Weather forecasting past about ten days is basically astrology with better graphics. The atmosphere is chaotic, small errors compound, and anyone telling you what next month looks like is selling confidence, not information. Which is roughly where we are with Nvidia projecting an AI-driven sales surge that runs through 2028.
That’s a three-year forecast in an industry where the product roadmap changes every eighteen months and the customers are a handful of companies that could collectively decide to slow down over a single bad quarter of earnings calls.
What Was Actually Said
Bloomberg’s coverage is the headline itself: Nvidia expects the AI-fueled sales surge to extend into 2028. That’s the fact. Not a breakdown by segment, not a customer-by-customer commitment schedule, not a hedge about macro conditions. A number on a horizon.
I review AI tools for a living, which mostly means separating what a vendor demoed from what the vendor shipped. The same discipline applies here. Nvidia has earned enormous credibility by being right about demand repeatedly. Being right repeatedly is also exactly how forecasts stop getting scrutinized.
The Counter-Forecast Nobody Put in a Headline
Two data points arrived in the same news cycle and pointed opposite directions. SanDisk fell 54%, and Morningstar’s read is that the AI boom is finite, with a downturn projected before 2029. Meanwhile, JLL published its 2026 Global Data Center Outlook, which is the kind of document that exists because a lot of capital is being committed to physical buildings on multi-year timelines.
Put those side by side and you get a genuinely interesting tension:
- Nvidia’s forecast says the surge runs through 2028.
- Morningstar’s view says a downturn arrives before 2029.
- Data center capacity planning is happening on timelines that stretch past both.
Notice that Nvidia’s projection and Morningstar’s warning are not actually contradictory. They’re nearly the same statement with different emotional framing. “Growth through 2028” and “downturn before 2029” can describe an identical curve. One party is selling the climb and the other is pointing at the cliff.
That’s not a gotcha. It’s a reminder that a forecast’s tone tells you more about who’s speaking than about what’s coming.
Why This Matters If You Buy AI Tools
Here’s where I’ll be more direct than the equity analysts, because my readers aren’t managing portfolios. They’re deciding whether to sign an annual contract with an AI agent startup.
Every AI tool you evaluate sits on top of compute someone else is paying for. When that compute is cheap and abundant, you get generous free tiers, unlimited-feeling usage caps, and startups that can afford to eat inference costs while chasing growth. When compute tightens or the funding environment cools, those same tools quietly introduce usage limits, raise prices, or shut down.
So a hardware sales forecast is not an abstraction to you. It’s the weather over the entire tool market you’re shopping in. Practical implications:
- Treat generous pricing as promotional until proven otherwise. If a tool’s economics only work with subsidized inference, the pricing page you signed up on is temporary.
- Prefer month-to-month over annual when the vendor is young. The discount for annual is rarely worth being locked into a company that might reprice mid-contract anyway.
- Ask about data portability before you need it. The tools most likely to disappear are the ones that made switching hardest.
- Watch for vendors whose pitch depends on compute getting cheaper forever. That’s a bet, not a plan.
My Honest Read
I don’t think Nvidia is lying. I think Nvidia has better visibility into order books than anyone writing about Nvidia does, and that visibility genuinely extends further than the average company’s. Multi-year data center construction, which JLL is documenting, creates real committed demand that doesn’t evaporate on a bad news day.
I also think a three-year forecast from the primary beneficiary of a boom deserves the same skepticism I’d apply to any vendor forecasting its own category. Not dismissal. Skepticism. The difference matters.
The SanDisk drop is the more instructive story, honestly. A 54% decline is what it looks like when the market decides one link in the AI supply chain isn’t as protected as it appeared. That can happen to any link. It’s happened before to companies that also had solid multi-year outlooks.
What I’d actually watch isn’t whether Nvidia hits 2028. It’s whether the AI tools you use start behaving like companies that expect the money to keep flowing, or like companies quietly preparing for the weather to turn. The second group is easier to spot than you’d think. Check your usage limits this month against last month.
🕒 Published: