Digital Colliers Daily Briefing — August 25, 2026
Consolidation, labor disruption, and infrastructure lock-in defined today's news cycle. A reported $13 billion approach for Hugging Face suggests core AI infrastructure is entering an M&A phase, while updated Stanford research quantifies the labor market damage to young workers in AI-exposed fields. Meanwhile, new figures show gas-fired power in development for US data centers has nearly doubled in eight months, a structural shift that will shape the country's climate and energy trajectory for decades.
1. Hugging Face weighs a $13B exit as AI infrastructure M&A accelerates
What happened. Hugging Face has been approached at a valuation of $13 billion or more, according to a Business Insider report surfaced by TechCrunch over the weekend. The startup — the de facto hub for open-source model sharing — has reportedly engaged banks to evaluate bids, though no counterparty has been named and no deal has been struck. The talks arrive after Hugging Face turned down a $500 million investment from Nvidia earlier this year at a $7 billion valuation, citing concerns about a single dominant investor. The company last priced at $4.5 billion in 2023 in a round led by Salesforce Ventures.
Why it matters. A $13 billion take-out would represent a near-tripling of Hugging Face's private valuation in under three years and mark one of the largest consolidations in AI infrastructure to date. It follows Stripe's $7 billion acquisition of OpenRouter and signals that the picks-and-shovels layer of AI — model registries, routing, evaluation — is now a strategic target for platform companies rather than a neutral commons.
Who is affected. Millions of developers rely on Hugging Face for model hosting, datasets, and Spaces. Any acquirer would inherit that trust relationship along with sensitive metadata about who is building what. CEO Clem Delangue has repeatedly framed the company as a steward of community assets, telling TechCrunch's Equity podcast that Hugging Face is "close to profitability" and optimizing for "long-term sustainability" rather than fundraising maximization — language that raises the question of whether the board is genuinely for sale or simply price-discovering.
What to watch next. The identity of the bidders. Hyperscalers (Amazon, Microsoft, Google) and enterprise platforms (Salesforce, IBM) all have strategic reasons to control the model distribution layer, but any deal would draw immediate antitrust scrutiny given Hugging Face's central position. Also worth watching: whether an announced sale triggers a fork or migration to alternatives.
Sources:
2. Stanford: entry-level employment gap in AI-exposed jobs widens to 19%
What happened. The August 2026 update to Stanford's "Canaries in the Coal Mine?" study finds that employment for workers aged 22 to 25 in the most AI-exposed occupations is now 19 percent below employment for peers in less-exposed fields, according to Ars Technica's coverage. That gap was 13 percent in the researchers' 2025 edition. Older workers in the same occupations appear largely unaffected in the data so far.
Why it matters. This is among the first empirical labor-market signals that generative AI is displacing specific worker cohorts rather than raising productivity uniformly. The concentration of the effect on the youngest workers — those typically hired into the codeable, summarizable, first-draft tasks that current models handle competently — matches the theoretical predictions economists have been debating for two years. The six-percentage-point widening in twelve months suggests the effect is not a one-off adjustment.
Who is affected. Recent graduates in software, customer support, paralegal work, and junior analyst roles face the sharpest exposure. Universities and coding bootcamps that market entry-level placement outcomes will feel downstream pressure. Employers face a pipeline problem: if entry-level hiring collapses, the mid-career talent bench in five to seven years will be thin, a dynamic that has historically produced wage spikes.
What to watch next. Whether the gap continues widening in the next data release, whether it appears in other advanced economies, and how policymakers respond — expect renewed debate around wage subsidies for junior hires, apprenticeship funding, and AI-adjusted unemployment metrics.
Sources:
3. Gas-fired capacity for US data centers reaches 189GW in development
What happened. Global Energy Monitor reports that gas-power projects tied specifically to US data centers reached 189 gigawatts in development as of mid-2026, up from 97GW at the end of 2025 and just 4GW in early 2024, according to Wired. The buildout is heavily weighted toward "behind-the-meter" plants — private facilities that bypass grid interconnection queues. The Trump administration has encouraged the pattern via a voluntary pledge signed by Microsoft, Meta, Google, and OpenAI, along with several Republican governors and major utilities. The shift has pushed the US back ahead of China as the country with the most gas projects in the pipeline; China's data-center expansion, by contrast, is oriented around solar and hydropower, Brookings fellow Kyle Chan told Wired.
Why it matters. Even discounting for projects that never break ground — and GEM analyst Jenny Martos cautions that many will not — a buildout at this scale locks in fossil emissions for decades. Wired has previously reported that some of these plants are permitted to emit more greenhouse gases annually than many small and mid-sized countries, and many are being deployed with less efficient turbines to compress schedules. The behind-the-meter approach also shifts costs and risk off ratepayers but concentrates local air-quality impacts near data-center clusters.
The gas story is not the whole picture. Ars Technica reports, citing S&P Global Energy, that US clean-energy additions will rise by a record 45GW this year — roughly 25 percent above 2024 — driven by data-center demand and developers racing to capture expiring tax credits. Separately, Ars reports that data centers have become the "killer application" for solid-state power transformers, potentially easing a multi-year backlog on conventional transformer hardware that has bottlenecked grid upgrades.
Who is affected. Hyperscalers gain speed-to-power but take on operator risk in a business adjacent to their core competency. Utilities face demand forecasts that swing by tens of gigawatts on individual siting decisions. Communities near proposed sites face permitting fights, and ratepayers face uncertain cost allocation as behind-the-meter and grid-connected loads intermingle.
What to watch next. Attrition in the 189GW pipeline as turbine supply, financing, and local opposition constrain projects; the pace of solid-state transformer commercialization; and whether the clean-energy surge sustains itself once current tax credits fully sunset.
Sources:
- Data Centers Are Driving an Alarming Gas Power Expansion in the US — Wired
- Trump tried to curb clean energy. It's booming anyway. — Ars Technica
- Data centers become "killer application" for new power transformer tech — Ars Technica
Today's three stories track the same underlying pressure at different layers of the stack. Capital is consolidating around the infrastructure that trains and distributes models; the labor market is absorbing the first measurable shocks from what those models can do; and the physical grid is being reshaped to feed the whole system. Each trend has its own timeline — deal announcements in months, employment shifts in years, gas plants in decades — but the compounding effects will define the operating environment for tech, energy, and policy for the remainder of this cycle.