Dion Hinchcliffe

1️⃣ Big news today: After two months of deep analysis, @nickpatience and I are releasing @TheFuturumGroup's new Agentic AI Signal. It's a definitive evaluation of the platforms powering the next era of enterprise AI. The space is young, explosive, and moving at lightning speed.

Use of AI Triples for Product and Enterprise Innovation in 2026 futurumgroup.com/press-release/… From my latest global CIO survey. The use of AI gets far more strategic this year.

8/ My assumptions (terse): • Sustained disruption (6–24 mo) • Partial (not 30% like some say) but persistent Qatar helium loss • No new helium supply step-change • 2–6 mo fab inventories → ~4–6 mo inflection • Helium critical for key etch/cooling steps (no true substitute) • HBM = binding constraint (SK hynix/Samsung concentration) • AI output = min(GPU, HBM) • LNG = secondary (cost/grid pressure, not immediate outage) • Korea/Taiwan have a buffer via nuclear, coal, alt LNG • No demand collapse (AI stays supply-constrained) • Gradual throughput decline, not sudden failure • Limited substitution + partial recycling gains only

7/ CIO implications: AI systems get hit twice: GPU supply tightens AND HBM constrains completion. Hedge now: • Lock AI capacity early • Diversify cloud/providers • Prioritize high-ROI workloads • Optimize token efficiency This is no longer just a software scaling problem.

1/ Here’s my baseline assessment of what happens to critical AI components if the Iran war continues. It shows a significant double digit decline. Helium, energy, and a narrow set of fab chokepoints. If the disruption persists, the risk isn’t collapse. It’s a steady ramp down.

@sarbjeetjohal @antgrasso @eric_kavanagh @CTOAdvisor @MarshaCollier @cleartechtoday @JimHarris @EvanKirstel @emollick @HaroldSinnott Yes, it absolutely does matter. To a degree. This is a large signal. I do want to see it cross-checked. Provocative and early, and without exactly how they classified AI spenders, so it’s hard to say. But nevertheless very interesting.

@dhinchcliffe @antgrasso @eric_kavanagh @CTOAdvisor @MarshaCollier @cleartechtoday @JimHarris @EvanKirstel @emollick @HaroldSinnott It depends on who’s in ramp customers what constitutes/defines AI spend :)

AI spend is directly correlated to significant revenue growth. Perhaps the most interesting chart yet about the era of AI. Companies with little AI spend track the broader economy. Those with high AI intensity are now seeing triple digit revenue growth. Sure to spur debate.

@joannefriedman @CPetersen_CS No worries. Just glad you make it when you are able to. And yes, I’m lucky with my locales. Some more good ones coming!

@dhinchcliffe @CPetersen_CS Sorry I've been AWOL, my plate runneth over... and I'm green with envy at your locals TBH. :)

Good afternoon, #ciochat, from bright and sunny Tennessee!

A3) So far, working that into project and product budgets and forecasts. As @nyike said, those forecasts have to stay pretty darn loose since estimation is rough at best. Model churn doesn't help. #ciochat

#CIOChat Good chat today. Join us at 2pm ET next Thursday. Topic: Using IT for Growth This is month 12 of our movie tie-ins to intro next week’s topic. Hope you’ve enjoyed them. Please suggest topics and invite your IT colleagues. Every Thursday at 2pm ET.

@joannefriedman @CPetersen_CS Thanks for stopping by too, Joanne!

@CPetersen_CS @dhinchcliffe hey there! hope all is well

@dhinchcliffe #CIOChat - use a context graph to allocate by correlated groups. (e.g. marketing and sales, finance and ops)

#CIOChat Q4: Endgame: If more IT workloads consume tokens, budgeting shifts from fixed to variable, a major $ issue. How are CIOs adapting financial models like FinOps for AI, chargebacks, or automated guardrails? How do you manage volatility while still enabling innovation?

A2) Prompt discipline / design and model choice more than in-line model routing. Maybe a path to fewer surprises, maybe not. #ciochat

@dhinchcliffe #CIOChat the tip is to really hone in on what needs Frontier models, which model versions, and what can be done with OS models (specific ones) and... look at inferencing.. and ways around it.

#CIOChat Q3: As tokens become a shared resource across the org (or maybe hoarded instead), how are you allocating them across teams and roles? Are you setting budgets per product, per developer, per agent? Or letting usage float and then finding ways to reconcile later?

@dhinchcliffe #CIOChat A2: All of the above, but primarily we don't let AI handle everything. Call me luddite but humans still have critical thinking skills and creativity.