Leki Research Notes 🐵 Subscriber sign-in

$BRUN — BOOST RUN — DEEP DIVE

Written Sep 15, 2026

Contents

Date: September 15, 2026 Price at time of writing: $14.57 (Nasdaq, Sept 14 close, −12.4% on the day in a sector-wide sell-off). 52-week range $10.30 – $42.00. The stock is −65% from the June 3 intraday high of $42.00, −36% from the $22.93 close on the day it reported Q2 (Aug 14), and back to where the SPAC traded on April 22 ($14.34) before the Dell agreement, the merger close, the Thinking Machines contract, the warrant redemption and the first public print. Framework: Framework_Memo.md (six dimensions, EV/spot asymmetry, Refinements 1-8 applied) Thesis check up front: On-thesis, demand side. Boost Run is the compute layer itself — the GPU cloud that buys the servers, optics, switches, cooling and power that the rest of the AI data center supply chain sells. It is the most direct exposure to GPU rental pricing and Blackwell deployment available at small-cap size, and it carries the sector's beta in both directions (it fell 12.4% on a day CoreWeave fell 7.6%).

THE SHORT ANSWER

What it is. A three-year-old Chicago-area GPU cloud (a "neocloud") that rents bare-metal NVIDIA GPU servers on multi-year take-or-pay contracts with a cash prepayment on every deal, houses them in leased colocation space rather than owned data centers, and finances the hardware with 30-36 month equipment leases whose amortization is matched to the customer contract. It went public on May 8, 2026 through the Willow Lane SPAC with zero redemptions and no PIPE. Six sites are live (Seattle, Eagan MN, Fort Worth, Richardson TX, Charlotte, Raleigh), three more are due within six months (Rock Island IL, Marietta GA, Pryor OK), and it has access to ~253 MW of colocation power. It is one of eight NVIDIA Exemplar Clouds and, at certification, one of two validated on B300. Named customers include Thinking Machines Lab (a $471.7M, 36-month, 5,000-GPU B300 contract), RunPod, Shadeform, Manifold Labs, Nebius, Baseten, DRW, fal, Fireworks AI, Fluidstack, Prime Intellect and Higgsfield.

The numbers. Q2 2026 revenue $31.1M (+270% YoY, +184% QoQ from $10.96M). ARR $145M at June 30, up from $30M at December 31, 2025, with management reaffirming an exit-2026 ARR of ~$400M. Total contract value $1.9B (over $1B signed in Q2 alone), average duration ~3 years, average prepayment 22% of TCV, $128.4M of customer deposits already collected. FY2025 revenue was $26.9M; Q2 2026 alone exceeded it. At $14.57 the company has an 83.85M share count, a $1.22B market cap and a ~$1.33B enterprise value pro forma for the warrant cash (~$1.45B if you capitalize the operating leases). That is 3.3x the guided exit-2026 ARR, 2.7x the $496M FY2027 consensus revenue, 0.7x contracted TCV, and $5.3M of EV per accessible megawatt — the last figure sits in the bitcoin-miner band of the EV/MW table while the fleet is almost entirely B300.

Why it is at $14.57. Five reasons, none of which is "the business broke." (1) The whole small-cap neocloud complex has been derated: WhiteFiber is −61% from its June price, CoreWeave −55% from its high, Nebius −28%, and September 14 was another sector-wide leg down on AI-safety and capex-overspend headlines. (2) Supply. The share count went from 61.4M at closing to 83.85M by August 25 as 10.97M earnout shares were issued (all three price tranches cleared inside one quarter) and 11.45M warrants were exercised for ~$130M of cash; the S-1 registering 58.7M shares for resale went effective July 10, and the lock-up on legacy holders released early once the stock spent 20 of 30 days above $12 — so there is no November cliff to wait for, the supply is already legal. (3) The first public print missed sell-side consensus on revenue ($31.1M vs ~$34M) and post-lease adjusted EBITDA (~$12M vs ~$22M) even as it beat on ARR trajectory and reaffirmed guidance. (4) The 10-Q discloses three unremediated material weaknesses in internal control, both quarterly filings this year were late (NT 10-Q), and the Q1 10-Q was amended for its controls disclosures. (5) Short interest is 15% of a 28.7M-share float with 1.9 days to cover, and a director sold 198,200 shares at $20.15 on August 27.

What the framework says. Probability-weighted EV $28.70 vs spot $14.57 = 1.97x EV/spot, in the "strong asymmetry" band. Composite 7.08. Passes every gate: Asymmetry 8 ≥ 6, Conviction Gap 7.5 ≥ 7 (Moat is 6, so CG carries the quality gate), Leadership 6 ≥ 5, composite > 5.5. Bear case is bounded: no bank debt, $1.9B of take-or-pay contracts with 22% prepaid, the equipment leases are self-liquidating inside each contract's term, and management states liquidity is sufficient for at least twelve months. It is not a high-conviction call, because Moat is 6 and Leadership is capped at 6 by governance flags, and the framework requires 8 and 7 for that.

Recommendation: Investable, initiate now, plan to add rather than chase. The asymmetry is at its widest before the two conversion events, not after: the Q3 print (mid-November) that shows whether the Q4 ARR ramp is on schedule, and the financing structure the CFO promised "additional color" on this quarter, which decides whether the $4-5B hardware procurement gets funded with prepayments and project debt or with the share count. Re-score triggers in both directions are in Section 11.

1. SNAPSHOT

Metric BRUN Notes
Price $14.57 Sept 14 close. −12.4% on the day (CRWV −7.6%, WYFI −8.1%). After-hours $14.81
52-week range $10.30 – $42.00 High on June 3, 2026; −65% since. Closed $22.93 on the Q2 print day (Aug 14)
Shares outstanding 83.85M 54.31M Class A + 29.53M Class B (10 votes/share) as of Aug 25, 2026. Was 61.43M at the May 8 closing
Voting control CEO ~90% of votes Controlled company. Class B held entirely by the founder; ~44.6% economic stake including 7.875M earnout shares
Market cap $1.22B
Cash $120.2M unrestricted + $13.0M restricted at June 30 ~$207M pro forma for the ~$87M of warrant proceeds collected after quarter-end (of ~$130M total; $43.4M was already in the June 30 cash), before Q3 lease payments and capex. Company cited $134.7M on Aug 12 with $74.5M of exercises realized to that date
Debt-like obligations $316.6M Finance leases $238.1M (GPU servers, 30-36 months) + $78.5M of financial obligations ($76.3M present value of a $100M five-year software licensing arrangement with the first installment in Oct 2026, plus $2.2M of financed insurance). No bank debt; bridge loans repaid at closing
Operating leases $119.5M Colocation and data center space
Net debt (pro forma) ~$110M ex operating leases ~$230M including them. On the reported June 30 cash alone it is ~$196M / ~$316M
Enterprise value (pro forma) ~$1.33B ~$1.45B including operating leases; ~$1.42B on reported June 30 cash. Multiples below use the pro forma figure
Q2 2026 $31.1M revenue, +270% YoY Q1 $10.96M. Operating lease income $30.4M of the $31.1M
Q2 costs Cost of revenue $1.7M; colocation $10.6M (34% of rev); adj. SG&A ~$6.4M (20%); D&A $18.1M GAAP SG&A $13.6M includes $7.3M of SBC vesting at closing
Q2 adjusted EBITDA ~$23M lease-adjusted (74%) / ~$12M after colocation (~40%) Company's definition excludes lease expense; the post-colocation figure is the one comparable to CRWV (59%) and NBIS (41%). Consensus was ~$22M
Q2 GAAP Operating loss −$12.9M; net loss −$75.0M Includes a $55.7M one-time deferred tax charge on conversion from LLC to C-corp, $7.3M SBC, $1.4M debt extinguishment
ARR $145M at June 30 $30M at Dec 2025, $96M at April 30. Exit-2026 target ~$400M ("$375M+" in the May listing release, "$400M+" in the June 2 deck, "approximately $400M" reaffirmed in August)
TCV $1.9B $120M at Dec 2025, $360M at April 30, $940M at the May 11 listing, $1.45B at June 1, $1.9B at June 30; >$1B signed in Q2. Avg ~3 years, avg 22% prepaid
Customer deposits $128.4M $34.9M current, $93.5M non-current. Cash already collected for 2026-27 delivery
H1 2026 cash flow CFO +$114.2M (of which +$112.9M deposits); investing outflows −$39.7M; finance lease principal −$99.9M Ex-deposits, operating cash flow was ~$1M, ~−$99M after lease principal and ~−$138M after investing outflows as well ($29.6M equipment, $10.0M lease prepayments) — the build is funded by customers and lessors, not by the P&L yet
FY2025 $26.89M revenue (+239%), FCF +$14.3M, GAAP net loss −$16.3M The Sept 2025 SPAC deck projected 2025E revenue of $31.5M — a ~15% miss
FY2026 consensus $174M revenue (BTIG $182.9M); EPS −$1.29 Management's exit ARR of ~$400M implies a Q4 run-rate near $100M
FY2027 consensus $496M revenue (BTIG $428M, high $551M); EPS +$0.96 BTIG FY27 adj. EBITDA $329M; consensus ~$294M
EV / revenue 7.7x FY26E, 2.7x FY27E consensus, 2.4x FY27 high 3.3x exit-2026 ARR, 9.2x June ARR, 0.7x TCV
EV / MW ~$5.3M per accessible MW (253 MW) NBIS ~$12.8M, CRWV ~$19M, WYFI ~$10M on the same table in June; the miners sit at $1-5M
Power ~253 MW accessible June 2 deck: 33 MW locked + 125 MW under LOI = 158 MW. BTIG (Aug 14): ~125 MW added this year, the bulk of it the 10X Infrastructure Partners lease (up to ~111 MW: 20 MW at an existing energized site in Q4 2026, scaling across two further sites through 2027). Rock Island (6.5 MW total / 4 MW critical, lease from Sept 1) and Pryor OK (144-month lease from Dec 15) signed July/Aug
Hardware B300 at five of eight sites, B200 at one, H200 at one Youngest fleet among listed neoclouds; no 2023-era Hopper estate
Hardware pipeline $1.44B Dell purchase agreement "essentially fully committed"; $4-5B additional procurement in process across OEMs Dell PA minimums: $578M year 1 (already exceeded), $215.5M/yr years 2-5, shortfall payable
Named customers Thinking Machines Lab, RunPod, Shadeform, Manifold Labs, Nebius, Baseten, DRW, fal, Fireworks AI, Fluidstack, Prime Intellect, Higgsfield 2025 concentration: RunPod ~45%, Shadeform ~16%, Manifold ~16% (top-3 76%, down from 94% in 2024)
Certifications NVIDIA Preferred Cloud Partner; Exemplar Cloud on Blackwell (1 of 8, April 13); SOC 2 Type I & II, HIPAA, ISO 27001, ISO 27701 Carahsoft (public sector) and CDW (enterprise) distribution
Analyst coverage BTIG $45 Buy; D.A. Davidson $45 Buy ($20 → $25 → $45); Craig-Hallum $45 Buy (initiated $30); Weiss Sell All three targets set in June; 0% PT velocity over the last 90 days; targets sit 3.1x above spot
Short interest 4.34M shares, 15.1% of float, 1.9 days to cover Float 28.7M. Insider ownership ~64%, institutions ~23%
Insider activity (12 mo) Director Sean Goodrich sold 198,200 shares at $20.15 (Aug 27, ~$4.0M), retains 4.94M. No open-market purchases. No Form 4 sale by the CEO found 336,000-share consulting grant to director Luke Weil fully vested in Q2
Internal controls Three material weaknesses (IT general controls, segregation of duties, accounting personnel); disclosure controls "not effective" NT 10-Q filed for both Q1 and Q2; Q1 10-Q/A filed Aug 18 for controls disclosures only, results unchanged
Next print Q3 2026: mid-November (Q2 was Aug 14) CFO promised financing-structure color "over the next quarter"

2. BUSINESS DESCRIPTION

Boost Run was founded in 2023 by Andrew Karos, who had previously co-founded Blue Fire Capital (a proprietary trading firm that ran its own low-latency compute across six countries and thirteen data centers, acquired by Galaxy Digital) and then ran electronic trading at Galaxy. The company self-funded its first GPU fleet from cash, started in the spot and on-demand market, and extended contract duration as it grew: two months, three months, one year, then three and four year agreements. The operating model has four inputs management repeats on every call: customers (direct sales, channel partners and an on-demand platform), colocation (it owns no data centers), hardware (Dell and Lenovo as OEMs under NVIDIA reference architecture, procured ahead of demand) and finance (customer prepayments, equipment leases and, where needed, balance sheet equity).

How revenue is booked. Under ASC 842 the company is a lessor: once a cluster is provisioned the customer controls the servers, so $30.4M of the $31.1M in Q2 was operating lease income. Contracts longer than a month require an upfront payment; month-to-month customers pay in arrears. Three channels: its own platform (direct GPU rental with orchestration, Kubernetes, storage and API access — this is also where ~12% of revenue comes from short-term on-demand contracts at higher pricing), third-party AI platforms that lack their own hardware (RunPod, Shadeform, Manifold — the resellers that dominated 2024-25 revenue), and GPU brokers. Excess capacity is monetized for blockchain rewards (Bittensor, Aethir), which is immaterial.

The unit economics as disclosed. New B300-class capacity costs roughly $35-40M per MW to deploy (BTIG's payback table assumes $40-60K per GPU and ~1.8 kW per GPU, i.e. ~$25-30M of GPU cost per MW before servers, networking and storage; the company has not published a per-MW figure). Management's contract-value-to-capex ratio is 1.4x "and rising" — the CEO's own worked example on the call used 2:1 — so a $35-40M megawatt carries ~$50-56M of TCV, or ~$16-19M of annualized revenue per MW over a three-year term. Nebius' own Blackwell deals were priced at $20-25M per MW, so the arithmetic is consistent with the market leader. The three contracts with filed cash terms imply $3.59/GPU-hour (Thinking Machines, 5,000 B300s, 36 months, $471.7M), ~$4.11/GPU-hour (June 8 agreement, 240 servers, three years, ~$207.6M with $56.9M paid at or before commencement) and ~$4.13/GPU-hour (July 9 order, 1,536 B300s, 48 months, ~$222.5M with a 25% non-refundable prepayment), against September on-demand B300 list prices of $7.10-7.89 at Modal, Nebius and RunPod. Long-term customers get a ~50% discount for underwriting the hardware's life; the discount is the financing.

The financing model, which is the actual product. The prepayment is quoted against TCV but funds capex, so a 22% average prepayment on a 1.4x TCV/capex contract covers ~31% of the hardware bill before any lease draws. The remainder is financed by 30-36 month finance leases on the GPU servers (34 new lease agreements in H1 2026; $268M of right-of-use assets added), with a purchase option at term-end, on contracts that run three to four years — so the lessor is repaid before the customer contract is 60-75% through. The company describes the outcome as a "sustainable net cash flow margin of 15-20%." The honest reading of the H1 cash flow statement is that this is a design, not yet a demonstrated result: operating cash flow ex-deposits was ~$1M, and after lease principal and investing outflows it was ~−$138M ex-deposits, because deposits arrive at signing and depreciation-heavy revenue arrives over 36 months. That mismatch is exactly what the prepayment structure exists to bridge, and it is why the balance sheet shows $128.4M of customer deposits against $238.1M of finance leases.

The site strategy. "Four 25 MW sites over one 100 MW site, every time." Smaller, already-energized brownfield colocation sites near population centers, using a templated cluster design that has passed NVIDIA's architecture review board. The stated advantages are time-to-revenue (the 10X site announced August 14 is already powered and targeted for production by end of Q4), containment of single-site failure, and inference latency. The stated cost is that nobody gets a gigawatt headline.

Why NVIDIA matters here more than at most neoclouds. Exemplar Cloud status requires NVIDIA to stress-test the full stack at FP4 and FP8 and certify a 95% performance guarantee; at the time Boost Run received it on B300, the only other holder was Oracle Cloud. Every cluster passes NVIDIA's architecture review board before production. Management ties that directly to two things: customer handoff quality (the sector's most common complaint is rented compute that underdelivers against spec) and lender comfort (a third party has validated the collateral). It also appears to be a channel: a company running ~$6.4M of quarterly SG&A with engineers rather than a sales force does not land a frontier lab on cold outreach, and the CEO's own language — "getting this backlog put in front of our face," "seeing all traffic from all angles come across our desk" — describes inbound referral, not outbound sales.

3. GROWTH DRIVERS

  1. $1.9B of contracted TCV converting to revenue through Q1 2027. Management expects the full book to be "live in production" by Q1 2027. At an average three-year duration that is ~$630M of annualized revenue from what is already signed, against $145M of ARR at June 30 and a $400M exit-2026 target. Evidence: Q2 press release, call, 10-Q Notes 18 and 21 (two contracts with filed cash terms totaling ~$430M, 25-27% prepaid). Counter-evidence: the June 8 contract starts November 30, 2026 and the July 9 contract starts March 2027, so a material slice of the $1.9B is 2027 revenue, and Q2 revenue missed consensus by ~9% on deployment timing.
  2. ARR $30M → $145M in six months, guided to ~$400M in six more. The Q4 exit rate implies roughly $100M of quarterly revenue by December against $31.1M in Q2. That requires roughly 22-33 MW of energized capacity depending on fleet mix, against six live sites plus 20 MW from 10X in Q4, Rock Island from September and Marietta in Q3. The capacity exists; the question is energization timing.
  3. 253 MW of accessible power against a fleet that today occupies a fraction of it. The 10X Infrastructure Partners lease (20 MW at an existing energized site in Q4, scaling to 111 MW across two further sites through 2027), Rock Island and Pryor, Oklahoma (144-month lease commencing December 15) took accessible power from ~125 MW at listing to ~253 MW. At $16-19M of ARR per MW that footprint supports $4B+ of run-rate revenue — the constraint is hardware financing, not megawatts.
  4. The $4-5B hardware procurement. Dell's $1.44B agreement is "essentially fully committed and allocated." Management is negotiating $4-5B more across multiple OEMs, sized against "active discussions with some of the largest GPU consumers in the world, with requirements ranging from 10,000-50,000 GPUs" — NVIDIA itself, investment-grade financial institutions and frontier labs. A single 50,000-GPU customer would be larger than the entire current fleet. Nothing is signed; management says it will announce contracts only when they are contracts.
  5. On-demand as a margin lever. More than 12% of revenue comes from short-term on-demand contracts through the platform, at roughly double the multi-year rate per MW. Lenders will not finance uncontracted capacity, so only self-funded operators can hold inventory on that shelf; Boost Run's day-one self-funding and prepayment model lets it "trade the whole curve." BTIG notes B200 spot pricing in August was +25-30% YTD with Q3 tracking +5-10% sequentially.
  6. Regulated-industry demand. SOC 2, HIPAA, ISO 27001 and ISO 27701 in place; Carahsoft for public sector; CDW (whose CEO named Boost Run on CDW's Q1 call) for enterprise distribution. Management flagged financial services and healthcare as diversification targets. Note the erosion: Nebius has held the same certifications since October 2025, and BTIG expects others to follow; this is a first-mover window, not a permanent moat.
  7. NVIDIA programs. The CEO referenced the AI Cloud Partner program and the $500B ecosystem financing consortium three times on the call ("at the appropriate time, we can get into more details"; "a plentiful amount of optionality"). NVIDIA has written rent-back backstops for CoreWeave ($6.3B), Lambda ($1.5B), Firmus and Sharon AI. Any comparable arrangement would convert the funding question from "can they?" to "how much?"

4. LEADERSHIP ANALYSIS

Key executives.

  • Andrew Karos, Founder, CEO and Chairman (since 2023). Former Managing Director and Head of Electronic Trading at Galaxy Digital and executive committee member (2020-23); co-founder and CEO of Blue Fire Capital, which operated across six countries and thirteen data centers before Galaxy acquired it. Derivatives and mathematics background; on the Q2 call he worked the prepayment-to-capex arithmetic live and described the FP4/FP8 certification protocol unprompted. Took no salary in FY2024 or FY2025 and has no employment agreement; his compensation was the 7.875M earnout shares, which vested on $12.50/$15/$17.50 VWAP hurdles. Holds all 29.53M Class B shares (10 votes each) plus the earnout shares: ~90% of votes, ~45% of economics. Lent the company $1.43M in November 2025 (repaid at closing) and received an $8.5M installment note at closing as partial consideration for his LLC units (settled at closing).
  • Harry Georgakopoulos, COO (since April 2024). Former MD at Galaxy Digital; Karos' co-builder at Blue Fire Capital. Author of Quantitative Trading with R (Palgrave Macmillan, 2015); MS Financial Mathematics, University of Chicago. The same bench, second company.
  • Erik Guckel, CFO. PhD Chemical Engineering (Illinois), Chicago MBA; has closed $2B+ of corporate transactions, managed a $3B debt portfolio and financed first-of-a-kind facility construction. Salary $250,000. He owns the financing-structure promise for this quarter.
  • Daniel Gormley-Rahn, CTO, runs deployment of thousands of GPUs across sites and the network/security architecture. Karim Ali, CIO, ~20 years of performance-sensitive infrastructure, built ultra-low-latency trading networks on four continents, led a FedRAMP certification at a SaaS security company.
  • Board. B. Luke Weil (former SPAC chairman/CEO; controls the Sponsor; consulting agreement signed January 13, 2026 for 336,000 price-vested shares) and Sean Goodrich (owns the SPV that received 1.97M earnout shares; sold 198,200 shares on August 27) are both directors. The company is a "controlled company" under Nasdaq rules.

Insider ownership and activity. Insiders ~64% of shares. The only open-market transaction in the last twelve months is Goodrich's $4.0M sale at $20.15 after exercising 1.1M warrants; he retains 4.94M shares. No open-market purchases by anyone. No CEO sale on file. TOMS Capital filed a 13G for 3.1M shares (9.8% of Class A at the time) in August.

Guidance track record (short, because the company is young).

Item Original Outcome Call
2025E revenue (Sept 2025 SPAC deck) $31.5M $26.89M actual Miss, ~15% (2H25 ~$14.3M vs $19.4M projected)
2026E revenue (SPAC deck) $170-190M Consensus $174-183M; H1 actual $42.1M with a back-loaded ramp On track, back-half weighted
2026 exit ARR $275M (original deal materials) → "$375M+" (May listing) → "$400M+" (June deck) → "~$400M" (Aug) — Raised, then reaffirmed
Contracted revenue $940M at listing (May) $1.9B at Q2 Beat, doubled in one quarter
Q2 2026 vs sell-side ~$34M revenue / ~$22M adj. EBITDA $31.1M / ~$12M post-lease (~$23M lease-adjusted) Miss on the quarter, beat on the trajectory
Net cash flow margin "15-20%" Not yet demonstrable; H1 CFO ex-deposits ~$1M Unproven

This is a raise-the-bar pattern on the forward metrics (ARR, TCV) and a miss on the one dated projection that has fully aged (2025 revenue), plus a quarterly consensus miss on the first print. It is not a reset-the-bar pattern; the 6/10 cap for that reason does not apply.

Capability for the next bound. The transition is from a $31M-a-quarter operator deploying a $1.44B hardware book with equipment leases to a company financing $4-5B of additional hardware through project-level debt, prepayments and possibly NVIDIA-linked structures, while remediating material weaknesses and reporting as a public company. Karos and Georgakopoulos have built one compute-heavy business together and sold it; Karos has "monetized billions in credit facilities" in prior roles; Guckel has run a $3B debt book. None of them has run a public company or a multi-billion-dollar project-finance program. Credentials and skin in the game, unproven at this scale: the framework floor is 7 before governance.

Red flags — and why Leadership is 6, not 7. The framework caps Leadership at 6 for governance red flags including related-party transactions and board-independence issues, and this file has several ordinary-for-a-SPAC but real ones:

  • Related-party transactions. A CEO loan to the company; an $8.5M installment note to the CEO at closing; a consulting agreement signed in January 2026 with the SPAC's then-chairman, now a Boost Run director (Weil), paid in 336,000 price-vested shares that fully vested inside the first quarter; earnout shares to entities controlled by two directors.
  • Controlled company with a dual-class structure. Ten-vote Class B gives the founder ~90% of votes at ~45% of economics. Minority holders are along for the ride on capital allocation and any future equity issuance.
  • Internal controls. Three material weaknesses (IT general controls, segregation of duties, insufficient qualified accounting staff), disclosure controls concluded "not effective," late filings for both quarters this year, and a 10-Q/A on the Q1 controls disclosure. Remediation is expected to "continue through 2026." There has been no restatement of results, so the 1-3 floor for material restatements does not apply.
  • Earnout design. All 10.97M earnout shares (~18% dilution) vested within one quarter of listing because the hurdles were set at $12.50-$17.50 against a stock that ran to $42. Defensible as pay-for-performance; it also means the founder was paid in full before the first public 10-Q.
  • Sponsor promote fully retained (4.63M founder shares, no forfeiture) and the SPAC's own bookrunner (BTIG) is one of the three covering analysts, all at $45.

Against that: zero redemptions, no PIPE, no discounted share class, an unused sponsor loan facility, warrants retired into $130M of cash rather than left as a five-year overhang, no cash salary for the founder, and a management team that talks about the business in FP4 validation runs and TCV-to-capex ratios rather than megawatt press releases. Leadership 6 is the governance cap, not a judgment on competence.

5. FORWARD OUTLOOK — TAILWINDS + GUIDANCE

Structural tailwinds (24-36 months).

  1. Compute demand is still supply-constrained at the leading edge. CoreWeave's Q2: revenue $2.575B (+112%), backlog ~$104B plus $25B added early in Q3, active power 1.5 GW, contracted 3.7 GW. Nebius' Q2: revenue $582M (+454%), ARR $3.0B, year-end contracted power target raised to 5 GW, guided 2026 exit ARR $7-9B, customer prepayments expected at $9B+. B300 on-demand capacity is "thin" because most of it is committed to long-term contracts. Captured: Boost Run sells into exactly this shortage at the small-to-mid deployment size (5,000-50,000 GPUs) the hyperscalers do not prioritize.
  2. Pricing has held or risen on Blackwell. BTIG: B200 spot +25-30% YTD through August, Q3 +5-10% sequential. The H100 precedent (≈$7-10/hr in 2023 to ≈$2/hr by late 2025, an ~80% decline) is the bear's exhibit, and it is the reason the framework discounts anything not on the leading edge. Captured while the fleet is current-generation; adjacent risk as B300 ages against Vera Rubin.
  3. Enterprise and regulated adoption. Deloitte survey data cited by BTIG: 84% of 3,000+ enterprises increasing AI investment, 73% citing data privacy/security as the top concern; ~30% of the Fortune 500 already paying for AI services. Adjacent: the compliance stack opens the door, but the named book today is AI-native (labs, inference platforms, resellers, a trading firm), not hospitals and banks.
  4. NVIDIA financing the channel. Rent-back backstops for CoreWeave, Lambda, Firmus and Sharon AI; a reported ~$2.5B NVIDIA investment in Thinking Machines Lab's $5-6B raise at a $40B+ pre-money (in talks as of September 4, not closed). Adjacent, with a named path: management has publicly signaled it is working on NVIDIA's AI Cloud Partner program.

Most recent guidance, exact language (Q2 call, Aug 14).

  • "We currently expect to exit fiscal 2026 with approximately $400 million of annualized reoccurring revenue." Reaffirmed in closing remarks.
  • "We expect to continue deploying capacity through fiscal year and into Q1 2027, bringing the full $1.9 billion of TCV live in production."
  • "We currently expect to achieve a sustainable net cash flow margin of 15%-20% moving forward."
  • "We are expecting to hand over four projects by year-end."
  • "We are in the process of a strategic procurement of an additional $4 billion-$5 billion of compute hardware with multiple OEMs."
  • "We continue to innovate on financing structures beyond the equipment financing approach used to date and expect to provide additional color and progress on this over the next quarter."
  • 10X site: "already powered, and we plan to bring this into production in end of Q4."

What management explicitly did not guide to: quarterly revenue, FY2026 revenue, FY2027 anything, GAAP profitability, capex for 2027, customer concentration for 2026, the terms or counterparties of the $4-5B procurement, or any NVIDIA financing arrangement.

Gap analysis. The June ARR of $145M annualizes to ~$36M a quarter; the exit target requires ~$100M a quarter by December, so Q3 needs to land somewhere near $50-60M (BTIG models $56M) and Q4 near $80M (BTIG $81M) for FY2026 of ~$175-185M. Consensus FY2027 of $496M is ~2.8x the FY2026 base and sits comfortably inside what $1.9B of three-year TCV plus modest new signings supports (~$630M annualized from the existing book alone). Applying the leadership discount from the one aged projection (a 15% miss on 2025 revenue) to the exit-ARR target gives ~$340M, which still implies FY2027 revenue above BTIG's $428M. At spot, EV/FY27 consensus of 2.7x is below every listed neocloud comp (Nebius 5.1x its $12.1B 2027 consensus; CoreWeave ~7x 2026E), so the valuation requires the book to convert on roughly the stated schedule, not that the $4-5B lands.

What we are watching on the Q3 call (mid-November).

  1. Q3 revenue at or above ~$50M and the exit ARR reaffirmed at ~$400M with the four year-end project handovers on schedule. Anything that moves the "full TCV live" date beyond Q1 2027 is a downgrade trigger.
  2. The financing structure: named lenders, a facility size, an NVIDIA program reference, or a project-level debt tranche against a named contract. This is the single item that converts the conviction gap.
  3. Customer concentration and named signings from the $4-5B pipeline. A second frontier-lab or investment-grade counterparty at ≥$200M TCV would move both Moat and Revenue Quality.
  4. Material weakness remediation progress and an on-time 10-Q.

6. BULL CASE

The book converts on schedule and the market re-learns what it is looking at. Exit-2026 ARR of ~$400M lands in the Q4 print in March 2027, FY2027 revenue clears $500M on the existing $1.9B plus new signings, and post-colocation adjusted EBITDA margins move from 40% toward 50%+ as revenue grows faster than colocation cost. The financing announcement arrives in the form of project-level debt or an NVIDIA-linked structure, and the first $1-2B tranche of the $4-5B procurement is contracted against a named frontier lab or investment-grade counterparty with a 20%+ prepayment. Thinking Machines closes its $40B+ round with NVIDIA in it, expands or renews with its incumbent validated-B300 supplier, and the market recognizes the referral channel. The sector multiple recovers from the September trough. On 12x BTIG's FY2027 adjusted EBITDA of $329M, or ~7x the high end of FY2027 consensus revenue, the stock is $40-45 — which is where all three covering analysts already have it. That is a 2.9x from spot with nothing in it that management has not already said it is doing.

The stronger version: the entire $4-5B is financed and contracted by mid-2027, exit-2027 ARR reaches $1.5B on ~80-95 MW of the 253 MW already accessible, and the company is valued the way Nebius is valued — 5x forward ARR — for $75, a 5x. That requires everything to land in a 15-month window, which is why it carries 12% and not 25%.

The structural argument underneath both: at $5.3M of EV per accessible megawatt, a pure B300 neocloud with NVIDIA Exemplar status is priced where the market prices bitcoin miners whose megawatts earn $1-2M a year. The megawatts here earn ~$12M blended and $16-19M on new builds. Either the power is not real, the contracts are not real, or the multiple is wrong. The power is leased and energized, the contracts have $128M of cash deposits against them, and the last twelve months have been a demonstration that the multiple is the variable.

7. BEAR CASE

The bear is not fraud and it is not going-concern. It is a slow version of the neocloud cycle playing out on a company with less capital than its peers:

  • Deployment slips. Q2 missed consensus revenue by ~9% on timing. If Q3 prints $45M instead of $55M and the exit ARR slides to "Q1 2027," the credibility of every forward number takes a haircut in a sector that has stopped paying for promises. The June 8 and July 9 contracts do not start until November 30, 2026 and March 2027.
  • The financing never arrives at scale. Equipment leases at $100-250M a tranche funded a $1.44B book. A $4-5B book needs $2-3B of project debt per year, ten times the historical pace, from lenders that just watched the 30-year Treasury touch a 19-year high. If it comes as equity instead — the founder controls 90% of the vote and needs no minority approval — the share count does the financing.
  • Concentration and counterparty quality. Top-3 customers were 76% of 2025 revenue and the largest, RunPod, is a reseller. The named additions (a frontier lab in the middle of a down-round-then-up-round cycle, a competing neocloud, inference startups) are better but not investment-grade. A single counterparty failing to pay past its prepayment is a real hole in a $145M ARR base.
  • GPU pricing rolls over. The H100 curve fell ~80% in 30 months. B300 pricing is rising today because supply is short; Vera Rubin in 2027 and hyperscaler self-build (Meta's compute-resale plans, Microsoft, SpaceX) are the supply. Contracted revenue is protected for its term; the renewal and the on-demand 12% are not.
  • Internal controls and disclosure. Three material weaknesses, two late 10-Qs and an amended filing in the first four months as a public company. The next disclosure surprise is more likely than at a company with clean controls.
  • Supply. 58.7M shares registered for resale, lock-ups already released, 15% of the float short, and a director already selling at $20. The founder has not sold, but nothing prevents it.

Bear midpoint $7 (−52%). Anchored to a real precedent: the stock itself fell 65% from June to September without any of the above happening, WhiteFiber fell 61% over the same period, and Applied Digital fell ~80% in 2022-23. At $7 the company trades at ~2.5x a bear-case FY2027 revenue of $330M (23% below BTIG) with net debt grown to ~$250M — roughly where a levered, decelerating small-cap infrastructure name sits at the bottom of a cycle. The tail below that (a counterparty default plus a failed financing plus an equity raise at the lows) exists and is why Bear carries 25% rather than 15%, but it is bounded by $1.9B of take-or-pay contracts with 22% prepaid, no bank debt, self-liquidating leases and a twelve-month liquidity statement in the 10-Q.

8. SCORING

Dimension Score Reasoning
Revenue Quality (RQ) 6 Real, audited, GAAP lease income growing 270%; $1.9B of take-or-pay TCV with $128M of cash deposits; ~12% on-demand at premium pricing. Against: top-3 customers 76% of 2025 revenue with a reseller at 45%; 2026 concentration not yet disclosed; three material weaknesses; the "net cash flow margin" is a construct the cash flow statement does not yet show. Descriptive, not predictive: a 6.
Revenue Growth (RG) 9 "Multi-year structural growth with confirmed customer commitments" is the 9-10 bracket by definition. $1.9B contracted over ~3 years against $145M ARR; exit ARR guided to ~$400M; FY2027 consensus $496M (~2.8x FY2026); the $4-5B procurement is upside not in these numbers. Not 10 because the H1 revenue miss shows timing risk on the ramp.
Moat / IP (M) 6 Base 5: no proprietary silicon or software of note; the durable advantages are process and capital-structure (NVIDIA reference-architecture templating, Exemplar validation, prepayment-on-every-deal, self-funded on-demand inventory, small-site speed) plus a compliance stack that competitors can and will replicate. Validator stack, counted strictly: "3+ strategic partners with capital at risk" does not fire (Dell has a purchase agreement and financing exposure, NVIDIA and Lenovo are vendors, CDW is a channel — no equity at risk); "2+ sovereign/hyperscale customers" does not fire (Thinking Machines is a lab, Nebius is a neocloud); "5+ recurring named deployments" fires (+0.5, twelve named customers); "reference design alignment with a platform leader" fires (+0.5, NVIDIA Exemplar and architecture review board). 5 + 1.0 = 6.
Leadership (L) 6 Floor 7 for credentialed, skin-in-the-game, unproven-at-scale founders (no salary, ~45% economic stake, prior compute business built and sold, technical fluency). Capped at 6 by governance: related-party transactions (CEO loan and note, director consulting shares, director-controlled earnout entities), controlled-company dual class, three material weaknesses and two late filings. One aged projection missed by 15% (2025 revenue); forward targets raised twice. Not a reset-the-bar pattern; no restatement.
Asymmetry (A) 8 EV/spot 1.97x (Section 9). In the 1.8-2.5x "strong asymmetry" band. Bear anchored to the stock's own June-September drawdown and to WhiteFiber's; Bull anchored to the three published $45 targets and 12x BTIG FY2027 EBITDA; Moonshot capacity-checked against 253 MW. Sensitivity: 30/37/23/10 gives 1.80x, 35/35/22/8 gives 1.67x — still above the 1.5x real-asymmetry line.
Conviction Gap (CG) 7.5 (a) Unknown today: whether the $400M exit ARR lands in Q4 or slips; what the financing structure is and whether it dilutes; who the $4-5B is for; 2026 customer concentration; whether the NVIDIA channel is real or coincidence. (b) Evidence: positive — >$1B signed in Q2 with cash prepayments, ARR up 4.8x in six months, an already-energized 20 MW site for Q4, three $45 targets against a $14.57 stock (3.1x spread between spot and the sell-side, wider than the 2.5x dispersion threshold even though the three targets agree with each other), and a frontier lab raising $5-6B; mixed — first-print consensus miss, controls, 15% short interest, the sector telling you it no longer believes any of these stories. (c) Conversion event: the Q3 print and the financing announcement, both inside the next 60-90 days, then the Q4 exit-ARR print in March 2027. (d) Timeline: 2-6 months. Wide gap, evidence directionally positive, near-term catalyst: 7-8 bracket. 7.5 rather than 8 because the controls and the consensus miss are real negative evidence, not just thin coverage.
Composite 7.08 (6 + 9 + 6 + 6 + 8 + 7.5) / 6

Gates. Investable threshold: Asymmetry 8 ≥ 6 ✔; Moat 6 < 7 but Conviction Gap 7.5 ≥ 7 ✔ (CG carries the quality gate); Leadership 6 ≥ 5 ✔; composite 7.08 > 5.5 ✔. The high-conviction gate requires Asymmetry ≥ 8 ✔, Moat ≥ 8 ✘, Leadership ≥ 7 ✘. Investable, not high-conviction.

9. PRICE TARGET FRAMEWORK AND VERDICT

Scenarios, 12-18 month horizon, ~87M shares (allowing for equity-plan issuance), valued on FY2027 or exit-2027 metrics.

Scenario Prob. Price vs spot What has to be true Comparable and multiple
Bear 25% $7 −52% Q3-Q4 deployment slips two quarters; exit ARR ~$300M lands in Q1-Q2 2027; financing arrives late or as equity; FY2027 revenue ~$330M; sector multiple stays at the September trough; resale supply and 15% short interest keep pressing. 2.5x EV/FY27 revenue on $330M less ~$250M net debt. Precedent: BRUN −65% Jun-Sep 2026; WYFI −61% same period; APLD ~−80% 2022-23.
Base 37% $19 +30% Exit ARR ~$400M delivered within a quarter of plan; FY2027 revenue ~$465M (between BTIG's $428M and consensus $496M); post-colocation adj. EBITDA ~45% (~$210M); a financing structure announced but conventional (project-level equipment debt); one or two new named contracts; multiples flat to slightly better. 4.5x EV/FY27 revenue less ~$300M net debt = ~$21; 8x post-lease EBITDA = ~$16; midpoint $19. Comparable: Nebius at 5.1x 2027E revenue today; CoreWeave at ~7x 2026E.
Bull 26% $42 +188% Exit ARR at or above $400M on time; FY2027 revenue at the high end of consensus (~$550M); a named non-dilutive financing facility (project debt or NVIDIA-linked) plus the first $1-2B of the $4-5B procurement contracted to a named counterparty; Thinking Machines round closes and the relationship expands; sector multiples recover partially. 12x BTIG FY2027 adj. EBITDA of $329M less ~$400M net debt = ~$41; ~7x FY27 high-consensus revenue = ~$37-43. Anchored to the three published $45 targets (BTIG, D.A. Davidson, Craig-Hallum). Comparable: Nebius traded at 6.9x 2027E in June 2026.
Moonshot 12% $75 +415% The full $4-5B is financed and contracted by mid-2027; exit-2027 ARR ~$1.5B on ~80-95 MW of the 253 MW accessible (feasible on power; needs ~$3-3.5B of hardware energized inside 15 months, which is the constraint); NVIDIA program participation or investment; re-rating to leader multiples. 5x exit-2027 ARR less ~$800M net debt = ~$77. Comparable: Nebius at ~5x forward ARR today and 22.6x 2026E revenue at its June peak; CoreWeave 8-10x forward revenue at its inflection. Cannot sit below the most bullish broker target ($45) and does not.

Probability-weighted EV: $28.70. EV / spot = 1.97x. Asymmetry score 8.

Capacity check (Refinement 8): $400M of ARR needs ~22-33 MW depending on fleet economics against six live sites plus 20 MW in Q4 — feasible. $1.5B needs ~80-95 MW against 253 MW accessible — feasible on power, constrained by ~$3-3.5B of hardware financing, which is why it is the Moonshot and not the Bull. Revenue anchoring (Refinement 6): Bull revenue ($550M) is at the high end of published FY2027 forecasts, not below them; Base ($465M) sits between BTIG and consensus; only Bear is below the lowest broker. PT velocity (Refinement 1): 0% over 90 days — targets were set at $45 in June and have not moved while the stock fell 65%, so "the stock is below consensus" is not a lag signal in either direction; the sell-side has simply not re-underwritten since the print. Dispersion (Refinement 2): the three published targets agree ($45/$45/$45, 1.0x spread), which would ordinarily read as a narrow gap, but the 3.1x spread between spot and every target, a Weiss Sell, and 15% short interest describe a market that disagrees with the sell-side entirely; the gap is between the sell-side and the tape, not within the sell-side.

Verdict per the decision rules. Asymmetry 8 (EV/spot 1.97x), bear case bounded, conviction gap wide with a conversion catalyst inside 90 days, Moat 6 and Leadership 6 below the high-conviction thresholds. That is investable, and at 1.97x it is the strong end of investable: initiate now. Not "wait to be safe": the framework forces a decision at ≥1.2x and the math is at 1.97x. Not high-conviction: Moat 6 and Leadership 6 fail those gates on the merits, not on caution. Plan to add only on the upgrade triggers below, because the two events that would justify more are also the two events that would eliminate most of the discount.

Self-check (Framework Memo questions 1-12). (1) EV/spot calculated at 1.97x; the 65% drawdown is context for the Bear anchor, not a scoring input. (2) Thin coverage is not the reason CG is 7.5; the controls and consensus miss are, and they are marked as negative evidence. (3) No "wait for the print" recommendation — EV/spot is above 1.2x so a decision is made now. (4) Bear at −52% is anchored to the stock's own realized drawdown and a peer's; it is severe because the precedent is severe, not to bias the EV. (5) Decided on conviction: "initiate now" is what the rules produce, not a hedge. (6) Structurally this is the thin-coverage, wide-gap, contracted-evidence, catalyst-ahead setup the framework was built to act on, not a name that has already confirmed. (7) PT velocity 0%; not used as a downward input. (8) Sell-side dispersion 1.0x but spot-to-target 3.1x; treated as wide. (9) Every Bull/Moonshot midpoint names a comparable and a multiple. (10) Bull revenue at the top of broker forecasts, not below. (11) Validator brackets applied literally: two half-brackets fire, +1.0, no interpolation. (12) Capacity reconciled against 253 MW and hardware financing.

10. PORTFOLIO FIT

Layer: Compute — the GPU cloud that is the customer of every other layer in the AI data center supply chain (servers, optics and interconnects, substrates, power, cooling, test). A book built around suppliers to AI infrastructure is short the demand side by construction; a neocloud is the demand side. That is a diversifying exposure in thesis terms and a concentrating one in factor terms: on a day the AI-infrastructure trade sells off, this name falls harder than the suppliers, not less.

Correlation and stacking. Boost Run trades as a high-beta version of the listed neocloud complex (CoreWeave, Nebius, IREN, WhiteFiber). Anyone already holding a neocloud should treat this as the same factor with higher volatility rather than as a separate idea, and manage the combined neocloud exposure, not each name. It does not stack with the optical interconnect or glass substrate layers except through the sector beta.

What it adds that the suppliers do not. Direct exposure to GPU rental pricing (which rose in 2026 on Blackwell while supplier stocks sold off), to NVIDIA's channel-financing decisions, and to the enterprise/regulated-industry adoption curve. What it lacks that the suppliers have: proprietary technology, qualification-cycle switching costs and a moat that survives a pricing downturn.

Liquidity. ~$30M of average daily dollar volume at current prices (~2M shares a day); liquidity is not a constraint. Float is 28.7M shares, so the stock gaps on news in both directions; use limits.

11. KEY CATALYSTS AND RE-SCORE TRIGGERS

Date Event What matters
Sept 2026 Rock Island (6.5 MW) lease commencement (Sept 1); Marietta online (Q3) Sites energized on schedule
Sept-Oct 2026 Thinking Machines Lab round ($5-6B at $40B+ pre; NVIDIA ~$2.5B) Closing removes the largest customer's funding question; any Boost Run expansion or renewal is a Moat and RQ upgrade
Oct 2026 First $20M installment on the $100M software licensing obligation Cash step-down; watch the Q3 cash balance against it
Mid-Nov 2026 Q3 2026 print Revenue ≥ ~$50M; exit ARR reaffirmed; four project handovers on track; 2026 customer concentration; material-weakness remediation; on-time filing
Q4 2026 Financing structure announcement ("additional color over the next quarter") The conversion event. Named facility, project debt, or NVIDIA-program reference = upgrade. An equity raise or ATM = downgrade
Q4 2026 10X site (20 MW) into production; four project handovers; Pryor OK lease commences Dec 15 Exit-ARR delivery mechanics
Ongoing $4-5B procurement announcements; new customer agreements ("we will announce them only when they are contracts") Any named counterparty ≥ $200M TCV with a ≥20% prepayment
Q1 2027 Full $1.9B TCV live; July 9 contract (1,536 B300s) commences March 2027 Run-rate revenue ~$100M+/quarter
Mar 2027 Q4 2026 print and first 10-K The exit-ARR number becomes a reported number; first audited public-company year
2027 10X capacity scaling to 111 MW; Vera Rubin transition Fleet-age advantage either extends or starts to erode

Upgrade triggers (add; re-score Moat and RQ): - A named non-dilutive financing facility of ≥$500M (project debt, DFS-style vendor financing at scale, or an NVIDIA-linked structure). - A second frontier-lab or investment-grade counterparty at ≥$200M TCV, or a Thinking Machines expansion; a hyperscale-class named customer would fire the "2+" validator bracket and move Moat to 7. - Q3 revenue ≥ $55M with exit ARR reaffirmed and an on-time, clean 10-Q. - Material weaknesses remediated (Leadership cap lifts toward the 7 floor).

Downgrade / exit triggers: - Exit ARR pushed beyond Q1 2027, or Q3 revenue below ~$42M with no timing explanation tied to a specific site. - Equity issuance to fund hardware while the stock is below $20, or any ATM program. - A restatement, auditor change, SEC inquiry or a counterparty default past its prepayment — Leadership floors and the position is exited regardless of price. - CEO open-market sales of size before the exit-ARR print. - Blackwell contracted pricing on new signings below ~$3/GPU-hour, or on-demand B300 below ~$5 (the beginning of the H100 curve).

APPENDIX — SOURCES

Boost Run Q2 2026 Form 10-Q (filed Aug 18, 2026), Q2 2026 results press release and 8-K (Aug 14), Q2 2026 earnings call transcript (Aug 14; Quartr/Yahoo), 10X Infrastructure Partners lease release (Aug 14), warrant exercise and redemption releases (July 6, July 27, Aug 26) and 8-K (July 27), Form S-1 resale registration (July 2; effective July 10) and 424B prospectus supplements, Q1 2026 10-Q/A (Aug 18), Investor Update presentation (June 2, 2026), SPAC investor presentation (Sept 14, 2025), Nasdaq listing release (May 11, 2026), business combination close release (May 8); Forms 4 and Schedule 13D/A for Sean Goodrich (Aug 28), Schedule 13D for B. Luke Weil (June 15), Schedule 13G for TOMS Capital (Aug 14); BTIG initiation (June 25, 2026, Buy $45) and Q2 note (Aug 14, 2026, Buy $45); D.A. Davidson (May 13, June 1) and Craig-Hallum (May 12, June 23) rating notes via Investing.com and GuruFocus; StockAnalysis (price history, statistics, forecasts) for BRUN, CRWV, NBIS, IREN, WYFI, SHAZ (Sept 14-15, 2026); MarketBeat analyst summary; Insider Monkey insider table; CoreWeave Q2 2026 results release (Aug 2026); Nebius Q2 2026 results coverage; Thunder Compute B300/B200 pricing guide (Sept 11, 2026); Yahoo Finance / Inc. / Seeking Alpha on the Thinking Machines Lab funding talks (Sept 4, 2026); Benzinga and Invezz on the Sept 14, 2026 neocloud sell-off; Yahoo Finance on the Aug 18, 2026 sector sell-off; Simply Wall St consensus tables for CRWV and NBIS.

Not financial advice. Prices are exchange closes. Author may hold a position.