We Have 90 Days
Why the next quarter matters more than the next five years.
Last week I wrote about the bill coming for AI. The convergence of rising energy costs, the infrastructure arms race, and a labour market flooding with displaced professionals. I said the downstream effects would start showing up soon.
That was seven days ago, and the timeline just compressed.
The Strait of Hormuz was already effectively closed. This weekend, a second major shipping chokepoint came under direct threat when armed groups in Yemen launched strikes that could foreshadow a closure of the Bab al-Mandeb Strait, the narrow passage off Yemen’s coast that controls access to the Red Sea and the Suez Canal. Maersk has already paused trans-Suez sailings. Oil closed at three-year highs on Friday, according to CNBC, before any of this happened. If both straits close simultaneously, a significant share of global energy and trade flows will be physically blocked.
The Philippines declared a national energy emergency on March 24. According to Philippine government statements reported by Bloomberg, the country has roughly 45 days of fuel remaining. South Korea, which imports around 70 per cent of its crude oil through the Strait of Hormuz, launched a nationwide energy-saving campaign asking citizens to take shorter showers, ride bicycles for short trips, and only use washing machines on weekends. These are major economies making emergency moves because the energy supply they depend on has been cut off at the source.
I grew up in the Philippines. I have family there. So when I say this feels real, I mean it in a way that goes well beyond market analysis. The numbers on the page become different when you know the people behind them.
And all of this is landing on an AI industry that was already stretched.
Last week, the question was whether AI would get more expensive. This week, two things happened that turned that question into a fork in the road.
The first is that Anthropic accidentally leaked details of its most powerful model ever. As first reported by Fortune, a misconfiguration in their content management system left nearly 3,000 unpublished assets in a publicly searchable data store, including a draft blog post about a model called Claude Mythos. Anthropic confirmed the model exists, calling it a step change and the most capable they have ever built. It sits in a new tier above their current Opus line, codenamed Capybara, and reportedly delivers major improvements in coding, reasoning, and cybersecurity.
What matters most for this conversation is the economics. The leaked materials indicate the model is extremely expensive to run, both for Anthropic and for customers, and that the company is working to bring costs down before any wide release. The most capable AI model ever built, and the company behind it has not yet figured out how to make the economics viable at scale.
The irony of a company building advanced cybersecurity capabilities while exposing its own model through a basic configuration error writes itself. But the bigger story is what it reveals about where the frontier is heading. The most powerful models are getting smarter and simultaneously more expensive, at a time when the energy powering them costs more than it did a month ago.
The second thing that happened is the opposite. Alibaba’s Qwen team released a series of small models that run on standard laptops. Mistral Small 4 topped open-source reasoning benchmarks earlier this month. Industry leaders including senior executives at AT&T and IBM have said publicly that fine-tuned small models will define serious AI adoption in 2026, because they can match frontier models in accuracy for specific business tasks while running faster and costing significantly less. IBM’s research team is framing this year as the inflection point between frontier and efficient model classes.
The AI industry is splitting in two directions at once. One path is getting bigger and more expensive at the exact moment energy costs are spiking. The other is getting smaller, faster, and cheaper at the exact moment businesses need to control costs.
I keep thinking about a finding from BCG’s AI Radar survey published earlier this year. Roughly half of the CEOs surveyed said their position depends on AI delivering results in 2026, and corporations across the board expect to double their AI spending this year, according to the same survey. The overwhelming majority of those executives believe AI agents will produce measurable returns before the year is out.
That is an enormous amount of pressure pointed in one direction. And most of it was planned before oil hit $100 a barrel, before two shipping straits faced closure, and before the infrastructure powering these investments got materially more expensive to run.
The budgets were set in a different world. The world changed.
The investment itself may still be right. But the strategy behind it needs to catch up to the reality on the ground. And for most businesses, that catching up needs to happen in the next 90 days.
Here is how I am thinking about it.
If you are a business that has already started integrating AI, the next quarter is when you find out whether your approach is resilient or fragile. Are you locked into a single provider whose pricing could shift? Are you running expensive frontier models for tasks that a smaller, fine-tuned model could handle? Are you measuring what AI actually saves you, or are you measuring how much you spend on it?
If you have not started yet, the calculus just changed in a way that might actually work in your favour. The rise of small, efficient models means the barrier to entry is lower than it has ever been. You do not need Mythos. You probably do not need Opus. You might need a 9-billion-parameter model running locally that does one thing brilliantly for your specific use case, costs almost nothing to operate, and does not depend on a data centre powered by gas that is stuck on the wrong side of a conflict zone.
The companies that will come out of this period strongest are the ones that get specific now. What AI does for them, what it costs, where they are exposed, and where they have options.
Five-year AI roadmaps are a luxury that the current moment does not support. The next 90 days will reshape pricing, capability, and access in ways that make long-range planning an exercise in guessing. The businesses that treat this quarter as the strategic window, and move with clarity instead of waiting for certainty, will be the ones that set the pace for everything that follows.
I build AI infrastructure for a living. I am watching the same energy markets, the same model releases, the same pricing signals as everyone else. And what I keep coming back to is that the opportunity here is precision. Knowing exactly what you need, finding the most efficient way to get it, and being ready to adjust when conditions shift again. Because they will.
The conflict is reshaping energy markets in real time. The labs are building models they cannot yet afford to deploy. And somewhere in between, there is a window for businesses that are willing to be honest about what they actually need and disciplined enough to act on it.
That window is about 90 days.
All the Zest 🍋
Cien