VAST Data CEO Renen Hallak describes a single customer whose three-year storage forecast moved by a factor of roughly five in one quarter — and the structural problem that number exposes has nothing to do with forecasting skill.
On Matt Turck’s podcast — the single source for everything that follows, relayed and not independently verified — VAST Data founder and CEO Renen Hallak described a conversation with a customer he identifies only as “one of these AI clouds, one of the smaller ones.” A quarter ago, VAST asked this unnamed customer to plan three years out, explicitly because of supply-chain lead times. The customer’s answer: roughly 500 petabytes over the next three years.
Last week, the same customer returned. The new ask was an additional two exabytes on top of that 500 petabytes. Using the decimal convention storage vendors apply — where one exabyte equals 1,000 petabytes — that additional request is itself 2,000 petabytes, which puts the revised total at roughly five times the original figure. That arithmetic is this publication’s, not Hallak’s characterisation. The customer is unnamed and will not be named, guessed at or hinted at here. One customer is not a market, and nothing in this article extrapolates these figures into an industry demand rate, a growth estimate or a market size.
Hallak’s own reaction was candid: “I’m not sure it gives me comfort and sometimes it scares me as to how much demand is there,” and “every quarter our customers are coming back to us and telling us they need a lot more than they thought they did.” Those are the only quotations from the podcast used in this piece. The word “scares” is a planning observation — addressed directly below — and not a market call. It does not express doubt that the demand is real; on Hallak’s account, it plainly is.
The key insight: A three-year forecast that moves by a multiple in a single quarter is not evidence of bad forecasting. It is evidence that the commitment horizon and the information half-life are running at incompatible speeds — and no amount of analytical skill closes that gap.
The supply chain needs three years of notice. The number moved in one quarter.
The Structural Read
The failure mode here is architectural, not analytical. Supply chains with long component lead times require planning horizons measured in years — which is precisely why Hallak says VAST asked its customer to look three years out in the first place. The demand signal that long horizon depends on is being revised quarterly, and revised by multiples. Those two clocks are structurally incompatible, and the customer did the responsible thing: it gave a three-year number when asked. The number still moved.
It is important to be explicit about what a forecast revising upward does and does not tell you. It is not evidence of a problem. It is equally consistent with demand being genuinely larger than anyone estimated — which is exactly how Hallak frames it. Volatility in an estimate tells you the estimate is unreliable. It does not tell you which direction the truth lies in. No conclusion is drawn here about whether any buildout is justified, and nothing here predicts bubble, overheating, shortage or glut.
This is also the demand-side twin of a separate structural problem in AI infrastructure: long-dated compute contracts being underwritten against residual values that nobody has yet observed, because no generation of this hardware has been watched all the way to retirement. Set the two together and the shared property becomes clear. Underwriting any long-dated commitment requires stable estimates at both ends — what the asset will be worth when the term expires, and how much of it will be wanted along the way. On this account, one of those ends has no history to draw on at all, and the other is revising by multiples every quarter.
Structural Pattern
The Two-Clock Problem
A supply chain commits resources on a horizon measured in years. The demand signal it is planning against revises on a horizon measured in quarters — and revises by multiples. No forecasting method resolves this, because the problem is not that the estimate was wrong. It is that the horizon of the commitment exceeds the half-life of the information it rests on. The customer acted responsibly. The clocks are simply incompatible.
Renen Hallak — Matt Turck’s Podcast
“I’m not sure it gives me comfort and sometimes it scares me as to how much demand is there.”
On the word “scares”: it is easy to misread as bearishness. It is not. For a supplier, demand above plan is not free. Unexpected upside consumes inventory, component allocation and delivery slots already promised to other customers — which means the upside arrives as an operational liability some time before it arrives as revenue. You have to source it, build it and ship it, and the customers you cannot fully serve in the meantime are real. A supplier and an investor looking at the same demand curve are looking at two different problems. Hallak’s remark is about the first problem, not the second.
A separate and shorter note from the same conversation cuts against a common assumption. Hallak said that “other than as an investor, there’s no legal document between us and NVIDIA” — NVIDIA does not resell the product and the two do nothing together from a legal standpoint — and yet he describes NVIDIA as their best partner by far, with a triple-digit number of developers collaborating on each side. The general property is worth stating plainly: a contract allocates risk and obligation. It does not create a collaboration. The two get conflated routinely because the contract is the only part of a relationship visible from outside. The absence of a formal agreement is not characterised here as risky, irregular or a problem — and the size and terms of the investment are not established and are not described.
Three Implications
IMPLICATION 1 — FOR SUPPLIERS
Demand above plan is an operational problem before it is a revenueopportunity. A supplier who cannot match upside demand with sourced inventory and available delivery capacity absorbs the liability of customer disappointment first. Planning buffers built for normal variance are structurally undersized when the unit of revision is multiples rather than percentages — and that is a capacity-management question, not a demand-quality question.
IMPLICATION 2 — FOR BUYERS AND PLANNERS
When the commitment horizon exceeds the half-life of the underlying demand signal, the responsible move is to build revision mechanisms into contracts rather than to ask for better estimates. A three-year number that revises by a multiple in a quarter is not a planning failure — it is a signal that the planning architecture itself needs to account for structural uncertainty. Locking in rigid multi-year quantities against a quarterly-revision environment transfers forecast risk without eliminating it.
IMPLICATION 3 — FOR ANYONE READING PARTNERSHIPS
The VAST–NVIDIA relationship — substantive collaboration, no reseller agreement, investment relationship, triple-digit developers on each side — is a useful reminder that the visible architecture of a partnership (contracts, press releases, co-sell agreements) is a poor proxy for the actual depth of it. Analysts and investors who map competitive positioning from public legal structures will systematically misread relationships where the collaboration is real but informal. The contract is the floor, not the ceiling.
Business Engineer Framework
The Map of AI — Where Storage Infrastructure Sits in the Stack
The two-clock problem described above is a layer-specific phenomenon. Storage infrastructure operates at a physical layer where lead times are measured in quarters and years. The demand it serves is generated at application layers where model architectures, training runs and inference workloads change on cycles measured in weeks. The Map of AI framework maps all nine layers of the stack — and makes visible exactly why commitments made at one layer routinely misprice risk generated at another. Understanding where VAST sits, and what sits above it, is how you read the structural tension clearly.
A single unnamed customer revising a three-year storage estimate by a multiple in one quarter is not a market data point — it is a structural signal about what happens when long-cycle physical infrastructure is asked to serve a demand environment whose half-life is measured in quarters. The forecast did not fail because the estimator was wrong. It failed because the commitment architecture was not built to accommodate the speed at which the underlying facts are changing. That is the problem worth sitting with — not the size of the number, but the speed of its revision.
Primary source: Matt Turck’s podcast (YouTube) — remarks by Renen Hallak, VAST Data CEO, relayed and not independently verified. The unnamed customer is not identified or speculated upon. One customer is not a market; no industry demand figure, growth rate or market size is implied. The ~5× multiple uses the decimal storage convention (1 EB = 1,000 PB) and is this publication’s arithmetic on Hallak’s figures, not his characterisation. A forecast revising upward is not evidence of a problem; it is equally consistent with demand being genuinely larger than estimated. Nothing in this article constitutes investment advice, and no prediction is made about any company, market or security.
Single source. Everything above is Renen Hallak’s account on Matt Turck’s podcast, relayed here and not independently verified. The customer is unnamed — he describes it only as one of the smaller AI clouds — and nothing above names, identifies or guesses at it. This is one customer and not a market. Nothing above turns it into an industry demand figure, a growth rate or a market size. The two exabytes are in addition to the original 500 petabytes; the “roughly five times” multiple is this publication’s arithmetic on the decimal convention storage vendors use, where one exabyte is a thousand petabytes, and is not his characterisation. A forecast that revises upward is not evidence of a problem — it is equally consistent with demand being genuinely larger than anyone estimated, which is how he frames it. His remark that the demand sometimes scares him is a planning observation, not a market call, and not doubt that the demand is real. His comment that there is no legal document with NVIDIA is not presented here as risky, irregular or a problem, and the investment’s size and terms are not established. Nothing above is investment advice, expresses a view on any company or security, or predicts anything. VAST’s revenue, bookings, backlog, pricing, margins, capacity and customer count, any contract, term or delivery date, and what the storage is for are not established and do not appear above.
Gennaro is the creator of FourWeekMBA, which reached about four million business people, comprising C-level executives, investors, analysts, product managers, and aspiring digital entrepreneurs in 2022 alone | He is also Director of Sales for a high-tech scaleup in the AI Industry | In 2012, Gennaro earned an International MBA with emphasis on Corporate Finance and Business Strategy.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
