If you've been to a data conference or opened a data newsletter this year, you've been told about the "modern data stack" — a constellation of cloud-native tools that, assembled, supposedly represents a generational leap in how we do data. A cloud warehouse at the centre, tools to load data into it, tools to transform it once it's there, tools to visualise it, tools to push it back out. It's the phrase of the year, and it arrives with the confident implication that if you're not on it, you're doing data wrong. So let me do my usual thing and sort the genuine improvement from the fresh coat of venture-funded paint, because there's real substance here and real hype, and telling them apart saves you money and disappointment.

What's genuinely real and good

I want to lead with the real, because there is a lot of it, and dismissing the whole thing as hype would be as wrong as swallowing it whole.

  • Cloud warehouses genuinely changed the economics. The thing at the centre of the modern stack — a scalable cloud warehouse that separates storage from compute — is a real advance. It removed a whole category of pain: the capacity planning, the "we've outgrown the box," the coupling of storage and compute that made scaling miserable. That's not marketing; that's a genuine improvement I feel in daily work.
  • ELT over ETL is a real reframe, and mostly a good one. The modern stack flips the old order: instead of transforming data before loading it (ETL), you load the raw data into the powerful warehouse first and transform it there (ELT). Because the warehouse is now cheap and scalable enough to handle it, this is often genuinely better — more flexible, easier to reprocess, less brittle. Real idea, real benefit.
  • Bringing software discipline to transformation is overdue and welcome. The rise of treating your data transformations as code — versioned, tested, reviewed, with proper engineering practices — is a genuine maturation. Data transformation used to be a wild west of undocumented SQL; the modern stack's insistence on treating it like real software is a real and valuable culture shift.

So: the cloud warehouse, ELT, and software discipline for transformation are genuine, and I'd not want to go back. If that's all "modern data stack" meant, I'd be an unqualified fan.

What's mostly hype

But "modern data stack" doesn't just mean those good ideas — it's become a marketing constellation, and around the real core orbits a fair amount of froth worth naming.

  • The tool proliferation is a solution and a new problem. The modern stack is many specialised tools, each doing one slice, each venture-funded and marketed as essential. Assemble them and you've solved some old problems and bought a new one: a sprawl of tools to integrate, pay for, and keep working together — the exact "fragile seams between services" problem I keep running into, now sold to you as modernity. Every tool is real; the number of them is partly hype, because a lot of teams need far fewer than the ecosystem implies.
  • "It fixes your data problems" is the oldest lie, relabelled. The implicit promise — adopt this stack and your data will finally be good — is the same promise every generation of data tooling has made and none has kept, because your data problems were never mostly tooling problems. The disagreements about what "revenue" means, the ungoverned sprawl, the gap between the data and the decision — the modern stack doesn't touch any of that, because those are human and organisational problems that no constellation of cloud tools can fix. Selling technology as the answer to a human problem is the eternal hype, wearing this year's logos.
  • The novelty is oversold. A lot of what's branded "modern" is a genuinely nice cloud-native repackaging of ideas that aren't new — staging and transforming and serving data in layers is old practice (I've said the same about "medallion" branding). The tools are better; the concepts are largely the sensible things good data people were already doing. "Modern" is doing a lot of marketing work.

How to be a grown-up about it

So what's the sane posture, if it's neither "adopt the whole stack because it's modern" nor "dismiss it as hype"? The same posture I try to bring to every wave: take the real advances, ignore the marketing, and never mistake tooling for the actual problem.

Concretely: adopt the cloud warehouse and the ELT model and the software-discipline culture, because those are genuine and good. Be ruthlessly sceptical about the tool proliferation — every additional specialised tool is real cost and real integration burden, and most teams need far fewer than the ecosystem is selling; add one only when a concrete pain justifies it, not because it completes the fashionable stack. And hold onto the hard truth that no stack, however modern, fixes the problems that actually make data hard, because those live in people and organisations, not in tools.

The modern data stack's real advances are the cloud warehouse, ELT, and treating transformation as code. Its hype is the implication that assembling enough venture-funded tools will fix problems that were never about tools. Buy the first. Ignore the second. And remember your definitions were never going to agree themselves just because you bought a warehouse.

The honest bottom line

The modern data stack is a genuine step forward wrapped in an oversold story, which is true of most things in this field and is exactly why the hype-check matters. The cloud-native core is real and worth adopting. The endless tool proliferation is a cost dressed as progress. And the deepest promise — that this finally makes your data good — is the same mirage every tooling generation sells, because good data was always mostly a matter of human agreement, governance, and communication that no stack delivers.

Adopt it for what it genuinely improves, with clear eyes about what it can't, and you'll get real value and avoid the expensive disappointment of the teams who bought the whole constellation expecting it to fix problems that were never technical. Modern is good where it's genuinely modern. It's just not magic, and the parts sold hardest as magic are the parts that aren't real.