Gold is trading around $4,489 per ounce as of this week - roughly $1,100 below its January 2026 peak of $5,602, but still more than double where it sat a year ago. J.P. Morgan sees it pushing toward $5,000 by the fourth quarter of 2026.
That number alone is not the point. The point is what gold's two-year parabolic run forces me to confront as an AI infrastructure investor: is money leaving the compute trade and parking in bullion? Or is this a completely different beast wearing the same safe-haven costume?
I believe the answer is the latter - and understanding why changes how you manage your AI allocation when gold headlines dominate.
The buyer is the signal
Here is the fact that separates gold from every other safe-haven narrative: Goldman Sachs updated its forecast on May 18, expecting central banks to buy an average of 60 tonnes of gold per month through 2026. That's not a speculative number - it's based on Q1 purchases that already clocked 244 tonnes, up from 237 tonnes in Q4 2025. Poland alone added 12 tonnes in April, pushing its reserves to a record 543 tonnes.
Put plainly: the margin buyer on gold is not a hedge fund rotating out of tech. It is sovereign nations diversifying reserves away from the dollar, amid a widening Middle East conflict and persistent geopolitical fragmentation. Central banks don't trade on 4-hour charts. They don't front-run earnings. They don't care about Nvidia's data center revenue guidance.
This is what separates gold's current cycle from the gold rallies of 2008 or 2013. In those episodes, retail panic and fund flows drove price - and when panic faded, price followed. This time, the structural buyer base is institutional in a different sense: nation-states with multi-year reserve rebalancing mandates. That demand floor doesn't get cancelled by a strong Nasdaq quarter.
The AI capex budget doesn't overlap
On the other side, the AI infrastructure spending machine is operating on a different register entirely. The four dominant hyperscalers - Microsoft, Alphabet, Amazon, and Meta - are collectively planning to spend between $630 billion and $700 billion on capital expenditures in fiscal 2026. That's up from roughly $300 billion in 2024. Quarterly earnings reports show combined capex exceeding $130 billion per quarter with no clear end in sight.
These budgets don't compete with gold. A treasury team at Microsoft doesn't look at gold breaking $4,500 and say "we should build fewer data centers." The AI capex commitment is contract-driven, multi-year, and already baked into multi-billion-dollar deals with Nvidia, TSMC, and equipment suppliers. Gold price action has zero transmission mechanism into the hyperscaler capital allocation process.
This is the structural distinction the market keeps missing: gold and AI infrastructure are running on parallel capital rails that rarely intersect.
Where the risk actually is
However, this doesn't mean gold is irrelevant to the AI portfolio. The connection runs through investor psychology, not through supply chains.
Here is what I watch: when gold prints headlines - $4,500, $5,000, new all-time highs - it triggers the rotation reflex in a specific audience. Retail investors and smaller funds see a parabolic asset they don't hold and start questioning whether their AI positions are too concentrated. The Nasdaq has already tested correction territory in Q1 2026, and some of that pullback coincided with accelerating safe-haven flows.
Goldman Sachs noted in mid-May that the recent tech stock pullback was "an up crash" - their term for the kind of correction that precedes further gains rather than marks a trend break. That's one perspective. But what matters for me is whether gold headlines are the reason I'm second-guessing my AI thesis. If the answer is yes, I'm reacting to the wrong buyer.
The central bank buying gold is not the same money that would be flowing into Nvidia or AMD. The AI capex budget is not shrinking because sovereigns want more bullion. These are separate systems.
The allocation question
So what does this mean for where capital goes?
I don't think gold demands a trim in AI positions. The AI thesis - hyperscaler capex acceleration, training-to-inference transition, architectural generation gaps between chip competitors - is driven by product cycles and enterprise adoption curves, not by what gold does on a Tuesday. I've been clear about this before: the debate is not whether a strong asset remains important. It is whether the return profile is still as compelling as what you can find elsewhere.
Gold is compelling - but it's compelling for a different reason, to a different audience, on a different time horizon. Central bank buying gives gold a structural floor that most equities don't have. For a portfolio hedge, that's valuable. But it doesn't replace the return engine of a company that's embedded in $700 billion of annual capex commitments with multi-year supply contracts.
What would change my view? If I started seeing hyperscaler capex guidance actually softening - not stock price softening, but the companies themselves saying "we're slowing the pace" - that would be the signal. Until then, gold hitting $5,000 doesn't break the AI trade. It just means the world is hedging against risks that AI infrastructure doesn't solve - and that's fine. Some of your capital should be on the wrong side of those risks.
The question isn't "gold or AI." It's whether you're holding both in proportions that match your actual conviction - not your fear of headlines.