NVIDIA Hits $228.52 as AI Compute Supercycle Rewrites Hyperscaler Budgets

NVIDIA (NVDA) closed at $228.52 on Wednesday, up 0.97%, as Wall Street digested fresh signals that the AI compute supercycle is accelerating rather than plateauing. AMD posted an even sharper single-day gain of 1.90% to $605.53, while the broader VanEck Semiconductor ETF (SMH) slipped 0.49% to $588.12 — a divergence that tells a specific story: investors are rotating into the most direct AI-infrastructure winners and trimming broader chip exposure where the AI narrative is murkier.

Hyperscaler Capex Is the Engine, and It’s Running Hot

Microsoft, Alphabet, Amazon, and Meta have collectively committed well over US$300 billion in capital expenditure for fiscal 2026, with the bulk earmarked for AI data centres, custom silicon, and network fabric. Microsoft closed at $543.45 (+0.43%) — a modest move, but the stock is up more than 40% since the company disclosed its expanded Azure AI infrastructure roadmap in early 2026. The investment thesis here is straightforward: every dollar a hyperscaler spends on a GPU cluster flows through NVIDIA’s order book first, then ripples into cooling, power, and fibre. The supercycle is not a single event; it compounds.

AI adoption metrics reinforce the capex case. Enterprise AI workloads — inference, retrieval-augmented generation, and agentic automation — are now consuming roughly three times the compute they did at the start of 2025, according to data-centre utilization reports cited by multiple cloud providers. That demand is structural, not cyclical. Enterprises that deferred AI deployment in 2024 are now signing multi-year contracts, pulling forward GPU demand that analysts had pencilled in for 2027 and beyond.

The Canadian Angle: Cohere, Vector, and a Quiet Infrastructure Play

Canada’s AI ecosystem is maturing fast, and the hyperscaler buildout is creating direct tailwinds for homegrown players. Toronto-based Cohere, founded by Vector Institute alumni including CEO Aidan Gomez, has emerged as the leading enterprise-grade large-language-model provider outside the U.S. hyperscaler duopoly. Cohere’s Command and Embed models are now embedded in workflows at several of Canada’s largest financial institutions, and the company is reportedly in advanced talks for a Series D that would value it north of US$6 billion — approximately C$8.55 billion at today’s USD/CAD rate of 1.4241. On the TSX, Constellation Software (CSU) surged 2.98% to C$3,093.29, its best single-day move in months, as investors re-rated the vertical-software compounder on expectations that AI-augmented pricing power will widen already-formidable operating margins across its 500-plus subsidiary portfolio.

Valuation Risk: The Denominator Problem

The bull case is compelling, but the math deserves scrutiny. NVIDIA now trades at roughly 38 times forward earnings — a premium that is only justifiable if data-centre revenue sustains 40%-plus year-over-year growth through at least fiscal 2028. Any deceleration in hyperscaler capex commitments, whether from margin pressure, regulatory intervention on AI energy consumption, or a sudden shift to custom silicon (Google’s TPUs, Amazon’s Trainium), could reprice NVDA sharply. SMH’s 0.49% decline on a day when NVDA rose is a quiet warning: the market is already differentiating between AI pure-plays and companies merely riding the semiconductor tide.

For Canadian retail investors, the more measured entry point may be infrastructure-adjacent equities — power, data-centre REITs, and fibre — where valuations are less stretched and the AI tailwind is just as real. The supercycle rewards patience as much as conviction.