The AI week, distilled.
Week 34 · 2026
This week in non-Microsoft AI

Nvidia and Google set the pace on AI capacity, pricing, and leadership

Nvidia moved beyond chips into financing and underwriting for large-scale AI data centers, reinforcing how capital markets now shape AI capacity and cloud pricing. Google pushed fast model iteration and reported mass-market adoption, while leadership changes at DeepMind signaled a new operating structure for Gemini delivery.

01

Nvidia backs OpenAI-linked Ohio data center with guarantee

Nvidia agreed to guarantee up to $105 billion to support leasing for an SB Energy data-center campus in Ohio and to invest $1.5 billion in the project that will serve OpenAI.

  • Expect AI service pricing to reflect long-dated infrastructure financing costs, especially for high-end inference and training workloads.
  • Treat Nvidia exclusivity at new mega-sites as a vendor concentration risk when negotiating cloud commitments and multi-year AI roadmaps.
  • Use this as a trigger to validate contingency options (alternative model providers, open models, or non-Nvidia accelerators) for critical workloads.
02

Nvidia pitches GPUs as financeable assets via $500B platform

Nvidia announced compute-financing platforms with major asset managers, aiming to mobilise more than $500 billion of third-party capital for AI infrastructure over time.

  • Track whether your cloud providers adopt GPU leasing or capacity reservation constructs that shift costs into predictable OPEX rather than ad-hoc consumption spikes.
  • Prepare procurement to evaluate AI capacity deals like infrastructure contracts (tenor, exit clauses, utilization floors), not only per-token pricing.
  • Reassess lock-in exposure because Nvidia’s role expands from hardware supply into financing leverage across the AI stack.
03

Google releases Gemini 3.7 Flash with halved pricing

Google launched Gemini 3.7 Flash, positioning it as a workhorse model for coding and agents and offering temporarily reduced token prices through year-end.

  • Benchmark real workload cost (per resolved ticket, per PR review, per automated workflow run) because token price cuts do not guarantee lower end-to-end cost.
  • Plan for frequent model refreshes in production by versioning prompts, regression-testing tool calls, and maintaining an approval path for model upgrades.
  • Use the model’s agent focus to pilot task decomposition (planner vs executor) and measure reliability under enterprise guardrails.
04

Gemini app reaches one billion monthly users

Google said the Gemini app surpassed one billion monthly users, with reports highlighting that a majority of interactions now happen via voice.

  • Assume employees and customers will bring prior expectations from consumer Gemini into enterprise deployments, which raises the bar for UX and response quality.
  • Treat voice as a first-class channel by validating data retention, consent flows, and transcription storage in Czech and EU compliance contexts.
  • Prioritise identity, logging, and policy controls because higher usage typically increases shadow-AI risk and operational support load.
05

Hassabis steps down as DeepMind CEO

Demis Hassabis moved from CEO of Google DeepMind to Alphabet chief scientist and DeepMind chair, while Koray Kavukcuoglu took over day-to-day leadership of DeepMind and the Gemini roadmap.

  • Expect potential reprioritisation between research-led releases and enterprise operational needs such as admin controls, region support, and auditability.
  • Update vendor risk reviews to include leadership and organisational changes that can affect roadmap predictability and support escalation paths.
  • Reconfirm contract protections for model changes (deprecations, pricing changes, regional availability) when committing to Gemini-based solutions.