Hyperscaler AI Capex Hits $300B in 2026, Fuelling a Global Compute Supercycle

The numbers are no longer speculative. As of September 2026, the four largest hyperscalers — Microsoft (MSFT), Alphabet (GOOGL), Amazon (AMZN), and Meta Platforms (META) — have collectively disclosed or guided toward more than $300 billion USD (approximately $421 billion CAD at the current 1.4044 rate) in capital expenditure for the fiscal year, with the overwhelming majority directed at AI compute, data centres, and supporting energy infrastructure. This is not a single-quarter spike. It is a structural re-platforming of the global economy’s digital backbone, and it is accelerating.

What the Capex Wave Is Actually Buying

Strip away the press releases and the dollars are flowing into three distinct buckets: GPU clusters (dominated by NVIDIA’s H100 and next-generation Blackwell architecture), custom silicon (Google’s TPUs, Amazon’s Trainium, Microsoft’s Maia), and physical data centre capacity tied to long-term power purchase agreements. Microsoft alone has committed to more than $80 billion USD in data centre construction this fiscal year, with significant buildouts announced in Virginia, Wisconsin, and — critically for Canadian investors — the greater Toronto and Quebec corridors, where hydroelectric power gives Canadian facilities a meaningful cost-per-watt advantage over U.S. counterparts running on natural gas peakers.

NVIDIA remains the fulcrum of the entire supply chain. The company’s most recent guidance pointed to data centre revenue sustaining above $30 billion USD per quarter, a run rate that would have been unthinkable three years ago. Gross margins have held above 70%, defying analyst fears that hyperscaler custom silicon would cannibalize NVIDIA’s dominance. Instead, the evidence suggests custom chips are handling inference workloads at scale, while NVIDIA retains a near-monopoly on frontier model training — a distinction that matters enormously for the durability of its earnings power.

The Canadian Angle: More Than a Bystander

Canada is not merely a passive beneficiary of cheap electricity. Cohere, the Toronto-based enterprise AI company co-founded by Aidan Gomez, has emerged as one of the most credible NVIDIA-alternative stack plays, building large language models optimized for private, on-premise enterprise deployment — exactly the architecture that regulated industries like Canadian banking and healthcare require. The company has raised over $500 million USD in cumulative venture funding and counts Oracle and Salesforce Ventures among its backers. On the public markets, Constellation Software (CSU) and Shopify (SHOP) continue to attract AI premium valuations on the TSX, with institutional investors treating both as proxy plays on enterprise AI adoption in North American software. The Vector Institute in Toronto, meanwhile, reported in its 2026 annual update that Canadian AI research talent placement into industry roles hit a record, with more than 60% of graduates fielding offers from hyperscalers or well-funded AI startups before graduation.

The Risk That Deserves More Attention

The supercycle thesis rests on one load-bearing assumption: that AI monetization catches up to AI infrastructure spending before the credit cycle turns. So far, the evidence is encouraging but uneven. Microsoft’s Copilot suite has crossed $10 billion USD in annualized revenue, and Meta has cited AI-driven ad targeting as the primary driver of its 2026 revenue re-acceleration. But Amazon’s AWS AI revenue remains harder to disaggregate from its broader cloud growth, and Alphabet has been notably cautious about providing AI-specific monetization metrics. If one or two hyperscalers guide toward capex deceleration in their Q3 2026 earnings calls — scheduled for October — the entire sector could reprice sharply. Valuation discipline matters: buying a supercycle narrative after a multi-year run requires accepting that much of the good news is already in the price.