undisclosed $4 million USD seed round, bringing total funding to $24 million USD, about $28 million CAD for the Series A alone. Y Combinator, E12 Ventures, and Next Frontier Capital also participate. The Vancouver company builds datasets, evaluations, and research infrastructure for training AI systems on sensory information such as audio and video, describing its mission as the data layer for perceptual intelligence.
Founded in 2024 by Jason Liao, Naijide Anwaer, Garreth Lee, and Kenneth Wu, Mundo started in multilingual training data and pivoted toward what it calls the missing half of AI, the ability to understand unstructured information from the physical world. The funding will expand the 30 person team across research, engineering, and operations.
- $20 million USD Series A plus $4 million USD seed; GreatPoint Ventures leads
- Datasets and evaluations for audio, video, and other perceptual inputs
- 30 person Vancouver team growing across research, engineering, and operations
Enterprise Impact: Text is largely solved as a training input; the next capability gains come from models that interpret cameras, microphones, and sensors, and that is where enterprise data is most abundant and least organized. Manufacturers, logistics operators, retailers, and utilities sitting on years of video and sensor archives should treat them as an asset to catalogue now, with retention, consent, and labelling questions answered before a vendor asks. A Canadian company building the evaluation layer for perceptual models is also a useful reference point when assessing whether a vendor's claims about vision or audio performance are measured against anything real.
Source: BetaKitRegina's ASI Engineering Gets $980,000 From PrairiesCan to Put AI Into Municipal Asset Management
ASI Engineering has received $980,000 from Prairies Economic Development Canada through the Business Scale up and Productivity Program to integrate AI into its ASI Software platform, which serves organizations managing public assets such as roads, municipal buildings, utility systems, and community facilities. The platform holds asset information in a cloud registry with dashboards that summarize condition, track budgets, and identify funding gaps and infrastructure deficits.
The AI work targets continuous asset monitoring and reporting, with climate adaptation modelling planned as a further capability. The company intends to expand into the United States and Europe over the next two years.
- $980,000 federal contribution to add AI monitoring and reporting to an existing asset platform
- Customers are municipalities and public bodies with roads, buildings, and utilities to maintain
- Climate adaptation modelling next; US and European expansion planned
Enterprise Impact: This is what most successful AI adoption looks like: a vertical platform that already holds the data adds inference to it, and customers get better decisions without a new system. Public sector and infrastructure heavy organizations should look at the platforms they already run for the same opportunity before funding a standalone AI project, because the integration and change management are already paid for. Vendors with a niche platform and a Prairies footprint should note the funding instrument; scale up and productivity programs are underused for exactly this kind of feature investment.
Source: BetaKitProtein Industries Canada Puts $2.2 Million Into Two AI Projects for Crop Disease Forecasting and Grain Bin Measurement
Protein Industries Canada announced more than $2.2 million across two agricultural AI projects. The first, a $2.6 million project with $1.1 million from the cluster, brings TerraVision360, Metos Canada, and Rocky Mountain Equipment together on a tool that combines crop data, disease information, and local weather to forecast Ascochyta blight risk at regional and individual farm level. The second, a $2.4 million project with $1.1 million from the cluster, pairs SuperGeoAi Technology with Southview Farms on an AI enabled drone carrying LiDAR to measure grain volumes in bins and silos.
The grain project removes the need for workers to climb and inspect bins while producing documented records of grain levels. Saskatchewan's Secretary of State for Rural Development framed both as practical tools to lower costs, improve safety, and raise productivity.
- Fungal disease risk forecasting from crop, disease, and localized weather data
- Drone mounted LiDAR for grain inventory without bin entry
- Industry partners co invest alongside the cluster in both projects
Enterprise Impact: Both projects replace a manual measurement or a judgement call with an instrumented one, which is the reliable formula for AI return in operations businesses. The grain project in particular converts a safety risk into a data record, and that pattern, where the AI's by product is documentation nobody had before, tends to unlock insurance, financing, and compliance value beyond the original use case. Operations leaders outside agriculture should ask where their own staff climb, count, or estimate, and whether a sensor plus a model would produce a record worth more than the labour saved.
Source: BetaKitNvidia Confirms a $12.9 Billion Deal for Hugging Face and Promises the Hub Stays Open to Every Chip and Cloud
Nvidia confirmed it will acquire Hugging Face for $12.93 billion. The platform hosts three million models, one million applications, and 500,000 datasets, and is used by more than 18 million developers. Jensen Huang pledged that Hugging Face will remain an open platform for the entire AI ecosystem, that developers keep their choice of models, frameworks, cloud providers, and compute, and that Nvidia hardware will not be required to build on or deploy through the service. Hugging Face chief executive Clem Delangue said the deal brings the compute, support, collaboration, and visibility needed to scale.
Nvidia has contributed more than 500 models and 250 open datasets to the platform. Hugging Face rejected a $500 million offer from Nvidia last year and recently reported about $150 million in annualized revenue. No closing timeline or regulatory review details were given. The strategic logic is an open ecosystem that Nvidia controls and can optimize for its silicon, plus a channel to package spare accelerator capacity for enterprise customers.
- $12.93 billion for the largest open model and dataset hub, roughly 86 times its reported revenue
- Public commitment to hardware and cloud neutrality for developers
- Path for Nvidia to sell compute directly through the platform
Enterprise Impact: If your model supply chain runs through Hugging Face, and for most teams using open weight models it does, the neutrality pledge is now a dependency to manage rather than a fact to assume. Mirror the models and datasets you rely on internally, record their licences and checksums, and make sure your deployment pipeline does not depend on hub features that could become Nvidia preferred over time. The upside is real too: expect tighter integration between hosted models and Nvidia inference stacks, and a simpler route to renting capacity for fine tuning without a hyperscaler contract. Treat this as a moment to formalize your open model governance, because the platform that made it informal just changed hands.
Source: TechCrunchSam Altman Calls the Astra Rollout Messy as Enterprise Customers Discover That Announced and Available Are Different States
OpenAI launched Astra on September 3 to members of its Daybreak cybersecurity program first, describing it as its strongest model for software engineering and a new frontier for computer and browser use. Broader access lagged the announcement: Pro, Enterprise, and Business Premium customers gained access through the Work and Codex products by September 5, API access opened to developers, and Plus and Business users were told to expect access within days. On September 7 chief executive Sam Altman apologized for what he called a messy rollout and said broad availability to API customers and subscribers should begin in the near future.
Greyhound Research summarized the lesson as four separate states, announced, available, entitled, and production ready, and advised enterprises to verify actual access rather than assume a uniform rollout. Gartner advised CIOs to strengthen governance controls and evaluate on specific use cases rather than early capability claims. The model's headline strengths are code, terminal task execution, and vulnerability discovery, which is why the security program received it first.
- Staggered access over at least a week across program, subscription, and API tiers
- Public apology from the chief executive for the rollout sequence
- Analysts advise verifying entitlement before planning production work
Enterprise Impact: Build your model roadmap on entitlement dates, not launch dates. Any team that planned work around a frontier model announcement this month lost a week, and the fix is contractual and architectural: an enterprise agreement that specifies when new models become available to your tier, and an application layer that lets you swap models without rewriting prompts and evaluations. The stronger signal is where the model went first. Cyber capable models are being released through vetted programs before general availability, so organizations that want early access to the most capable systems should expect to qualify for it, with attestations about identity and intended use, rather than simply subscribe.
Source: CSO OnlineThe Seattle Times and Newsday Join the Publishers Suing OpenAI and Microsoft Over Training Data
The Seattle Times and Newsday filed suit against OpenAI and Microsoft in the Southern District of New York, alleging their journalism was used to train AI systems without permission and describing generative AI as a snake eating its own tail that threatens to destroy the organizations producing the content it depends on. They join The New York Times, which has been litigating since 2023.
A Microsoft representative expressed surprise and said the company is always happy to sit down and explore solutions to this type of dispute. Both defendants have previously funded journalism initiatives and fellowships at The Seattle Times, which makes the filing notable for the relationship it interrupts as much as for the claims.
- Two more regional publishers in federal court over training data
- Same venue and claims as The New York Times case
- Defendants signal willingness to negotiate rather than contest publicly
Enterprise Impact: Training data litigation is now a steady stream rather than a single case, and the settlement patterns it produces will define what a licensed model costs. For buyers, the relevant contract terms are indemnification for output that infringes, transparency about training sources for any model you fine tune, and the vendor's obligation to notify you if a court order changes what the model may produce. For organizations that own content, from media to research to product documentation, the same litigation is establishing that training use has a price, and it is worth knowing what your archive is worth before a vendor makes an offer for it.
Source: TechCrunch