CoreWeave In-Depth Value Research Report

Hunting ten-year five-baggers among great growth stocks — pressing the upside question: "Can it get much bigger?"

• QHow high is its market ceiling? Is it enlarging an existing pie, or creating an entirely new market?6/10

  A

  The ceiling is very high, and it has a dual nature of both "enlarging an existing pie" and "staking out a new form"; but CoreWeave is grabbing position in a race that is still expanding violently, rather than being a pioneer that has already defined the endgame.

  First, how big is the pie itself. What CoreWeave sells is "AI-native cloud": packaging high-density data centers, Nvidia GPUs, and in-house scheduling software into multi-year compute contracts. The reality of this demand no longer needs to be argued: the report discloses that company revenue rose from $229 million in 2023 and $1.915 billion in 2024 to $5.131 billion in 2025, with Q1 2026 revenue of $2.078 billion, up another 112% year over year (Company Q1 2026 results disclose revenue of $2.078B, doubling). What better illustrates the ceiling is contracted stock: as of the end of Q1 2026, unfulfilled RPO was about $98.8 billion and the company's own revenue backlog reached $99.4 billion (StockTitan cites Q1 2026 8-K: revenue backlog near $99.4B, including Meta's new $21B commitment), equivalent to a contract reserve of nearly five times 2025 full-year revenue. In terms of total contracted power, the company says it has expanded from over 850MW of active power at the end of 2025 to over 1GW active and over 3.5GW contracted in Q1 2026; power is the hardest physical constraint for AI data centers, and contracted power is itself a proxy variable for the ceiling.

  But one must honestly distinguish "enlarging the pie" from "creating a market." CoreWeave did not create new demand out of nothing; the pie of AI training/inference compute was enlarged jointly by Nvidia, OpenAI, and hyperscale cloud providers, and CoreWeave, within this already existing and rapidly swelling pie, grabbed a sizable share through "bringing the latest generation of GPUs online faster, in-house orchestration software, and switchable infrastructure for training/inference." What genuinely carries a hint of "new form" is carving "AI-dedicated cloud" out of the hyperscalers' general-purpose cloud as a standalone business; this is a redefinition of a niche market, not the creation of an entirely new market from scratch. The report's own characterization is restrained: it leads in the AI-native GPU cloud niche, but once the competitive boundary is widened to all of cloud computing, it faces "hyperscalers with far greater resources than its own" (Amazon, Google, Microsoft, Oracle).

  So the key to the ceiling is not "whether it is high enough," but "how large a share CoreWeave can hold under such a high ceiling." The race's ceiling may be AI infrastructure spending on the scale of hundreds of billions of dollars (CoreWeave's own 2026 capex guidance alone is as high as $31–35 billion, showing the industry's investment intensity), but this also means supply is expanding wildly and the strongest competitors hold lower funding costs and more complete ecosystems. From a Baillie Gifford perspective, "enlarging an existing pie" is actually higher risk, because the endgame share is undetermined and the possibility of price wars and customer internalization always hangs overhead. The conclusion is: this is a genuinely long, snowy slope, but CoreWeave is a strong executor in the race, not the definer of the ceiling; how much of the ceiling's dividend it can enjoy depends on whether its moat can transform from "first-mover execution" into a "structural barrier," and that has not yet been proven.

  Jun 10, 2026
• QCan its revenue at least double over the next five years? Is growth driven mainly by volume, price, or new businesses?7/10

  A

  It can almost certainly double, and most likely far more than double; growth is driven mainly by "volume" (newly added compute capacity coming online), with "price" a secondary or even pressured factor, and "new businesses" (the software layer) currently a garnish.

  First, the certainty of doubling. This is the question CoreWeave answers most decisively among the ten. The company's own 2026 full-year revenue guidance is $12–13 billion; relative to 2025's $5.131 billion, in 2026 alone it will grow to 2.3–2.5x—the hurdle of doubling in five years is essentially meaningless for it. What supports this growth is not expectation but signed contracts: as of Q1 2026, unfulfilled RPO was about $98.8 billion and revenue backlog about $99.4 billion (Company Q1 2026 discloses backlog near $99.4B), of which 36% is expected to be recognized within the next 24 months, and that portion alone is about $35.6 billion. In other words, a fairly large chunk of the high growth of the next two to three years has already been "locked in" by contracts.

  Next, breaking down the drivers of growth—this is the "growth quality" Baillie Gifford truly cares about:

  • Volume is the absolute mainstay. CoreWeave's revenue ≈ deployed compute capacity × utilization × unit price. Its revenue surge is essentially the physical expansion of data centers and GPU clusters: active power expanded from over 850MW at the end of 2025 to over 1GW in Q1 2026, with total contracted power over 3.5GW. To turn this contracted power into revenue, it must keep bringing equipment online, which is exactly why it plans to pour in $31–35 billion of capex in 2026. Growth is a capital-heavy model of "trading money (and debt) for capacity, and capacity for revenue."

  • Price is a pressured factor, not a tailwind. This must be honest. The report notes that GAAP gross margin has fallen from 74.3% in 2024 to 65.5% in Q1 2026, and Adjusted EBITDA margin also declined from a higher level to 56% in Q1 2026; "margins compress" was a market focus for the quarter. The report judges that the company's high past profitability "more likely came partly from the dividend of supply-demand mismatch," and once GPU supply eases, unit price is more likely to be a headwind. So growth relies not on price increases; on the contrary, it wins through volume in an environment where prices may loosen.

  • New businesses (the software layer) are currently strategic positioning rather than a growth engine. The company acquired Weights & Biases in 2025, and also acquired Monolith and Marimo, aiming to extend from infrastructure into higher-value-added software layers. The direction is right, but the scale is still far from enough to drive overall revenue: revenue is still 98% from committed contracts (the report discloses that in 2025 and Q1 2026, committed contracts accounted for about 98% of revenue in infrastructure contracts), and the software layer has not yet become a "second engine" capable of supporting a doubling on its own.

  To sum up in one line: doubling in five years is not the problem; the problem is whether this "volume-driven, capital-hungry, price-pressured" growth can, at some point, switch from "the more it grows the more it burns" to "the more it grows the more it earns." Baillie Gifford wants growth stocks that can deposit high growth into owner earnings; CoreWeave's growth engine is strong and real, but the structure of its drivers (volume-driven, capital-heavy, price-pressured) means its growth quality takes a discount: the growth rate is an A, but growth quality can only get a B for now.

  Jun 10, 2026
• QFive years from now, what will take over as the next growth engine? Does this "second curve" exist today?4/10

  A

  The "second curve" today can only be counted as seeds planted, not yet sprouted; the most promising successor is the value-chain move up from "selling compute to selling software/platform," but for now it is neither large enough nor proven able to profit independently. In other words, the first curve (compute expansion) is still in an accelerating phase, and the second curve has not yet reached the point where it needs to take over; but this is precisely the weakest link in CoreWeave's long-term narrative.

  First, define the problem. Baillie Gifford's question of "what takes over five years from now" is especially pointed for a capital-heavy company whose growth relies mainly on physical capacity expansion, because capacity expansion has limits: power has a ceiling, capital has a cost, and GPU generations iterate. When the backlog has been converted, and when AI compute supply and demand move from tight to balanced, growth from simply "racking up more" will slow. What sustains it then?

  CoreWeave currently has three second-curve candidates, ranked by credibility:

  First, moving up the value chain—extending from the infrastructure layer to managed software and application software layers (the most hoped-for, but the earliest-stage). The report discloses that the company's platform was designed as three layers from the start: infrastructure services, managed software services, and application software services; the in-house CoreWeave Mission Control, CKS, and bare-metal capabilities are the middle layer, with acquired developer tools such as Weights & Biases layered on top. In 2025 the company also acquired Monolith and Marimo, with the clear intent of climbing toward high-value-added software. The appeal of this curve is that the software layer is asset-light, high-margin, and has strong switching costs; if it can be pulled off, it could upgrade CoreWeave from a "capital-heavy compute lessor" to an "AI development platform," fundamentally changing the nature of the business. But today it is still just a seed: revenue is still 98% from committed contracts (Company Q1 2026 discloses committed contracts at 98%), the software layer's contribution is not yet at scale, and there is no evidence it can profit independently of the underlying compute.

  Second, inference demand taking over from training demand. This is an industry-level potential second curve, but for CoreWeave it is a double-edged sword. When AI switches from "the capital-intensive boom of training large models" to "inference-first," inference workloads are more distributed, more latency-sensitive, and lower in capital intensity per unit of compute. The report lists exactly this as a risk: "AI demand shifting from the training boom to lower-capital-intensity inference or in-house customer capability, driving down utilization of high-priced capacity" would trigger a reassessment. So inference could be a new growth pole or the killer of its high-priced capacity utilization, depending on whether CoreWeave can successfully turn infrastructure optimized for training toward being optimized for inference (the company does say it "provides switchable infrastructure for training and inference," but this is only a statement, not yet verified through a cycle).

  Third, geographic and customer expansion. The backlog is diversifying from sole reliance on Microsoft toward OpenAI (about $22.4 billion), Meta (a new commitment in March 2026 of $21B), Anthropic, Jane Street, and others. A widening customer base can itself extend the first curve, but it is still "selling the same business to more people," and does not qualify as a second curve of a different nature.

  The honest Baillie Gifford conclusion: CoreWeave's second curve (moving up the software layer) is directionally correct and logically sound, but today it remains at the "acquire-and-integrate + strategic-statement" stage, with no quantifiable revenue evidence and no proof it can profit independently. Compared with the great growth stocks whose second curves are already clearly visible, CoreWeave's long-term growth five years out depends heavily on the unverified bet of "whether the software transformation can be delivered"; and until the transformation is delivered, its growth is essentially an extension of the first curve of "buying capacity with capital." This is one of the important reasons it does not deserve the "simple and great" label.

  Jun 10, 2026
• QWhat is its core competitive advantage? Will this moat widen or narrow over the next three to five years?3/10

  A

  The core advantage is the stage-specific supply/scale lead created by "speed + execution": bringing the latest GPUs online faster, in-house orchestration software, and strong financing and build-out capability; but this moat is more like a "sandcastle" built on the current supply-demand tightness and first-mover advantage, and over the next three to five years it will most likely narrow rather than widen.

  First, be clear about what it actually has today. After breaking the moat down item by item, the report gives a "moat strength score" of only 2.5/5, with a status of "still forming, but not verified as able to widen over the long term." The most credible advantages concentrate in three places:

  • Speed advantage: CoreWeave's foundation is being "born for AI," able to bring the latest generation of Nvidia GPUs into production faster than general-purpose cloud. In a window where GPUs are in short supply and whoever racks first eats first, this is a hard-cash advantage; the company being named NVIDIA Exemplar Cloud by Nvidia in 2025 is also a side endorsement.

  • Execution/operational culture: The report considers this "the most credible item for now"; the company has indeed demonstrated the ability to bring huge amounts of capital into the business and complete data-center deployment quickly. From 43 data centers and over 850MW at the end of 2025 to over 1GW active and over 3.5GW contracted in Q1 2026, the pace of expansion is itself proof of execution.

  • Financing channels: The capital market's recognition of it is part of a real barrier. In January 2026, Nvidia privately placed about 22.958 million shares at $87.20/share, injecting about $2 billion; in March 2026 it secured a $8.5 billion DDTL 4.0 non-recourse delayed-draw term loan, and that financing received an A3/A(low) investment-grade rating, the first investment-grade GPU-collateralized financing. Being able to finance GPU assets at investment-grade cost is indeed a rare capability among peers.

  But why will this moat narrow over three to five years? The report's argument is solid, and I fully agree:

  First, the moat is not sufficiently "self-generating." It is built partly on the two external conditions of "current AI compute supply-demand tightness + its own first-mover execution." Once supply and demand ease, the first-come-first-served speed advantage depreciates. The report states plainly that the company's high past Adjusted EBITDA "more likely came partly from the dividend of supply-demand mismatch and accounting conventions, rather than from mature structural pricing power."

  Second, the strongest competitors happen to have exactly what it lacks most. CoreWeave's moat weaknesses are network effects (very weak), switching costs (moderate, far from locked in), and brand (an industry brand rather than a consumer brand). Its competitors are hyperscalers—Amazon, Google, Microsoft, Oracle—that "have more capital, lower funding costs, and more complete ecosystems." They can both press on price and tie customers with integrated ecosystems, and CoreWeave's speed advantage will be diluted once competitors get serious.

  Third, it is bottlenecked at both ends. On the supply side, the report discloses that all of the company's current GPUs are Nvidia GPUs, and many customers directly specify Nvidia in their contracts; this means its "supply advantage" is really the advantage of "Nvidia being willing to supply it," with half the bargaining power in Nvidia's hands. On the demand side, the large customers it is deeply tied to—Microsoft, Meta, OpenAI—all have the ability to build their own data centers or switch across multiple clouds, and can pull their workloads back at any time. A moat whose upstream and downstream gates are both not fully in its own hands can hardly be called "durable."

  Baillie Gifford wants a moat that "widens over the next three to five years." CoreWeave's assessment is the opposite: right now, when supply and demand are tightest and competitors have not yet fought at full force, its execution moat looks very effective; but as supply expands, competitors push, and customers internalize, it is more likely to be eroded than widened. What could truly widen the moat is whether it can successfully complete the transformation to the software/platform layer, upgrading customers from "renting compute" to "depending on an entire tool chain"—but that path (see the second-curve question) has not yet worked out today. So the honest answer to this question is: there are advantages, but their nature leans "cyclical and stage-specific" rather than "structural and self-reinforcing"; the trend judgment is that the risk of narrowing outweighs the opportunity of widening.

  Jun 10, 2026
• QIf its core business is disrupted, does it have the genes to reinvent itself? How does it treat mistakes and bad news?4/10

  A

  The genes for self-reinvention are "showing signs, but unverified": the company has indeed demonstrated one successful transformation from crypto mining to AI cloud, and is actively extending into the software layer; but its way of treating bad news is still "continuous disclosure yet slow to resolve," and in particular the material internal-control weaknesses remain unremediated to date, which puts a question mark over its ability to "face mistakes squarely and self-correct."

  This chained question requires two implicit premises: first, whether the company has "the genes for self-reinvention" if its core business is truly disrupted; second, how it treats mistakes and bad news day to day—the latter being the everyday rehearsal for the former.

  First, the positive evidence for "self-reinvention genes." The most persuasive point about CoreWeave is that it is itself the product of a successful reinvention: the company's predecessor did cryptocurrency mining, and it was precisely because the founding team keenly sensed AI training demand in the GPU asset and decisively turned the entire company toward AI cloud that today exists. This history proves the team is not attached to old businesses and dares to reallocate assets to new tailwinds—exactly the "regenerative capability" Baillie Gifford values. Continuing to today, the company is also actively positioning for the next evolution of form: in-house Mission Control and CKS orchestration software, acquiring Weights & Biases, Monolith, and Marimo to climb toward the software layer, and describing itself as "providing switchable infrastructure for training and inference"—all of which leave a fallback in advance for "the core compute business being disrupted" (such as training demand shrinking or the inference paradigm changing).

  But "whether it can reinvent when disrupted" is highly questionable, because of the physical nature of the business. The crypto-to-AI reinvention succeeded because the core GPU asset happened to be reusable across scenarios. But if future disruption comes from GPU generational iteration being so fast that installed assets depreciate, or from AI shifting to lower-capital-intensity inference / in-house customer capability, what CoreWeave holds are heavy assets entirely betting on Nvidia GPUs—the report discloses that technology equipment has an estimated life of only about 6 years, software about 3–6 years, and many data-center build-outs last only 12 years or less. The combination of heavy assets + high leverage means the cost and speed of "changing course" are far worse than for an asset-light software company. A company carrying Q1 2026 net debt of about $22.6–24.9 billion (the report's figure is $24.859 billion) may not have the financial room to reinvent calmly when the cycle reverses—more likely being forced to raise capital at deep discounts or dilute shareholders. So its "reinvention genes" are real on the asset-light dimension (team instinct, software positioning), but limited on the heavy-asset dimension (a big ship is hard to turn, heavy debt weighs it down).

  Now look at "how it treats mistakes and bad news," where the answer is more subtle: candid disclosure, but slow correction. On the positive side, the report assesses that the company "generally does not shy away from problems in its risk disclosure": customer concentration, supplier concentration, material internal-control weaknesses, and high-leverage risk—it continuously and truthfully discloses all of them, without hiding anything. This transparency is itself a plus for governance quality. But on the flip side, candid disclosure ≠ effective resolution:

  • The company has identified material weaknesses since its 2024 annual report, and the report clearly notes that these weaknesses "continued through the end of 2025 and March 31, 2026, without remediation being completed." A mistake disclosed continuously for over two years and still not fixed indicates a shortfall in correction execution, not a problem of attitude toward "facing mistakes."
  • In January 2026 the company also faced securities litigation alleging that it made false and misleading statements (the report notes it is still an allegation, with no court finding yet). This at least shows that outsiders have doubts about the quality of its disclosure.

  The Baillie Gifford wrap-up: CoreWeave has the reinvention genes to "switch tailwinds" (crypto→AI has been verified once), and it has the awareness to actively evolve toward the software layer—this is its good side; but its way of treating its own mistakes is "seen, stated clearly, yet slow to fix" (the internal-control weakness has dragged on for over two years unresolved), plus heavy assets and high leverage limit its room to turn around if truly disrupted. A great growth stock needs not only "the courage to transform" but also "the ability to correct quickly under pressure without harming shareholders"—on the latter, CoreWeave has yet to produce reassuring evidence. On this dimension I give a "neutral-to-weak": the genes are there, but discounted by the nature of the business and execution shortfalls.

  Jun 10, 2026
• QDoes management (especially the founders) have long-term vision and interests deeply aligned with the company? Are they willing to sacrifice current profit for five to ten years out?6/10

  A

  The founders' long-term vision and control are real, and their interest alignment is indeed deep: the three co-founders firmly hold the helm through Class B super-voting power; the company also clearly "sacrifices current profit for long-term reinvestment" (continuous heavy-loss expansion, no dividends, no buybacks). But this "long-termism" is a double-edged sword for ordinary shareholders: governance checks are weak, and a large part of the "current profit" it sacrifices is borne by ordinary shareholders through dilution and added leverage.

  First, the positive facts on alignment and long-term vision. The report discloses that the company is led by three co-founders: Mike Intrator (CEO), Brian Venturo (Chief Strategy Officer), and Brannin McBee (Chief Development Officer); they hold super-voting power through Class B stock with 10 votes per share, and as of March 31, 2026 they still held all outstanding Class B shares, continuing to control key voting power at the company. This structure does have positive implications for "long-term orientation": the founders will not be forced by short-term price swings or activist shareholders into killing-the-goose decisions, exactly the "founder at the long-term helm" trait Baillie Gifford prefers.

  On the question of "whether it is willing to sacrifice current profit for five to ten years out," CoreWeave's answer is an emphatic "yes, and it is doing so on a large scale":

  • The company clearly prioritizes reinvestment over dividends and buybacks. The report's "rough free cash flow" was deeply negative in 2023, 2024, 2025, and Q1 2026 (about -$7.2 billion in 2025 and about -$4.7 billion in Q1 2026), because it poured operating cash flow and large amounts of external financing entirely into capex—2026 capex guidance is as high as $31–35 billion. This is unquestionably "sacrificing the present, betting on the future."
  • It is even willing to borrow specifically to expand for a single large-customer contract—the $8.5 billion DDTL 4.0 loan was established "mainly to fulfill a certain customer contract, purchasing GPU servers and related infrastructure." This "build capacity first, tie down long-term contracts" approach is a hallmark of long-term vision.

  But the essence of Baillie Gifford's question is not just "willing to sacrifice or not," but "whose profit is sacrificed, and for whom is value created"—here CoreWeave's answer is not so pretty, and three points must be honestly noted:

  First, governance checks are weak. The report states plainly that the Class B super-voting structure "means weaker governance checks and balances for ordinary shareholders." When the founders both hold absolute voting power and make aggressive high-leverage + high-dilution decisions, ordinary shareholders have almost no brake.

  Second, the "sacrificed current profit" is largely paid for by ordinary shareholders. The company's growth "depends heavily on continued external financing, and shareholder dilution is a real rather than theoretical risk"—Nvidia's private placement of about 22.958 million shares at $87.20/share in January 2026 was real dilution. The founders lock in control through Class B (their voting power is not diluted), while the equity financing needed for expansion dilutes the economic interest of ordinary Class A shareholders—the "bet" of long-termism is funded by all shareholders, while control is highly concentrated in the founders' hands, and this asymmetry is something ordinary shareholders must watch out for.

  Third, strong execution ≠ capital allocation proven rational. The report gives "management and capital allocation" only 2.5/5, and judges integrity and governance quality as "moderately below average." The reasoning is clear-eyed: the team's execution in financing, contracting, and construction is beyond doubt, but it "has not yet proven it can convert rapid expansion into sustained growth in per-share intrinsic value, rather than simply making scale ever larger, debt ever higher, and ordinary shareholders ever more diluted." Adding the material internal-control weakness that has continued unremediated for over two years and the January 2026 securities-litigation allegation, one can only give "uncertain" on "whether management can be trusted."

  The Baillie Gifford conclusion: this is a founding team with long-term vision, willing to sacrifice the present for the future, and with interests and control highly aligned—consistent with Baillie Gifford's preference for "founder at the helm." But its long-termism carries the structural flaw of "concentrated control + economic interest supported by diluting ordinary shareholders to fund expansion," plus unrepaired internal controls and unverified capital-allocation returns. So on this dimension I give "neutral": the long-term alignment is real, but the cost of "sacrificing current profit for five to ten years out" falls disproportionately on ordinary shareholders, and whether this sacrifice can produce per-share value growth remains unanswered.

  Jun 10, 2026
• QIf it disappeared tomorrow, how much would customers miss it? Is its growth model sustainable and free of reliance on harming society and regulation?5/10

  A

  A group of large customers would miss it quite a bit in the short term, because migrating compute contracts of hundreds of MW carries real switching costs and time costs; but its "indispensability" is "indispensability within the contract term," not "irreplaceability." The growth model itself does not harm society, and regulatory sustainability is passable; the biggest sustainability variables are energy/power and export controls, rather than ethical or compliance blemishes.

  This chained question requires two layers: first, "indispensability"; second, "whether the growth model relies on harming society and regulation"—both must be answered honestly.

  First, "if it disappeared tomorrow, how much would customers miss it"—the answer is "much in the short term, replaceable in the long term." CoreWeave's stickiness with customers comes from three places: (1) it locks up customers' scarcest resource—compute capacity that is already racked, already powered, and fitted with the latest Nvidia GPUs, which in a tight supply-demand period is something "you may not be able to buy immediately even with money"; (2) large customers deeply tied to its infrastructure and tool chain have moderate switching costs, with the report judging switching costs to be moderate "especially for large customers deeply tied to its infrastructure and tool chain"; (3) customers have signed multi-year contracts on a take-or-pay basis, with a weighted-average contract term of about 5 years, so within the contract term customers "missing it" is written into the contract.

  But "indispensable" must be discounted, precisely because of who the customers are. The report discloses that in 2025 Microsoft, a single customer, accounted for about 67% of revenue, and in Q1 2026 "Customer A" still accounted for 45% of revenue—yet customers like Microsoft, Meta, and OpenAI have the ability to build their own data centers or switch across multiple clouds. The main reason they use CoreWeave is that their own capacity cannot keep up with demand, so they outsource temporarily; once their own capacity catches up or industry supply eases, how much they "miss" CoreWeave would fall rapidly. In other words, what customers cannot do without is "compute they need right now," not "CoreWeave and no one else." This is fundamentally different from the truly indispensable great companies (where a customer switch would be excruciating and there is no equivalent substitute)—CoreWeave is "one of the key suppliers," not "an irreplaceable lifeline."

  Now look at "whether the growth model is sustainable and whether it relies on harming society and regulation"—on this layer CoreWeave is actually relatively clean, but there are two real external constraints.

  • No obvious blemish on the ethical/social dimension. What CoreWeave sells is compute infrastructure; it does not make money through regulatory arbitrage, harming consumers, or data abuse. Its growth is not built on "exploiting loopholes in society"—which is better than many growth stocks that expand through regulatory gray zones or externalizing social costs. The securities litigation it faces (alleged false statements in January 2026, which the report notes is still an allegation with no court finding) is a disclosure dispute, in the realm of corporate governance, not "a growth model that harms society."

  • But energy/power is a real social and sustainability constraint. AI data centers are heavy power consumers, and CoreWeave's total contracted power already exceeds 3.5GW. The report explicitly lists "power and environmental policy," "electricity-price changes," and "energy policy" as risks that could affect cost and supply. As the scale of AI data centers' power use draws social attention (grid pressure, electricity prices, carbon emissions), future power access and environmental compliance could become hard constraints on its expansion—this is not "it harming society," but rather "society's tolerance for heavy power consumers like it, and regulation, possibly tightening."

  • Export controls/geopolitics are another regulatory red line. The report discloses that the company explicitly flags that export controls and tariffs could affect cost and supply. Because it depends entirely on Nvidia GPUs, and high-end AI chips are exactly the core target of export controls, regulatory changes could directly choke its supply.

  The Baillie Gifford wrap-up: CoreWeave's "missing-it factor" is moderate-to-strong but time-limited—customers cannot do without it within the contract term, but outside the contract term customers could build their own or switch at any time, because its core large customers happen to be the ones most capable of replacing it. Its growth model does not harm society and does not rely on regulatory arbitrage, which is its clean side; the real variables for regulatory sustainability lie in the two external constraints of "energy/power" and "export controls," not in ethical blemishes. On balance it is "neutral": it is an important supplier of AI compute, but it does not yet qualify as an indispensable player that both society and customers "cannot do without"—and the truly great growth stocks are usually the latter.

  Jun 10, 2026
• QHow is this business's unit economics (gross margin, incremental returns)? Does it get better or worse as scale grows? Where does the money earned go?2/10

  A

  Unit economics is where this business is most easily misled by "high gross margin": the reported GAAP gross margin looks very pretty (65%–74%), but that is caused by accounting conventions moving large amounts of infrastructure depreciation out of the cost line; the real incremental return is constrained by "every dollar of new revenue must first be met with capital expenditure several times operating cash flow." As scale grows, unit economics is currently "the bigger it gets the more it burns" rather than "the bigger it gets the more it earns"; the money earned (and borrowed) is almost entirely spent on buying GPUs and building data centers.

  First, poke through the "pretty gross margin" paper window, which is the "truth of incremental economics" Baillie Gifford cares about most. The GAAP gross margins the report discloses—69.9% in 2023, 74.3% in 2024, 71.7% in 2025, falling to 65.5% in Q1 2026—look on their own like a high-margin software business. But the report points out incisively: this is because "large amounts of depreciation and infrastructure expense related to servers, networking equipment, and internal software are placed in technology and infrastructure rather than the traditional cost of revenue." For a capital-heavy infrastructure business, moving equipment depreciation out of the cost line naturally inflates gross margin—"looking at gross margin alone overstates economic profit." So this "high gross margin" is not at all evidence of strong unit economics.

  What truly reflects unit economics is "how many dollars must be invested to earn each dollar." The data here is startling:

  • The report discloses that in Q1 2026 the company's single-quarter purchases of property and equipment reached $7.695 billion, far above the $2.984 billion of operating cash flow over the same period. That is, single-quarter capex was 2.6 times operating cash flow.
  • On a full-year basis it is even more extreme: 2026 capex guidance of $31–35 billion against full-year revenue guidance of only $12–13 billion—capex is about 2.5 times revenue. This means for every dollar of revenue the company takes in, it must pour in about 2.5 dollars to buy capacity. This is an extremely capital-intensive business, exactly the opposite of the "asset-light, high-incremental-return" that Baillie Gifford loves.
  • The report's calculated "rough free cash flow" is therefore deeply negative year after year: about -$5.9 billion in 2024, about -$7.2 billion in 2025, and about -$4.7 billion in Q1 2026.

  Does unit economics get better or worse as scale grows? Bulls will say "scale effects + a rising share of the software layer → margins will improve," which is the core bet of the optimistic scenario. But the actual data gives the opposite signal:

  • GAAP gross margin fell rather than rose (74.3% → 65.5%), and Adjusted EBITDA margin also compressed from a higher level to 56% in Q1 2026—scale is expanding while unit profitability is retreating.
  • Even using the company's broadest Adjusted EBITDA measure (the report discloses about $3.093 billion in 2025 and about $1.157 billion in Q1 2026), the report cautions that it "adds back large amounts of depreciation and stock-based compensation and cannot be directly equated with cash distributable to ordinary shareholders." And on a GAAP basis, the 2025 operating loss was $46 million while net interest expense was as high as $1.229 billion—a 2025 net loss of $1.167 billion, so interest coverage does not hold at all. The larger the scale, the more debt and interest burdens are amplified in tandem.

  Where does the money earned (and borrowed) go? The answer is extremely clear: almost entirely into capacity expansion. Operating cash flow + huge external financing (Nvidia's $2 billion private placement, the $8.5 billion DDTL 4.0 loan) → all poured into buying GPU servers, building data centers, and connecting power. The company pays no dividends and does no buybacks, plowing every available dollar back into expansion—which is not wrong during an industry expansion phase, but it has yet to prove these investments can produce returns that flow back to ordinary shareholders.

  One accounting detail warrants extra caution: the report notes the company's operating cash flow is "very strong," but "comes mainly from customer prepayments, deferred revenue, and changes in working capital"—the 2025 increase in operating cash flow "came mainly from upfront payments on committed contracts." This means a considerable part of the good-looking operating cash flow is a one-off/unsustainable working-capital tailwind of "money customers paid in advance," rather than the business generating cash on its own. The report therefore conservatively estimates its 2025 "true owner earnings" was most likely "only close to zero, or even slightly negative."

  The Baillie Gifford wrap-up: CoreWeave's unit economics is "pseudo-high-margin, truly capital-heavy": the reported gross margin is dressed up by accounting conventions, and the real incremental return is swallowed by "every dollar of revenue requiring 2.5 dollars of capital first"; as scale grows, what shows up is margin compression, debt amplification, and deeply negative free cash flow, not scale effects. The money earned and borrowed all goes back into expansion, but the inflection point of "the more it expands the more it earns" has yet to appear. On this dimension, within the Baillie Gifford framework, it is a clear weakness: a good growth business should show improving unit economics as it scales, and CoreWeave has so far gone the opposite way—which is exactly why it does not deserve "simple and great," and the fundamental financial reason I am highly wary of its valuation.

  Jun 10, 2026
• QFor it to rise fivefold in ten years, what conditions must hold simultaneously? Are these conditions realistic? What expectations does today's share price imply?3/10

  A

  For CoreWeave to rise fivefold in ten years from the current price (about $98, market cap about $53.7 billion) to a market cap of about $268.5 billion, at least six conditions must hold "simultaneously"—and each one alone is not easy, so stacked together the realistic probability is not high. Worse still: today's share price has already prepaid a fairly large part of the optimistic script of "high growth + major margin improvement + falling financing costs," leaving zero margin of safety.

  First, anchor the starting point. I anchor all answers in this piece uniformly to the June 9, 2026 close of about $98.45, market cap about $53.7 billion, about 545.6 million shares, TTM revenue about $6.23 billion, and a price-to-sales ratio of about 8.6x (the report body uses the May 20 snapshot of $99.81 / about $52.6 billion; the two are consistent in convention, with the difference coming only from the point in time). Fivefold in ten years means market cap reaching about $268.5 billion, an annualized rate of about 17.5%.

  Conditions that must all hold for fivefold in ten years (none can be missing):

  1. Revenue sustains high growth for many years. The report's optimistic scenario assumes revenue CAGR of about 35% over the next 5 years and about 12% for the following 5 years. Given that 2026 guidance is already $12–13 billion, sustaining this slope for ten years means revenue ten years out reaching the scale of high tens of billions or even over a hundred billion dollars—requiring AI compute demand not to recede for ten years.

  2. Margins must undergo a qualitative change. The report's optimistic scenario assumes a year-10 owner earnings margin of 20%. But the reality is the company is still losing money on a GAAP basis today (a 2025 net loss of $1.167 billion and a Q1 2026 net loss of $740 million), with gross margin still compressing (74.3%→65.5%). Jumping from "true owner earnings near zero or even negative" to "a 20% owner earnings margin" is a perilous leap from losses to high-quality profitability.

  3. The capital-heavy model must switch to "the more it grows the more it earns." Today every dollar of revenue must first be met with about 2.5 times capex (2026 capex guidance of $31–35 billion vs. revenue of $12–13 billion). Fivefold in ten years requires capital intensity to fall significantly and incremental returns to turn positive.

  4. Leverage and financing costs are contained. The company's net debt has reached about $22.6–24.9 billion, 2025 net interest expense was $1.229 billion, and interest coverage (GAAP) does not hold. Fivefold in ten years requires it to keep expanding through massive borrowing while keeping the interest burden and dilution under control—the $8.5 billion DDTL 4.0 receiving an investment-grade rating is a good sign, but the financing environment must not reverse.

  5. Customer concentration falls sharply without attrition. In 2025 Microsoft accounted for 67%, and in Q1 2026 Customer A still accounted for 45%. Fivefold in ten years requires customer diversification (Meta's new $21B, OpenAI, Anthropic point the direction), while these large customers with in-house build capability do not pull their workloads back.

  6. The moat solidifies from "first-mover execution" into a "structural barrier," and is not crushed by a hyperscaler price war. This is the hardest one—it must hold its share and pricing power under siege from competitors like Amazon/Google/Microsoft/Oracle that have "more capital, lower costs, and more complete ecosystems."

  Multiplying these six conditions together, each single probability is not high, so the joint probability of "all holding simultaneously" is even lower. This is exactly why the report summarizes the most fragile valuation assumption as "high growth over the next decade + clear margin improvement + falling financing costs can all hold simultaneously"—it is not impossible, but it requires too many things to go right at once.

  What expectations does today's share price imply? This is the soul of Baillie Gifford's question. The answer is: the current price has already prepaid most of the optimistic script, leaving almost no room for "error." The report's three-scenario intrinsic-value estimates make it very clear:

  • Conservative scenario (5-year revenue CAGR 20%, year-10 margin 8%, discount rate 12%): $0–25 per share;
  • Neutral scenario (CAGR 28%, margin 15%, discount rate 11%): $35–65 per share;
  • Optimistic scenario (CAGR 35%, margin 20%, discount rate 10%): $95–135 per share.

  The current price of about $98 has already fallen at the lower edge of the optimistic-scenario range—that is, what the market currently prices in is essentially the assumption that "the optimistic script will be delivered." Against the neutral and conservative scenarios, the current price has no margin of safety at all. The report therefore gives an "ideal buy range of $25–45, an acceptable holding range of $45–70, and above $90 falls into clearly overvalued territory"—the current price is above the overvaluation line the report defines. Relative valuation confirms this: CRWV's price-to-sales ratio is about 8.6x, comparable to Oracle's about 9–10x price-to-sales with FY2025 revenue of $57.4 billion and genuine profitability (Oracle's market cap is about $592 billion)—a company still losing money with unverified unit economics enjoys a sales multiple on par with a mature, profitable giant, which itself shows "cheap" is out of the question.

  The Baillie Gifford wrap-up: fivefold in ten years requires six high-difficulty conditions to hold simultaneously, with a low joint probability; and the current share price is already priced for the optimistic script, sitting at the lower edge of the optimistic scenario and above the overvaluation line the report defines. This means that even if the company executes quite well (delivering the neutral scenario), buying at the current price may not make money or may even lose money—because the good outcome has already been prepaid. Baillie Gifford looks for "great growth the market has not yet fully priced in"; CoreWeave's problem is not insufficient growth, but growth already overpriced. This dimension (including the expectations implied by valuation) is its clearest deduction.

  Jun 10, 2026
• QWhy hasn't the market realized all this yet? Is it that it can't understand, looks down on it, or can't see far enough? What will become the "narrative inflection point"?3/10

  A

  The premise of this question needs correcting first: the market has not "failed to realize" CoreWeave's growth—on the contrary, the market sees it very clearly and may even have priced it too full. So the question is not "why can't the market understand this flower," but "the market has already priced the flower as certain to bloom in the future, leaving investors an extremely small knowledge gap." The real "narrative inflection point" is not when the market discovers its merits, but whether it can deliver the optimistic expectations already prepaid—if confirmed, the high valuation holds; if disproven, the valuation contracts sharply.

  First, aim this question in the right direction. Baillie Gifford's "can't understand / looks down on / can't see far" framework applies to "great companies undervalued and hidden in a corner by the market." But CoreWeave's market status is exactly the opposite—it is one of the most closely watched AI names since its March 2025 IPO, with dense analyst coverage, a consensus rating of "Buy" and an average price target of about $132.68. This is not a neglected stock but a hotly pursued one. So diagnosing it with Baillie Gifford's three-way split:

  • "Can't understand"? Partly true, but in the opposite direction. The market's "not understanding" is not about underestimating growth, but about not necessarily fully grasping how much its growth quality is discounted. Several "easily misread" points the report repeatedly stresses—the high GAAP gross margin is actually inflated by accounting moving infrastructure depreciation out of the cost line, operating cash flow is "very strong" but a large part comes from customer prepayments and working-capital tailwinds, and Adjusted EBITDA adds back large amounts of depreciation and stock-based compensation—are exactly the details the optimistic narrative easily overlooks. The market understands "revenue doubling," but does not necessarily see through "true owner earnings near zero." This is a knowledge gap, but it points to "the market possibly overvaluing" rather than "undervaluing."

  • "Looks down on it"? Entirely untrue. Far from looking down on it, the market has given it a valuation on par with a mature, profitable giant—CRWV's price-to-sales ratio is about 8.6x, comparable to Oracle's about 9–10x price-to-sales, a genuinely profitable company with FY2025 revenue of $57.4 billion.

  • "Can't see far"? Also untrue, even the opposite—the market may "see too far." The current price of about $98 already sits at the lower edge of the report's optimistic scenario (5-year revenue CAGR 35%, year-10 owner earnings margin 20%), meaning the market has already discounted the far-off optimistic outcome of "a margin transformation ten years out" into today's price. This is not failing to see far, but treating the best far-off outcome as the base case.

  So where is the narrative inflection point? Since the market is already priced on the optimistic narrative, the inflection point is not "good news being discovered," but the key moment when "the optimistic narrative is confirmed or disproven." The report has actually listed these inflection signals quite clearly, which I group into two sets, positive and negative:

  Upward narrative inflection points (confirming the optimism, possibly maintaining/pushing up the valuation):

  • owner earnings / free cash flow clearly turning positive—the strongest catalyst for rewriting the "capital-heavy cash-burn story" into a "cash-cow story";
  • customer concentration falling significantly (the 67% reliance on Microsoft diluted by diversification into Meta's $21B, OpenAI, Anthropic, and others, without losing old customers);
  • the 2026 annual report showing the material internal-control weakness remediated and the securities litigation cleared—removing the governance-risk discount;
  • the software/platform layer (W&B and others) beginning to contribute quantifiable high-margin revenue, proving the "second curve" exists.

  Downward narrative inflection points (disproving the optimism, possibly triggering sharp valuation contraction):

  • RPO/backlog growth slowing while the company still needs large-scale borrowing or deeply discounted issuance;
  • AI demand shifting from the training boom to lower-capital-intensity inference or in-house customer capability, driving down utilization of high-priced capacity;
  • large-customer orders delayed/canceled, or financing costs rising significantly;
  • internal-control weaknesses long unfixed, or the 2026 audit running into problems.

  The Baillie Gifford wrap-up (which must also give an honest risk gauge): CoreWeave is not an undervalued stock the "market has not realized," but a hot stock the "market has fully or even excessively realized"—the knowledge gap is extremely small, and the market is more likely to be leaning optimistic than pessimistic. So its "narrative inflection point" is two-way, and the downside risk cannot be ignored: the report estimates that in the worst case ordinary shareholders face a 60%–90% risk of permanent loss (stemming from its combination of thin equity, high debt, and high capital intensity). For a Baillie Gifford framework of "finding great growth the market has not yet discovered," CoreWeave's answer is the opposite: its growth has long been discovered and generously priced, and what is left for new investors is not the opportunity to "understand what others haven't," but the risk of "betting that an already-priced optimistic script will not go wrong." This is exactly why the report ultimately gives "Watch" rather than "Buy," and compresses the ideal buy range to $25–45.

  Jun 10, 2026