Cerebras Drops to $225.85 as $180.1M Revenue Miss Overrides a 103% Core Revenue Gain

Cerebras Systems opened Thursday's session under heavy distribution, trading down 13.82% to $225.85 after an initial dip to $232.98 failed to hold. The stock closed Wednesday's regular session at $262.06, up 11.63% or $27.30 on the day, having traded an intraday range of $240.88 to $265.63 with an opening print of $248.82. Market capitalization at that close stood at $59.37 billion against average daily volume of 5.30 million shares.

The reversal began immediately after the 4:09 p.m. Eastern release. Extended trading took the stock to $238.96, a 9.26% decline, then accelerated to $221.01 and $220.08 — losses of 15.66% and 16.02% respectively. Pre-market quotes ran between 15% and 18% lower. The regular session opened into that gap and has not recovered it.

The round trip is severe. Wednesday's 11.63% advance came in sympathy with the broader AI infrastructure melt-up that lifted CoreWeave 20.2%, Nebius 17.1%, Super Micro Computer roughly 19% and Lumentum 14%. Cerebras participated fully, closing at its highest level in weeks and 42% above its May IPO price of $185. At $225.85 that premium has compressed to roughly 22%.

The wider context is a stock that has been broken since May. Cerebras peaked near $350 shortly after listing, then declined roughly 33% into the August 12 print. The recovery from that low carried it back above the 20-, 50-, 40- and 80-period moving averages and through the 0.382 Fibonacci retracement at $247.24 before stalling at the 200-day exponential average near $267.74, which converged with daily R1 resistance at $267.45. Daily relative strength read 63.32 with the one-hour reading at 82.46 — deeply overbought on the intraday timeframe — the session before earnings.

That configuration matters. The stock entered the print extended into a defined resistance shelf, having rallied roughly 12% in a single session on sector sympathy rather than company news. A rally of that character into a binary event is the most fragile setup available, and the market took it apart in one after-hours session.

Cerebras completed the largest semiconductor initial public offering in history on May 14, pricing at $185 and raising $6.4 billion in gross proceeds.

The Headline Miss Was Real: $180.1 Million Against $194 Million

GAAP revenue for the quarter ended June 30 came in at $180.11 million against a consensus of $193.6 million to $194.2 million — a shortfall of roughly 7%. That number grew 74.3% year over year, which is extraordinary in isolation and insufficient against the bar the market had set.

The bar was the problem. Company guidance issued alongside the first-quarter release had pointed to second-quarter core revenue of approximately $194 million, representing 88% year-over-year growth. First-quarter GAAP revenue had been $193.4 million with core revenue of $191.3 million, up 92%. Sequential GAAP revenue therefore declined from $193.4 million to $180.1 million — the first quarter-over-quarter contraction since the company began reporting publicly.

For the six months ended June 30, revenue reached $373.5 million against $202.8 million in the comparable prior-year period. That is 84% growth across the half, which is the cleaner read on trajectory than either individual quarter.

The bottom-line figures require careful separation because the reporting produced two entirely different numbers depending on the measure. GAAP loss per share was $2.98 against a consensus estimate of $0.18, a miss on the order of 1,555%. Adjusted loss per share was $0.05 against $0.17 expected — a beat. GAAP net loss for the quarter totaled $450.5 million.

Most of the coverage that hit the tape within minutes of the release compared the $2.98 GAAP figure to the $0.17 to $0.18 adjusted estimate, which is not a valid comparison and drove much of the initial algorithmic selling. The stock fell 9.26% within the first minutes of extended trading and then extended to 16% as headlines multiplied.

That said, the revenue miss stands on its own. A 7% shortfall against a number management itself had guided to nine weeks earlier is a guidance miss, not an estimate miss, and for a company trading at more than fifty times forward sales, execution against its own targets is the entire investment case. The market has now seen hardware demand described as lumpy in the same quarter that the company failed to hit a figure it set.

Core Revenue Doubled to $209.9 Million and Nobody Cared

The number management wants investors to focus on is core revenue, which excludes data center pass-through revenues, customer warrant amortization and certain other items. On that basis the quarter was a record.

Core total revenue reached $209.9 million, up 103% year over year and above the roughly $194 million the company had guided. Core cloud and other services revenue jumped 287% year over year to $127.7 million. On a GAAP basis, cloud and other services revenue reached $126.0 million, up 281%.

The composition shift is the most important operational fact in the release. A company that began life selling wafer-scale hardware systems now derives roughly 60% of core revenue from cloud services, and that segment nearly quadrupled year over year. Management attributed the majority of core revenue growth in the quarter to the cloud business, driven primarily by expanded OpenAI deployments and increased usage from other cloud customers.

That transition is strategically correct. Recurring inference capacity revenue carries better visibility, better margins at scale, and a valuation multiple that hardware sales never command. Chief executive Andrew Feldman framed the quarter around it directly, describing core revenue more than doubling to $210 million and the cloud business nearly quadrupling, and arguing that speed changes what AI can do — making it more useful and productive while opening new markets.

The gap between the two revenue measures is where the market's skepticism lives. GAAP revenue of $180.1 million against core revenue of $209.9 million is a $29.8 million adjustment, or roughly 17% of the reported figure. When a company's preferred metric exceeds its reported metric by that margin, and the preferred metric beats while the reported metric misses, the market discounts the preferred metric. That is precisely what happened.

Guidance compounds the issue because it is delivered in core terms. Third-quarter core revenue is guided to $214 million to $216 million, above consensus. Full-year core revenue was raised to $880 million to $890 million from a prior range of $855 million to $865 million. Both figures are unmatched by any GAAP equivalent, which means the market must model a moving reconciliation to arrive at reported results.

The $2.98 Loss Was $377 Million of Stock Compensation

The headline loss is almost entirely a post-IPO accounting event rather than an operating deterioration, and the distinction determines whether Thursday's decline is justified.

Stock-based compensation totaled $377.0 million in the second quarter and $386.6 million year to date. That means roughly 97.5% of the year's entire stock compensation expense landed in a single quarter. A large portion relates to restricted stock units and performance restricted stock units whose liquidity and performance conditions were satisfied in connection with the initial public offering, plus executive grants issued at listing.

Against a GAAP net loss of $450.5 million, the $377 million charge accounts for 84% of the total. Strip it out and the underlying operating loss falls to roughly $73 million on core revenue of $209.9 million — a loss margin of about 35% on a business growing above 100%.

That is a normal profile for an infrastructure company scaling ahead of demand, and it is the reason adjusted loss per share came in at $0.05 against a $0.17 estimate. The company beat on the measure that reflects cash economics and missed catastrophically on the measure that reflects the one-time cost of going public.

Two caveats keep this from being a clean dismissal. First, stock compensation is a real cost to shareholders even when non-cash, because it dilutes ownership. A $377 million charge against a $59 billion market capitalization is 0.64% of the company transferred to employees in one quarter, and the run-rate beyond the IPO catch-up remains unknown. Second, the market has demonstrated repeatedly this reporting season that it will not do the work of separating one-time items in real time. Cisco delivered a record quarter and lost 7% on a single margin guidance line the same evening.

The practical read for holders is that the GAAP loss will normalize sharply in the third quarter as the IPO-triggered vesting washes through. Year-to-date stock compensation of $386.6 million against a second-quarter figure of $377.0 million implies a first-quarter run rate near $10 million. Even allowing for substantially higher ongoing grants as headcount scales, the third quarter should show a fraction of this charge — which makes the next print the cleanest look at underlying economics the market will have had.

Margins Beat the Guide by a Mile: 40.6% Core Gross, Minus 16% Core Operating

The most underappreciated element of the release is that margins came in dramatically better than the company had told investors to expect.

Guidance issued in June called for second-quarter core gross margin of 36% to 38%, down sharply from 46.5% in the first quarter, with core operating margin between negative 30% and negative 32%. The stated reason for the compression was that meeting surging inference demand before self-built data centers came online required temporarily leasing capacity and relying on third-party facilities at unfavorable economics.

Core gross margin came in at 40.6%, roughly 41%, improving approximately 940 basis points year over year and landing 260 basis points above the top of the guided range. Core operating margin registered negative 16%, an improvement of approximately 2,600 basis points year over year and 1,400 to 1,600 basis points better than guided.

Management framed the operating margin improvement as evidence of structural leverage — delivering that gain while more than doubling revenue and simultaneously stepping up investment across people, manufacturing, data center capacity and company infrastructure.

The full-year guidance raise covered revenue, gross margin and operating margin, not just the top line. That is a three-line raise from a company that had guided conservatively nine weeks earlier and then beat its own margin targets by a wide margin.

Under any normal reading, a business that guides margins down 900 basis points sequentially, delivers 260 basis points better than the top of the range, improves operating margin by 2,600 basis points year over year, and raises full-year guidance across every metric is executing well. The market's response was to sell it 14% to 16%.

The disconnect is explicable only through the revenue miss and the hardware line. Investors had spent the pre-earnings window focused on five questions: whether core revenue would clear $194 million, whether margins would hold the 36% to 38% range, how full-year and 2027 outlooks would change, how the OpenAI ramp and AWS scaling would progress, and whether customer concentration would improve. Four of the five were answered favorably. The fifth — the revenue and concentration picture — was not, and that single failure overrode the rest.

Hardware at 1.8% GAAP Gross Margin Is the Number That Broke the Stock

Buried in the supplementary disclosure is the figure that explains Thursday's persistence rather than the initial gap.

GAAP hardware gross margin in the second quarter was 1.8%. Core hardware gross margin was 38.8%. The 37-point gap between the two reflects adjustments for customer warrant amortization, stock-based compensation and data center expenses — a reconciliation wide enough that reasonable analysts can disagree about which figure represents economic reality.

Hardware revenue declined in the quarter, which management characterized as evidence of lumpy demand. That is a defensible description of a business selling multi-million-dollar wafer-scale systems in discrete units, where a single deployment slipping a quarter moves the reported number by tens of millions. It is also exactly what a market pricing 100% growth does not want to hear.

The sequential math tells the story. Core revenue of $209.9 million with core cloud at $127.7 million implies core hardware and other revenue near $82 million. Against first-quarter core revenue of $191.3 million with a substantially smaller cloud contribution, hardware volumes clearly contracted. GAAP revenue declining from $193.4 million to $180.1 million sequentially while cloud grew 281% year over year confirms that hardware absorbed the entire decline and more.

This is the pivot point for the entire investment thesis. Cerebras is transitioning from a hardware vendor with lumpy multi-million-dollar deal timing to a cloud infrastructure provider with recurring inference capacity revenue. That transition is strategically superior in every respect except one: during the transition, reported revenue becomes unpredictable, because a declining lumpy business is being replaced by a scaling recurring business and the crossover produces exactly the kind of sequential contraction this quarter delivered.

Feldman positioned the company at the close of the call as one of only four organizations — alongside Google, Amazon and Nvidia — capable of building processors, systems and data centers while delivering AI cloud services. The claim is defensible on capability. It also implicitly concedes that the hardware-only identity is over, and that the market must now value Cerebras against infrastructure operators rather than chip designers.

The $25.4 Billion Backlog and the 80% Problem

Remaining performance obligations stood at $25.4 billion as of June 30. Management described the figure as a sign of extraordinary future demand providing visibility sufficient to invest ahead of revenue.

The recognition schedule is disclosed and worth modeling precisely. Approximately 22% is expected to be recognized over the next 24 months, 43% over the following 24 months, and the balance thereafter. That translates to roughly $5.6 billion across 2026 and 2027 combined, $10.9 billion across 2028 and 2029, and $8.9 billion beyond. Against full-year 2026 core revenue guidance of $880 million to $890 million, the backlog implies a revenue base an order of magnitude larger by the end of the decade.

The concentration is the problem. The backlog derives largely from the OpenAI master relationship agreement, and independent analysis at the time of the IPO put OpenAI's share near 80% of what was then a $24.6 billion figure. The underlying contract is the multi-year deal announced in the first quarter covering 750 megawatts of inference compute valued at more than $20 billion.

Feldman addressed concentration directly when questioned on the call, acknowledging that OpenAI will remain an important customer next year while expecting its revenue share to decline gradually as AWS, coding, security and other businesses scale.

Gradually is the operative word. A supplier deriving roughly 80% of contracted future revenue from a single counterparty carries counterparty risk that no growth rate compensates for. If OpenAI's capacity requirements shift, its model architecture changes in ways that reduce inference intensity, or its own capital position tightens, the entire backlog reprices.

The relationship is also financially entangled beyond the supply contract. Cerebras carries a $918.2 million outstanding Working Capital Loan from OpenAI. A customer that is simultaneously the source of 80% of backlog and a nine-figure creditor is not an arm's-length relationship, and the disclosure invites questions about how much of the reported growth reflects genuine third-party demand versus a vertically financed arrangement between two parties with aligned incentives to show scale.

Guidance Raised Across Every Line: $880–890 Million and a Tripling in 2027

Forward guidance was the strongest element of the release and the reason the decline may prove temporary.

Third-quarter core revenue is guided to $214 million to $216 million, above consensus and representing sequential growth of roughly 3% at the midpoint against second-quarter core revenue of $209.9 million. Full-year 2026 core revenue was raised to $880 million to $890 million from a prior $855 million to $865 million — a $25 million increase at both ends of the range delivered in the same release as a revenue miss.

The 2027 statement is the one that matters for valuation. Management expects revenue to triple in the next fiscal year. Applied to the midpoint of 2026 core guidance at $885 million, that implies roughly $2.65 billion in 2027. Feldman framed the operating leverage case around scale: more efficient manufacturing, better component pricing, and amortization of the manufacturing organization across more units.

Capacity commitments underpin the projection. Data center capacity live and under contract exceeds 600 megawatts for delivery by the end of 2027. Manufacturing capacity is set to grow more than tenfold during 2026 through new lines at several contract manufacturers. Operating leases require $690.3 million in future payments with a 5.5-year weighted-average term, which is the balance-sheet footprint of that expansion.

The arithmetic connecting guidance to backlog is internally consistent. Roughly 22% of $25.4 billion recognized over 24 months is $5.6 billion, and a 2026 base of $885 million growing to $2.65 billion in 2027 sums to $3.5 billion — leaving room for the remainder to land in the first half of 2028 within the disclosed window.

The gap between guidance credibility and guidance content is the whole trade. This is the same management team that guided second-quarter core revenue to approximately $194 million and delivered $209.9 million on that measure while missing GAAP consensus by 7%. It guided gross margin to 36% to 38% and delivered 40.6%. It guided operating margin to negative 30% to negative 32% and delivered negative 16%.

On its own stated metrics, the company beat everything it guided and raised everything forward. On the metrics the market uses, it missed. Both statements are true.

AWS, AMD and the Hyperscaler Optionality Not in the Backlog

The most valuable disclosure on the call received the least attention: the $25.4 billion backlog contains no contribution from AWS or any other hyperscaler.

The Amazon partnership was launched alongside the first-quarter results and is now scheduled for general availability through the Amazon Bedrock platform in the first quarter of 2027. Management expects first revenues from hyperscaler channels starting in mid-2027 with a ramp through 2028 and beyond, and characterized discussions with additional hyperscalers as going well.

That timeline puts meaningful hyperscaler revenue eighteen months out, which is too distant to support the current multiple but close enough to matter for a two-year view. Distribution through Bedrock provides reach into essentially every enterprise AWS serves without Cerebras building a direct sales organization at that scale.

The AMD relationship, announced July 23 at the Advancing AI event, covers disaggregated inference with products entering production later this year. Partnering with the primary alternative to Nvidia in accelerated computing positions Cerebras as a complement to the challenger rather than a competitor to the incumbent — a defensible strategic posture given that Nvidia's dominance in training makes direct confrontation unwinnable.

OpenAI's usage has deepened on the technical side as well, with Cerebras silicon serving GPT-5.6 Sol. Management's argument is that fast tokens command a market premium and that fast tokens combined with frontier intelligence are available only through the OpenAI-Cerebras pairing. CrowdStrike appears among named partners, extending the footprint into security workloads.

The business outside OpenAI and the hyperscalers continues to grow, per management commentary, though the company has not quantified it separately — which is itself informative. A company able to demonstrate rapid non-OpenAI growth would disclose the figure.

The optionality is genuine and unpriced within the backlog. It is also entirely prospective. Between now and mid-2027 the reported numbers depend on one customer, and the market has just demonstrated how it responds to a single soft quarter within that dependency.

The Lockup: 60 Million Shares Released Into This Print

Cerebras structured an unusually aggressive lockup release, and the timing collides directly with the current decline.

More than 60 million shares were scheduled for release from lockup restrictions in connection with the second-quarter earnings publication — a structure flagged at the time of the IPO as unusually fast for a newly listed company. Against roughly 226 million shares outstanding implied by the $59.37 billion market capitalization at Wednesday's close, that represents more than a quarter of the float becoming eligible for sale within days of a 14% earnings decline.

Insider activity ahead of the release supports the concern. Filings show $20.7 million in shares sold over the prior three months with no insider purchases. Coverage in early July flagged an executive shake-up in insider transactions, and additional reporting has noted senior executives cashing in through significant stock movements.

Insider selling at a company three months post-IPO where much of the compensation vested at listing is expected and not by itself a signal. The $377 million stock compensation charge in the quarter is precisely the mechanism through which those shares were created. Employees who received RSUs whose liquidity conditions were satisfied at IPO now hold taxable positions and will sell some portion to cover.

The problem is sequencing. A newly public stock trading 35% below its May peak, having just missed revenue consensus, with a broad lockup release arriving in the same window, faces mechanical supply into a market with no established institutional holder base to absorb it. That is the structural explanation for why Thursday's decline accelerated from $232.98 to $225.85 rather than stabilizing at the initial gap.

Average daily volume of 5.30 million shares against 60 million newly eligible shares means even modest participation from released holders represents multiple days of normal turnover. The supply overhang does not require panic selling to matter — it requires only that a fraction of holders who acquired stock below $185 choose to realize gains after watching the price fall by a third from its high.

Balance Sheet: $8.6 Billion Liquidity Against a $918 Million OpenAI Loan

Whatever the operating uncertainties, Cerebras is not capital constrained.

As of June 30 the company held $7.43 billion in cash, cash equivalents and restricted cash plus $1.18 billion in investments, for total liquidity above $8.6 billion. An $850 million revolving credit facility remains entirely undrawn. That balance derives principally from the $6.4 billion raised in the May IPO.

Against a quarterly cash operating loss near $73 million excluding stock compensation, the runway is effectively unlimited on an operating basis. The capital consumption comes from expansion: more than 600 megawatts of contracted data center capacity for delivery by end-2027, tenfold manufacturing capacity growth during 2026, and $690.3 million of committed operating lease payments over a 5.5-year weighted-average term.

At $59.37 billion market capitalization, cash and investments of $8.6 billion represent roughly 14.5% of the enterprise. At $225.85 and a market capitalization near $51.2 billion, that rises to nearly 17%. Backing out net cash after the $918.2 million OpenAI loan gives an enterprise value near $43.5 billion.

The OpenAI Working Capital Loan is the balance-sheet item requiring the most scrutiny. A $918.2 million principal balance owed to the customer generating roughly 80% of contracted backlog creates a circularity that conventional supplier-customer analysis does not handle well. Capital flows from OpenAI to Cerebras as debt, funds capacity that serves OpenAI, and returns as revenue. The arrangement is disclosed and legitimate, but it means the reported growth and the financing of that growth originate from the same counterparty.

Liquidity of $8.6 billion against that loan and against the capacity commitments means the company can fund its 2027 buildout without returning to markets. That removes the equity-issuance overhang that typically accompanies a high-growth infrastructure story and is the strongest argument against the current decline extending into a structural derating.

Valuation: 58x Forward Core Revenue in a Market Repricing AI Margins

At $225.85 and roughly 226 million shares, market capitalization sits near $51.2 billion. Against full-year 2026 core revenue guidance of $885 million at the midpoint, that is approximately 58 times forward sales. Against the implied 2027 figure of roughly $2.65 billion under a tripling, it is approximately 19 times.

At Wednesday's $262.06 close and $59.37 billion, the same multiples were 67 times and 22 times. At the May peak near $350 they were roughly 89 times and 30 times. Independent analysis at the time of the IPO put the stock at about 87 times 2026 sales, describing it as extremely expensive. Screening data has flagged price-to-sales at 291.5 times on a trailing GAAP basis against a historical median near 238.6 times, a figure that reflects the small trailing revenue base rather than any meaningful valuation signal.

Fifty-eight times forward sales is not defensible on its own. It requires the 2027 tripling to occur, the backlog to convert on schedule, and the gross margin trajectory to continue improving from 40.6% toward the 46.5% posted in the first quarter and beyond. Every one of those requires OpenAI to perform.

The timing is unhelpful. The market spent this week systematically repricing AI infrastructure on margin structure rather than revenue growth. Cisco delivered $17.3 billion in revenue, 18% growth, $4 billion in quarterly hyperscaler AI orders and guidance $1.2 billion above consensus, then fell 7% because first-quarter gross margin guidance came in 100 basis points below expectations. Coherent beat and fell 6%. Cerebras missed revenue by 7% while beating margins by 260 basis points and fell 14%.

The pattern across three consecutive evenings is consistent: revenue growth is treated as given and fully priced, while any imperfection in the margin or execution picture is repriced immediately and violently. That is a market where good news carries no information and only disappointments move price.

Analyst positioning has not adjusted. Eleven analysts maintain a Strong Buy consensus with ten buy ratings and one hold. The average twelve-month target sits between $292 and $297.56, with a high of $340 and a low ranging from $209 to $273 depending on source. One firm raised its target to $310 from $300 in late July. At $225.85, the average target implies roughly 30% upside — but those targets predate the print and will be revised.

Technical Structure: $267 Ceiling, $247 Fib, $209 Analyst Floor

The chart offers unusually clean levels because the stock has only three months of trading history and each one is anchored to an identifiable event.

Overhead resistance is concentrated at $267.45 to $267.74, where the 200-day exponential average converges with daily R1 pivot resistance. That zone rejected the pre-earnings rally at $265.63 on Wednesday and stands as the most important level on the chart. Above it, the next reference is the $273 low analyst target, then $292 to $297 at the consensus average, and $340 at the high.

The 0.382 Fibonacci retracement at $247.24 was reclaimed during the pre-earnings advance and has now been lost on the downside. That level becomes first resistance on any recovery attempt. Wednesday's opening print of $248.82 reinforces the zone.

Immediate support sits at Wednesday's intraday low of $240.88, which has already broken. Below that, the after-hours lows at $220.08 and $221.01 form the first defended shelf, and the session low near $225.85 sits between them. A break of $220 opens the $209 level marked by the lowest published analyst target, then the pre-rally base near $199 and the $185 IPO price.

The $185 line is the structural floor for the entire equity story. It represents the price at which $6.4 billion of institutional capital was allocated in May, and it is the level below which every IPO participant is underwater. Given the lockup release, that level carries more weight than any technical retracement.

Momentum has reset hard. Daily relative strength read 63.32 into the print with the hourly at 82.46 — deeply overbought — which is why the initial decline found no support until the mid-$220s. Those readings will now sit in oversold territory, but oversold conditions in a stock with three months of history and 60 million shares emerging from lockup provide limited mean-reversion signal.

The practical decision levels: reclaiming $247.24 neutralizes the damage, and a close above $267.74 restores the pre-earnings structure. Losing $220 targets $209 and then $199, with $185 as the line that would confirm the IPO trade has failed.

Verdict: The Business Beat Its Own Guidance and the Stock Broke Anyway

Cerebras at $225.85 has given back 13.8% from a $262.06 close and now trades roughly 35% below its May peak near $350, retaining a 22% premium to the $185 IPO price. The decline came in a quarter where the company beat its own core revenue guidance, beat its own gross margin guidance by 260 basis points, beat its own operating margin guidance by roughly 1,500 basis points, and raised full-year guidance across revenue, gross margin and operating margin.

The three things that broke it are specific and none of them are margins.

The first is the GAAP revenue miss. $180.11 million against $194 million is a 7% shortfall and a sequential decline from $193.4 million in the first quarter. For a company at fifty-eight times forward sales, missing a number it set nine weeks earlier is a credibility event regardless of what the core metric shows. The 17% gap between GAAP revenue at $180.1 million and core revenue at $209.9 million means the market must accept management's reconciliation to see the beat, and this week the market has been unwilling to accept any adjustment from anyone.

The second is hardware. GAAP hardware gross margin at 1.8% against core hardware gross margin at 38.8% is a 37-point reconciliation gap on a segment where revenue declined and management used the word lumpy. The cloud transition is strategically right — core cloud revenue up 287% to $127.7 million proves it is working — but the crossover produces exactly the sequential unpredictability this quarter delivered, and it will produce more.

The third is supply. More than 60 million shares released from lockup in the same window as a 14% earnings decline, against average daily volume of 5.30 million and $20.7 million of insider selling over three months with no purchases, is a mechanical overhang that will take weeks to clear.

Against those, the structural case is intact and unusually well-documented. Remaining performance obligations of $25.4 billion with a disclosed recognition schedule. Total liquidity above $8.6 billion with an undrawn $850 million revolver. Over 600 megawatts of contracted data center capacity for delivery by end-2027 and tenfold manufacturing expansion during 2026. Guidance for revenue to triple in 2027. AWS Bedrock general availability in the first quarter of 2027 with no hyperscaler contribution in the current backlog. An AMD inference partnership entering production this year.

The unresolvable risk is concentration. Roughly 80% of a $25.4 billion backlog derives from one customer that is also a $918.2 million creditor. Management expects that share to decline gradually. Gradually does not protect a fifty-eight-times multiple.

Base case is stabilization between $220 and $247 through the lockup absorption, with the third-quarter print — the first clean look at economics without $377 million of IPO-triggered stock compensation — as the catalyst. Bull case requires reclaiming $267.74 and targets the $292 to $297 consensus. Bear case breaks $220, targets $209 and then $199, with $185 as the line that invalidates the IPO thesis entirely.

That's TradingNEWS