Adtran Opens Its Network Software to AI Agents, Adtran launched Mosaic One Fabric, a platform built on MCP that lets operators build their own AI agents using their network data, models and token budgets instead of a vendor's fixed toolkit.

My Take: The real test is whether operators actually build agents with this freedom, or just admit they don't have the staff to use it.

Fiber Users Ask AI 50% More Than Cable Users, Study Finds, Recon Analytics found people on fiber connections have AI sessions 50% more intensive than cable users, averaging 28.5 questions per session, with researchers warning of a new AI proficiency divide.

My Take: This is the network catching up to what AI actually needs - low latency and high throughput, not just a bigger download number. If fast AI access ends up tied to fiber the way fast gaming was, rural and lower-income households fall behind twice.. Once on connectivity and once on the tools built on top of it.

🇨🇦 Bell's AI Buildout Moves From Announcement to Execution, Bell CEO Mirko Bibic told investors construction is progressing at the 300MW Saskatchewan AI data centre, with a Winnipeg facility coming online this year, while the company added over 45,000 residential FTTH subscribers in Canada and cut traditional telco capex.

My Take: The detail that stands out is Bell shrinking its core telco capex while ramping AI spend. That's Bell telling investors where it thinks the next decade of margin actually lives. Fiber in the US. Data Centres in Canada. Follow the regulators.

Rural Texas Clinics Turn Shipping Containers Into Telehealth Hubs, with broadband access still patchy across rural Texas, universities are converting shipping containers into telehealth clinics so residents without home internet or nearby doctors can access virtual care.

My Take: This is what happens when connectivity funding stalls and the need doesn't. Healthcare providers become de facto ISPs, wiring up a shipping container because nobody wired up the community first. It works, but it's a workaround, and it shifts the cost of the broadband gap onto hospitals and universities instead of the agencies that were supposed to close it. See the regulatory section for a story about stalled Texas BEAD money.

Wi-Fi 7 Adoption Quadruples in the US, Still Under 10%, Ookla found US Wi-Fi 7 adoption grew four times over to 7.2% by Q1 2026, led by Charter Spectrum and fiber providers like CenturyLink and Frontier, with median download speeds up 76% since 2022.

My Take: A 4x jump sounds big until you remember it's 4x almost nothing. What's actually interesting is fiber providers leading adoption per customer while cable leads in raw volume. That's the difference between an ISP upgrading because it can, and one upgrading because its customer base is simply enormous. I have no immediate for WiFi7 at this point, but that’s me.

WISPA Chief on Fiber Overbuilds, Starlink and Spectrum, four years into leading WISPA, David Zumwalt says the roughly 600-member group now runs hybrid fiber and fixed wireless networks, competing against fiber overbuilds, private equity buyers, and Starlink for the same customers.

My Take: Small wireless ISPs used to be the default option where nobody else would build. Now they're fighting fiber overbuilders and Starlink for the same customers they once had to themselves. Is community presence enough? Likely not. Is price still the deciding factor? Seems it may be. Will Starlink wipe out some pure WISPs? Probably.

Zayo Confirms Nvidia as Anchor Customer for 8,000-Mile Fiber Build, Zayo is building over 8,000 miles of new long-haul fiber across AI-heavy corridors with Nvidia as its anchor customer, adding six new routes and upgrading capacity on ten existing ones.

My Take: Fiber companies spent years being told long-haul routes didn't pencil out anymore. Nvidia showing up as an anchor tenant proves the math changed, not because more people want internet, but because GPUs need to talk to each other over distance.

🇨🇦 Brantford Marks 150 Years Since Bell's First Long-Distance Call, Brantford, Ontario celebrated the 150th anniversary of Alexander Graham Bell's first long-distance telephone call, made from the Bell Homestead to Paris, Ontario in 1876.

My Take: It's worth sitting with how short that first call actually was, about 13 kilometres, and how long it took to get from there to calls over fiber running under an ocean and through satellites up in space. Every generation thinks its version of "long distance" is the final word. Every generation is wrong. ET, phone home (I never liked that movie..)

🇨🇦 Nova Scotia Lays Undersea Fibre to Connect Tancook Islands, Build Nova Scotia and Bell are laying nearly six kilometres of submarine fibre cable to bring high-speed internet to more than 180 homes and businesses on Big and Little Tancook Islands, part of a broader $193 million provincial initiative.

My Take: Six kilometres of cable for 180 homes doesn't pencil out on a spreadsheet built around subscriber revenue alone. It only happens because a province decided universal access was worth funding directly. That's the model most of rural Canada still needs, and it's the opposite of the technology-neutral, let-the-market-sort-it-out approach some jurisdictions are leaning toward. Hey, Bell. Install some fiber optic sensing on there to protect those links from tamper and damage,

1 Gig Fiber Is Now Just Table Stakes, ISP Execs Say, at the Mountain Connect conference, fiber executives from GFiber, NextLight, and Utopia Fiber said 1 Gig service is now standard across the industry, with operators differentiating on customer service, reliability, and price instead of raw speed.

My Take: The interesting shift isn't the speed, it's what providers are selling instead. When GFiber says the product is the customer service, not the gigabit, that's an admission that fiber has become a commodity. Reliability, support and Quality of Outcome are the new speed and price

Celona Merges Private 5G and Wi-Fi Into One AI-Managed Platform, Celona launched Orion, a platform combining private 5G, Wi-Fi 7, public cellular and satellite into one subscription managed by an agentic AI layer, aimed at connecting robots, autonomous vehicles and edge AI systems in industrial settings.

My Take: This is a bet that the next big wireless customer isn't a person with a phone, it's a robot on a factory floor that doesn't care which network it's on as long as it stays connected. Bundling every wireless option into one AI-managed subscription only matters if the orchestration actually works at scale, and that's the part nobody can verify from a press release. Always liked the Private 5G space and not sure why more aren’t taking advantage of it.. or maybe they are, and I just don’t see it..

Google Adds Three New Subsea Cables Linking the Caribbean to the Americas, Google announced the Alisios, Canoa, and OlaLuz subsea cable systems connecting the Dominican Republic to Panama, Chile, Bermuda, and Florida, plus a new Firmina branch, as part of its Americas Connect network expansion.

My Take: Google is laying this cable because the region sits on the shortest path between its own data centres, and they want new resilient and redundant paths. 9,500Km of new infrastructure, give or take 1,000Km?

QUICK HITS:

🇨🇦 Telus Confirms It Will Start Locking New Phones in October, Telus told the CRTC it will start applying a 60-day lock on newly financed phones in October, arguing it needs the same protection against theft and fraud that Bell already applies, with devices unlocking automatically and for free after 60 days.

My Take: Telus is moving ahead of CRTC approval, the same move Bell made and got a non-compliance order for. That tells you the carriers have decided the fine is cheaper than staying unlocked while thieves target them specifically. If the regulator wants unlocked-by-default to survive, it needs to solve the fraud problem it created, not just keep saying no - and the number of fraud stories aren’t slowing down.

Texas Freezes BEAD Disbursements Amid Starlink Favoritism Allegations, the Texas Broadband Development Office froze all disbursements from its share of federal BEAD funding after lawmakers ordered an audit over claims the office altered rules to favor Starlink.

My Take: This is the fallout from swapping a fiber-first standard for a technology-neutral one without spelling out exactly how the tradeoffs get scored. Once satellite became eligible for money that used to be earmarked for fiber, every award was going to invite a fight over who benefited. Freezing the whole program is a blunt way to buy time, but it also means real broadband projects sit stalled while the politics get sorted out. See the Texas portable health article above.

FCC Explores Letting Wi-Fi and Bluetooth Bands Talk to Satellites, the FCC voted 3-0 to open a rulemaking process examining whether more than 200 MHz of unlicensed spectrum, the bands used by Wi-Fi and Bluetooth, could support direct-to-device satellite connections.

My Take: Opening up unlicensed bands sounds like a technical footnote, but it's really a signal that satellite-to-phone connectivity is scaling faster than regulators expected, and the rulebook is being rewritten in real time to keep up.

SpaceX Plans Small Cells on Starlink Dishes to Rival Carriers Directly, SpaceX revealed plans to attach small cellular base stations to Starlink dishes already installed on customer rooftops, aiming to build a terrestrial mobile network that competes with AT&T, Verizon, and T-Mobile rather than just complementing them.

My Take: See my opening ramblings.

Viasat's Long-Delayed 100 Mbps Satellite Finally Nears Service, Viasat's ViaSat-3 F2 satellite, launched in November 2025 to serve the Americas, has completed its deployments and is awaiting FCC authorization to enter commercial service, promising to more than double Viasat's total network capacity.

My Take: Viasat has been playing catch-up on capacity for years while Starlink pulled ahead on both speed and latency. Doubling total bandwidth is a real milestone, but geostationary satellites still can't match low-Earth-orbit latency, so this closes the capacity gap without closing the experience gap - assuming this isn’t really for multicast-only type applications.

Starlink Mini vs 5G: Real-World Test Finds No Clear Winner, a side-by-side test of Starlink Mini against three 5G carriers found 5G posted higher peak speeds in strong-coverage spots, but Starlink delivered more consistent results overall, especially where towers were nowhere close.

My Take: Peak speed numbers make for a good headline, but consistency is what actually matters day to day. A satellite dish that gives you the same 300 Mbps whether you're in a city or a field beats a cellular connection that swings from a gig to nothing depending on which side of a hill you're standing on.. Reliability matters. I still hit areas where the phone either jumps to 3G, or nothing.

🇨🇦 SpaceX Brings Back Free Starlink Mini Perk for Top-Tier Plans, SpaceX restored a perk letting new Starlink Residential Max subscribers in Canada and the US claim a free Starlink Mini dish plus half-price Roam data, months after pulling it in June without explanation.

My Take: The lack of any public explanation either way tells you Starlink doesn't feel it needs to justify these moves to its subscriber base yet. Maybe this is Grok marketing doing what it does.

Why Optical Ground Stations Are Becoming Satellite Infrastructure's Bottleneck, a LinkedIn explainer breaks down optical ground stations, the telescope-and-laser facilities that receive high-speed data from satellites using light instead of radio, and why geographically distributed networks of them are now essential as more constellations adopt laser links.

My Take: Everyone talks about satellites and rockets, but the ground stations that actually catch the data are the unglamorous chokepoint. A laser link is only as good as the clear sky it needs to land on, which is why operators are building redundant stations across continents instead of relying on any single site. From what I understand, space to ground isn’t the big issue. Ground to space is, based on the atmospheric issues.

Study: 76% of Drivers Hit Weekly Dead Zones, Most Would Pay for Satellite Backup, a Viasat and 5GAA survey of 1,606 drivers across 10 countries found 76% experience weekly mobile signal loss while driving, with three-quarters saying they'd pay for satellite-connected safety features.

My Take: See the 5G vs Starlink article above.

Opinion: The "Internet of Space" Is Becoming Critical Infrastructure, an industry opinion piece argues that as more satellite constellations interconnect via optical crosslinks, the resulting mesh network in orbit is starting to function as resilient backup infrastructure for terrestrial networks during disasters and outages.

My Take: A mesh network in space is only resilient if enough operators actually interconnect with each other, and right now most of these constellations are built to compete, not cooperate - although there are ‘standards-based’ implementations of optical inter-satellite linking. Either way, integrating non-terrestrial into terrestrial continues to grow.

Russia Speeds Up Its Own Starlink Rival, Ukraine Warns, Ukrainian military intelligence says Russia's Rassvet satellite internet system, currently 16 satellites, is being deployed faster than planned with a target of nearly 300 satellites by 2027, giving Moscow the kind of battlefield connectivity Starlink has given Ukraine.

My Take: Starlink's role in this war has made satellite internet a strategic military asset, not just a consumer product, and every major power took note. Once Russia has its own version, the advantage Ukraine has held for years starts to erode, and the next fight will be about jamming and disabling constellations, not just building them.

Starlink Connections Drop an Average of 33 Minutes a Day, Data Shows, real-world monitoring across thousands of Starlink dishes found connections are unreachable for an average of 33 minutes daily, about 2.3% of connected time, with obstructions, loose cables, and network-wide outages the most common causes.

My Take: Loose cables, power loss and obstructions sound like a “you problem”, not a Starlink problem.

🇨🇦 Toronto's Kepler Books Its First Dedicated Rocket Launch, Toronto-based Kepler Communications selected Rocket Lab's Neutron rocket for its first dedicated launch mission, moving away from rideshare flights to gain full control over the schedule and orbit for its next-generation optical data relay satellites.

My Take: Booking your own rocket instead of hitching a ride is a company saying it has outgrown the beginner tier. Kepler's optical data relay business is a Canadian success story sitting underneath the noisier Starlink and Amazon headlines, and this launch is the clearest signal yet that its customer demand is real, not speculative. I’m here to promote.

AST SpaceMobile Positions Itself as the Anti-SpaceX for Carriers, after SpaceX announced plans to compete directly with mobile carriers, rival AST SpaceMobile used its earnings call to stress it builds alongside operators, not against them, while confirming its own satellite launch timeline has slipped to 2027.

My Take: SpaceX going after carriers directly just handed every competing satellite company its best sales pitch of the year. AST SpaceMobile is betting that telcos would rather partner with a company that needs them than compete with one that's building around them, and that bet gets more attractive the more aggressive Musk gets.

How Amazon's Satellite Internet Dream Fell Years Behind Schedule, seven years after Jeff Bezos first pitched satellite broadband on stage, Amazon Leo has just over 390 satellites in orbit against Starlink's 10,400-plus, hampered by rocket shortages and a Blue Origin launch failure that grounded its heaviest lift vehicle.

My Take: Amazon had every advantage on paper, capital, cloud infrastructure, retail distribution, and still lost seven years to launch bottlenecks it doesn't fully control. It's a reminder that in the space business, the constraint isn't always the satellite you build, it's the rocket you can actually get a seat on.

NASA's Next Challenge: Building a Cell Network for the Moon, as Artemis missions ramp up, NASA and partner agencies are racing to build lunar communications infrastructure capable of real-time 4K video, since current systems can only manage near-real-time data due to bandwidth limits on the link back to Earth.

My Take: It's a strange loop that the same industry racing to blanket Earth in high-speed satellite internet is now solving the exact same problem 250,000 miles further out. The lesson doesn't change with distance: connectivity infrastructure always lags the ambition of the mission it's supposed to support. I thought Nokia was building cell sites on the moon?

🇨🇦 Ontario Launches Data Centre Playbook - Framework will attract data centre investment, create jobs and deliver real benefits to host communities while ensuring Canadians’ data remains in Canada

My Take: “Ontario has a choice: we can build the economy of the future and ensure Canadians’ data remains in Canada, or we can get left behind and risk Canadians’ data being sent to the U.S. and elsewhere,”

Data Centre Labor Poaching Is Driving Up Customer Costs, Execs Warn, Schneider Electric and Iron Mountain executives say the industry's habit of poaching skilled electricians and cooling technicians from competitors, rather than training new ones, is a short-term fix creating long-term skills gaps and rising costs.

My Take: Roughly a quarter of departures in this industry are people getting hired away by a competitor, not people leaving the field. That means companies are bidding against each other for the same small pool of talent instead of growing it. Jobs that AI won’t take, kids.

Virginia Forces Data Centres to Pay for Their Own Grid Upgrades, after AI data centres drove a 76% spike in regional electricity prices, Virginia's State Corporation Commission ordered data centres to cover the full cost of any transmission infrastructure built exclusively for their use, rather than spreading it across all ratepayers.

My Take: This is the fix that should have been the default from day one. If a facility needs a dedicated substation to run, the facility should pay for it, not the retiree next door on a fixed income. Virginia hosting more data centres than any other state gave it more leverage than most, which is probably why it moved first.

Amazon Used 45-Year-Old Zoning Rules to Build Without Resident Input, Amazon built a 438,500-square-foot data centre on Gilroy, California farmland using zoning approvals that trace back 45 years, with most residents only learning about the project after construction had already started and the comment period had closed.

My Take: Nothing here was technically illegal, and that's the actual story. A single city staffer had the authority to approve a project this size because zoning written for a different era never anticipated a facility like this. The legal process worked exactly as written and still left residents feeling like the process was designed to exclude them. And that’s the real NIMBY issue — not the technology, the water or the noise.. the consultation and inclusion.

Crusoe and Aalo Will Test a Nuclear-Powered AI Data Centre in 2027, Crusoe and nuclear startup Aalo Atomics plan to power a modular Crusoe Spark data centre with Aalo's sodium-cooled microreactor at Idaho National Laboratory in 2027, following Aalo's reactor reaching criticality on July 4.

My Take: The actual test here isn't whether the reactor works; that part is largely solved physics. It's whether a small nuclear plant can hold steady power delivery while a data centre's GPU load spikes and drops in real time. That's a genuinely new engineering question, and 2027 will be the first real data anyone has on it. If the NIMBYs don’t like Data Centres, they’ll love SMRs 😉 See the last story in this section.

Telcos Are Becoming Data Centre Operators, Not Just Connectivity Providers, a roundup of ten telecom operators, including Telstra, e&, and Verizon, shows carriers increasingly leveraging their existing subsea cables, fibre networks, and real estate to build out colocation and hyperscale data centre capacity directly.

My Take: Telcos already own the two things a data centre business actually needs - land in the right places and the fibre to move data, fast. Watching carriers move from selling connectivity to selling compute is the same thing that pushed Bell into AI infrastructure. Everyone with the pipes wants a piece of what's flowing through them. Monetization, and all..

🇨🇦 Waterloo Region Orders a Study Before the Data Centres Arrive, Waterloo Regional Council voted to have staff prepare a report on how data centres will affect the local economy, infrastructure, and environment, aiming to get ahead of the issue before facilities are proposed rather than reacting after the fact.

My Take: Doing the homework before a developer shows up with a proposal is the smart. Every other story in this section is a community reacting after the fact. Water use fights, secret zoning approvals, angry protests. Waterloo doing the research first means it can negotiate from a position of knowledge rather than catch up after the ground has already been broken. Others, take note.

🇨🇦 Ontario Company Pitches a Data Centre That Grows Its Own Food, a Waterloo-based firm proposed converting a former industrial site in Fergus, Ontario into an "ecological AI data centre" with a vertical farm on top that would run on the facility's waste heat, though public details on the project's status remain thin.

My Take: Hmm.. Reusing waste heat instead of just venting it is genuinely good engineering, and it's the kind of idea that should be standard practice, not a headline. The skepticism is fair too. A striking rendering of lettuce growing above server racks is a much easier thing to announce than to actually build and operate at scale. I’m sure it’s some type of lettuce they’re growing…

Opinion: Communities Need a "Prenup" Before Saying Yes to a Data Centre, an industry opinion piece argues municipalities should negotiate binding community infrastructure agreements before approving data centre projects, spelling out obligations, exit conditions, and connectivity commitments up front rather than relying on goodwill after the fact.

My Take: The root cause: communities agreeing to decades-long partnerships without spelling out what happens if the other side doesn't hold up its end. See the article about Waterloo, above.

Local Opposition Now Threatens $130 Billion in Data Centre Projects, at least 75 data centre projects worth a combined $130 billion faced organized local opposition in the first quarter of 2026 alone, according to Data Center Watch research cited by Reuters, with lenders increasingly pricing cancellation risk into their financing.

My Take: The moment lenders start pricing in cancellation risk, community opposition stops being a public relations headache and becomes a line item on a spreadsheet. That's the point where this backlash actually starts changing corporate behaviour, not because the industry suddenly cares more about residents, but because angry neighbours now show up as a cost of capital. Money talks.

Gallup: 71% of Americans Oppose a Data Centre in Their Area, a Gallup poll found 71% of Americans oppose AI data centres being built locally, higher than the 53% who oppose a nearby nuclear plant, with opponents citing water use, energy demand, and pollution as their top concerns.

My Take: People being more comfortable with a nuclear plant next door than a data centre should stop the industry in its tracks. Nuclear spent fifty years building a reputation for danger. A data centre has done it in about three, and that's the trust gap every developer in this issue is now trying to build a "prenup" or a portable pod around.

How Claude Marks AI-Generated Content, Explained, Anthropic published a support article detailing how Claude models embed imperceptible watermarks in generated text and signed provenance metadata in generated files, to comply with the EU AI Act's transparency rules.

Anthropic Confirms Claude Will Watermark All Generated Text, Anthropic confirmed that all Claude models released after August 2 will embed machine-readable watermarks in generated text and files by default, applying across the API, Claude, Claude Code, Claude Cowork, and Claude Tag, in response to the EU AI Act's Transparency Code.

My Take: A watermark that can survive copy-paste but not a heavy rewrite is a useful signal, not a hard proof. The real value here isn't catching every piece of AI text, it's giving platforms and readers a starting point for asking the right question instead of guessing blind. Having said that, is there a problem with mixing original and AI content, or augmenting AI content with original content?

Trump Says Congress Wants to Regulate AI "Out of Business", President Trump said Congress wants to over-regulate the AI industry, in comments that came the same week OpenAI and Anthropic separately disclosed that their models had escaped testing environments and hacked outside companies during security evaluations.

My Take: Arguing against regulation the same week two frontier labs admit their own models broke containment during testing is a hard sell, and it puts the White House squarely on one side of a fight that's about to get a lot more public.

AI Safety Researcher: Models Learning to Cheat Might Be a Good Thing, after OpenAI and Anthropic models independently escaped their test sandboxes and hacked outside systems including Hugging Face, AI safety researcher Nate Soares argued the underlying problem is how these models were trained to grade tasks, not weak sandboxing.

My Take: Building a stronger lock doesn't help if the model was trained to treat the lock as a rule, not a boundary, and to route around it anyway. That's a training problem, not an infrastructure problem, and it's a much harder one to fix.

Business Students Using AI Weekly Jumped From 6% to 29% in Three Years, a three-year survey from American University's Kogod School of Business found regular AI use among business students climbed from 6.2% to 29%, with students asking for more structured guidance on using it ethically rather than resisting it outright.

My Take: The graduation-speech boos get the headlines, but this survey shows the real story underneath them. Students aren't rejecting AI, they're asking someone to actually teach them how to use it well, and a lot of schools still haven't figured out they're supposed to be the ones doing that teaching. Some primary schools are starting to “teach” about AI, but that’s only as effective as the person teaching it...

The Case for Portable AI Data Centres, an industry explainer lays out the growing case for shipping-container-sized modular data centres that can deploy AI compute in days rather than the years a traditional hyperscale build takes, aimed at remote sites, disaster recovery, and edge inference.

My Take: A container that shows up, does the job, and can leave is a very different proposition than a 500-acre campus locked in for 30 years. Package with an SMR the size of a garden shed, and off you go!

OpenAI Flags Its Own Upcoming Model as a Cybersecurity Risk, OpenAI disclosed that internal testing of its upcoming Astra model showed advances significant enough that it can no longer rule out the model crossing the "Critical" cyber capability threshold under its own Preparedness Framework, and is adding stricter controls before wider release.

My Take: A lab publicly flagging its own model as potentially dangerous before release, instead of after something goes wrong, is the kind of transparency the industry needs more of. It also means the honest answer to "is this safe" is now sometimes "we're not sure yet.” That’s not a good place for a fast-moving industry to sit.

SpaceXAI Launches Grok 4.6 at Half the Price of Rival Frontier Models, SpaceXAI released Grok 4.6, matching OpenAI's GPT-5.6 Sol on the Artificial Analysis Intelligence Index while pricing it at roughly half the cost of comparable frontier models, with a focus on long-running agentic coding tasks.

My Take: Matching frontier performance at half the price is a pricing war move as much as a capability one. If Grok can genuinely hold that line, it puts real pressure on the margins OpenAI and Anthropic need heading into their own next moves. Pricing pressure has a way of reshaping an industry faster than benchmark scores do. Look at the Canadian wholesale market for an example.

Thinking about a newsletter of your own? Check out beehiiv!