Daily Investment Brief — June 5, 2026

All four charts rendered. Here is the final daily brief:

GPU & Cloud Compute Daily Brief — June 5, 2026

Cross-provider pricing data · SEC filings · Regulatory intel · Equity signals

Market Pulse

The dominant narrative in GPU compute markets is a structural geographic bifurcation that is no longer a theory — it is a live, 90-day-documented pricing phenomenon. H100 SPOT in clean-power corridors (Montreal +104.6% over 30 days to $4.26/hr; Stockholm +63.1% to $3.36/hr; Madrid +35% to $2.81/hr) is in an uninterrupted multi-week recovery, while grid-constrained markets (Dublin -49.9%, Frankfurt -26.8%, Singapore -18.4%) remain in free-fall. The mechanism is explicit: power infrastructure is the binding constraint, and capital is repricing toward where it already exists. Layered over this, a generation-transition dynamic is hardening — AWS custom silicon is bifurcating sharply between Inferentia SPOT (Tokyo +772%, Mumbai +881% over 30 days as pool capacity tightens) and Trainium SPOT (Oregon -70% over 30 days as first-gen training silicon depreciates toward zero). The H100 training-era compute cycle is not over, but the next cycle's winners — clean-power-anchored infrastructure and Blackwell-era liquid-cooled density — are already being priced in by the equity market.

Key Movers

Investable Insights

H1 — Power-Secured Markets Command a Permanent Greenfield Premium Confidence: 5 / 5

Thesis: What started as a cyclical observation has hardened into a structural market fact: compute capacity in clean-power/greenfield corridors commands a measurable, durable pricing premium over grid-constrained markets. The Montreal H100 SPOT 90-day history makes this explicit — prices held a stable $4.99–$5.01 plateau through mid-March even as other markets softened, then derating began only when a single provider cut floor pricing unilaterally in April. The trough of $1.90 on May 11 was sharp but brief; the 25-day uninterrupted recovery to $4.26 represents the fastest and most sustained H100 SPOT recovery in the current dataset across any region. Crucially, Montreal ON_DEMAND has sat immovably at $7.68/hr throughout this entire cycle — flat 0.0% over 30 days — confirming providers are making no structural concessions on list price while spot capacity tightens back toward it. Stockholm's spread_pct explosion from 21.9% to 62.2% within a single week is the most precise real-time signature of a market mid-tightening, not post-tightening; the move there still has runway. The mechanism driving this bifurcation — PJM's $23B capacity cost shock, 14-year Northern Virginia grid interconnection timelines, Florida SB 484 removing utility cost socialization, and NERC's new "emerging large loads" regulatory category — creates rising barriers to new capacity in constrained markets while Quebec hydro, Nordic hydro, and Southeast Asia greenfield sites face no such friction. Capital follows physics: compute will physically relocate toward power.

Key Evidence:

1. Montreal H100 SPOT: $1.90/hr trough May 11 → $4.26/hr June 5, +124.2% in 25 days; two-provider market with spread_pct now 140% — both providers simultaneously raising floor pricing (live pricing data, June 5, 2026)
2. Stockholm H100 SPOT: +63.1% over 30 days, spread_pct 21.9% → 62.2% in one week — independent clean-power market corroborating the pattern approximately 3 weeks behind Montreal in its recovery cycle (live pricing data, June 5, 2026)
3. Dublin H100 SPOT: -49.9% over 30 days; Frankfurt -26.8% over 30 days. These are single-provider or shallow-pool grid-constrained EU markets with no sign of recovery on a 7-day horizon (live pricing data, June 5, 2026)
4. PJM market monitor: data center load growth added $23B to PJM capacity market costs since 2025; capacity shortfall widening from 208.7 MW (2026/2027) to 6.5 GW (2027/2028), with 3,300+ projects pending in Pennsylvania interconnection queue alone (Data Center Knowledge, June 3, 2026)
5. Florida Governor DeSantis signed SB 484 prohibiting utilities from passing data center grid upgrade costs to residential customers — new data centers in Florida must self-fund grid buildout (FERC intel, May 2026)
6. IEA: global data center electricity consumption could double to 945 TWh by 2030, exceeding Japan's total national consumption (Reuters, June 3, 2026)
7. Gulf Development $4.3B Thailand expansion targeting 2,000MW with Microsoft, Google, SingTel as anchor partners — capital explicitly migrating toward power availability (Financial Post, June 4, 2026)
8. HIVE Digital (Canada/Nordic clean-power): +98.6% over 3 months. Hut 8 (Canada): +148.8% over 3 months. Equity markets are already expressing this premium for clean-power-anchored operators (equities data, June 5, 2026)

Implied Action: Long neocloud operators and data center REITs with concentrated Canadian or Nordic clean-power footprints over Virginia/Northern Illinois-heavy operators. HIVE Digital (+98.6%, 3m) and Hut 8 (+148.8%, 3m) are the clearest expressions of this trade already in motion; Applied Digital (APLD, buy-rated, analyst target $66.77 vs. current $44.15, ~51% upside) with Northern U.S./Canada exposure offers the best risk-adjusted entry with significant target gap still to close. IREN (buy-rated, $61.86) has clean-power-first positioning. Structurally, long infrastructure plays with grid-bypass capability (nuclear SMR developers, direct hydro access, onsite generation) over operators dependent on public grid interconnection queues. This is a multi-year positioning, not tactical — the JLL 100GW build-out forecast through 2030 means compute will be built; the question is geography, and geography is now priced by power access.

H2 — Montreal/Stockholm SPOT Recovery Propagates Through Clean-Power Corridors Confidence: 4 / 5

Thesis: The Montreal and Stockholm H100 SPOT recoveries are not isolated — they represent the leading edge of a sequential tightening wave that will propagate through other clean-power or recovering markets over a 4–10 week window. The sequence is empirically visible: London had its recovery cycle earlier (troughed in late February at $2.08/hr, recovered to $4.75/hr by May 12) and is now softening at -13.5% over 7 days — it is post-cycle, not mid-cycle. Montreal is in the heart of its recovery. Stockholm is approximately 3 weeks behind Montreal (trough May 8–12 vs. Montreal's May 11, but earlier spread_pct explosion). This phase-shifted sequence is precisely what a propagation thesis would predict: clean-power markets recover in order of how tightly they were oversupplied, not simultaneously. The next candidates in the propagation chain are Tokyo (H100 SPOT +7.2% over 30 days, spot discount tightening from ~65% toward 57%) and Ohio (+44.4% over 30 days, multi-provider, still carrying a 60.9% spot discount — implying significant upside before the market is fully tight). The propagation does not extend to Dublin or Frankfurt, where grid constraints make supply responses impossible on relevant timescales.

Key Evidence:

1. Montreal H100 SPOT: 90-day history shows plateau → derating → sharp recovery with an uninterrupted 25-day rally. Spread_pct 140% with upper provider at $6.02/hr — that provider is accelerating markups, not stabilizing (live ticker history, June 5, 2026)
2. Stockholm H100 SPOT: independently reached trough May 8–12, recovery 3 weeks behind Montreal — consistent with propagation model, not simultaneous drivers (live ticker history, June 5, 2026)
3. London H100 SPOT: peaked $4.75/hr on May 12, now $4.12/hr (-13.5% in 7 days), spread_pct declining 165% → 124% — post-cycle softening confirms London already completed its recovery wave, sequencing consistent (live ticker history, June 5, 2026)
4. Ohio H100 SPOT: +44.4% over 30 days, $2.26/hr, multi-provider market, spot discount still wide at 60.9% — in early-to-mid recovery phase; fastest-moving US market (live pricing data, June 5, 2026)
5. Tokyo H100 SPOT: +7.2% over 30 days, discount beginning to compress toward 57% — early signal, not yet confirmed tightening (live pricing data, June 5, 2026)
6. Montreal ON_DEMAND: $7.68/hr, 0.0% delta over 30 days — providers holding list price firm while spot compresses toward it; this is the canonical capacity tightening mechanism (live pricing data, June 5, 2026)

Implied Action: Monitor London H100 SPOT for stabilization above $4.00/hr as a signal that the mid-market has found a floor. Watch for Tokyo H100 SPOT spot_discount to compress below 50% as the trigger for APAC tightening. Ohio spread_pct expansion above 60% would confirm US multi-provider competition for capacity. For tactical positioning, the Montreal momentum trade has an explicit exit signal: if spread_pct re-compresses below 100%, it signals one provider is adding capacity and the near-term recovery is complete. Propagation confirmation would be incrementally bullish for IREN and Applied Digital (primarily US/Canada exposed neoclouds) before the more structural greenfield premium trade (H1) fully matures.

H3 — CoreWeave Valuation Is Fragile to Competitive Intensity and Pricing Normalization Confidence: 4 / 5

Thesis: CoreWeave (CRWV) carries a 14.7x EV/Revenue multiple and $35.1B in total debt on a business that burned $4.7B in free cash flow in Q1 2026 alone, with a capex-to-revenue ratio that worsened from 2:1 in 2025 to 3.7:1 in Q1 2026. Revenue growth of +111% YoY to $5.13B is real and the $100B backlog (including $21B Meta commitment and Anthropic anchor) provides visibility — but the competitive structure surrounding CRWV is deteriorating faster than consensus models. Google and Blackstone announced a $5B joint venture (May 25) targeting 500MW of AI compute-as-a-service capacity within one year, explicitly competing with CoreWeave using Google's own TPUs at-cost. NVIDIA invested $2B directly in Nebius (NBIS) — a competing neocloud with $9.3B cash and zero debt, now deploying Vera Rubin NVL72 in H2 2026 — creating a direct, NVIDIA-endorsed alternative to CRWV in the premium compute tier. Meanwhile, Anthropic is diversifying compute sourcing (to SpaceX Colossus, Google, AWS Trainium with up to 5GW commitment), reducing customer concentration protection. CoreWeave's June 3 Vera Rubin NVL72 first-validation announcement (with proprietary "Valvey" liquid cooling and "Racky" rack control platforms) is a genuine strategic achievement and validates the cooling moat thesis — but the 14% stock surge it generated partially prices in a forward that still requires capital markets access at current rates for multiple years. With four consecutive EPS misses (including -16.3% in May 2026, -30% in February 2026) and an Aug 12 next earnings date where the estimate is -$1.24 EPS against deteriorating cost ratios, the risk/reward of the current $108 price is asymmetric.

Key Evidence:

1. CRWV Q1 2026: revenue $2.1B, capex $7.7B (3.7:1 ratio), FCF burn -$4.7B in a single quarter, total debt $35.1B (equity fundamentals, June 5, 2026)
2. Four consecutive EPS misses: May 2026 -16.3%, Feb 2026 -30%, Aug 2025 -30%, May 2025 -422%. Only Nov 2025 was a beat (+76.6%) (earnings history, June 5, 2026)
3. Google + Blackstone $5B JV announced May 25, targeting 500MW AI compute within one year using Google's own TPUs — direct competitive displacement of CRWV's core market (news feed, May 25, 2026)
4. NVIDIA $2B direct investment in Nebius (NBIS): Nebius has $9.3B cash, zero debt, deploying Vera Rubin NVL72 H2 2026; 3-month equity return +171% vs CRWV -15% — market is already expressing a quality preference within the same neocloud theme (equities data, June 5, 2026)
5. Amazon deployed 2.1M AI chips in 12 months; plans to deploy 1M NVIDIA GPUs beginning 2026 — AWS building competing capacity at scale at hyperscaler cost structure (news feed, 2026)
6. H100 SPOT in Dublin (-49.9%) and Frankfurt (-26.8%) over 30 days — CRWV's European customer markets are in pricing compression, diluting the realized $/GPU-hr on non-backlogged capacity (live pricing data, June 5, 2026)
7. CRWV Vera Rubin NVL72 first validation (June 3) with Valvey/Racky proprietary platforms — genuine competitive moat in next-gen infrastructure, but priced in via 14% single-day surge to ~$108 (news feed, June 3, 2026)
8. H200 on-demand listings emerging across 18+ regions at $8.60–$15.74/hr — as H200 becomes default capacity, H100 pricing power will compress on contract renewals (live pricing data, June 5, 2026)

Implied Action: Cautious neutral-to-short on CRWV at current levels ($108), with Aug 12 earnings as the primary catalyst. The bear case: another EPS miss on the back of Q1's 3.7:1 capex/revenue ratio, combined with any signal of customer concentration erosion or margin compression. The bull case (Vera Rubin first-mover, $100B backlog) is real but partially priced after the June 3 surge. The cleaner expression of neocloud exposure is Long NBIS / cautious on CRWV — NBIS carries the same Vera Rubin upside with $9.3B cash, no debt, direct NVIDIA endorsement, and a PEG of 0.63 (vs. CRWV's unattractive valuation framework). Watch for analyst target upgrades on NBIS in the next 2–4 weeks as models update for the NVIDIA investment; analyst mean target is currently $238 vs. $259.67 current price, suggesting models are lagging the investment thesis. For investors already long CRWV: hedge via puts or trim into the Montreal/Stockholm SPOT momentum wave if it generates a bullish neocloud coverage event before August 12.

H4 — Blackwell Cooling Infrastructure Is a Proprietary Operational Moat Confidence: 4 / 5

Thesis: Cooling infrastructure is transitioning from a data center operating cost into a performance variable that directly determines tokens-per-GPU-hour yield at Blackwell densities. The physics are not marginal: at 1,000W+ TDP per GPU in the Blackwell/Vera Rubin architecture, direct liquid cooling is not optional — it is the prerequisite for sustained rated performance. Frore's LiquidJet Nexus delivers 10% throughput improvement on Blackwell GPUs versus default cooling solutions; at $10/hr H200-class pricing, that 10% yield gain translates to $1/hr per GPU in effective revenue — $87.6M/year at 10,000-GPU scale without adding a single chip. NVIDIA's explicit alignment to 45°C warm-water single-phase DLC as next-gen platform standard locks this dynamic in structurally. CoreWeave's June 3 announcement of "Valvey" (programmable liquid cooling management) and "Racky" (unified rack control), developed specifically for the Vera Rubin NVL72 72-GPU rack, confirms that at least one neocloud operator has internalized cooling as a first-party competitive asset rather than a third-party procurement decision. The operators who deploy Blackwell at superior cooling standards will extract more compute per dollar spent than competitors — creating a widening cost-per-token gap that will become visible in utilization disclosures and token pricing over the next 12–18 months.

Key Evidence:

1. Frore LiquidJet Nexus: 10% GPU throughput improvement on Blackwell vs. default cooling — at $10/hr H200 pricing, implies $1/hr per GPU effective revenue premium (~$87.6M/year at 10,000 GPUs) (news feed, June 2026)
2. CoolIT Systems 15kW coldplate: 4x performance improvement over its March 2025 model; used in 7 of the top 10 supercomputers; ranked #1 direct-to-chip cooling 2026 — technology validated at hyperscale but not yet commoditized (news feed, June 2026)
3. NVIDIA platform alignment: 45°C warm-water single-phase DLC as next-gen platform standard — structurally mandates DLC for rated performance at Blackwell/Vera Rubin density (news feed, June 2026)
4. CoreWeave Valvey + Racky platforms (June 3): first-party programmable liquid cooling management deployed at Vera Rubin NVL72 launch — confirms cooling moat is being built by operators, not just hardware vendors (news feed, June 3, 2026)
5. Marvell Teralynx T100 (102.4 Tbps, 3nm, 25% lower power) shipping this quarter — advanced cooling + advanced networking fabric creates a compound performance advantage for early deployers (news feed, June 2026)
6. H200 on-demand across 18+ regions at $8.60–$15.74/hr ($12.94/hr London, $12.57/hr Tokyo): as H200/Blackwell becomes default capacity, performance differentiation at the cooling layer becomes the marginal competitive variable (live pricing data, June 5, 2026)

Implied Action: Long liquid-cooling infrastructure suppliers as the purest expression of this theme without the balance sheet risk: Vertiv (VRT) and Modine Manufacturing (MOD) are the primary publicly-traded DLC exposure names (VRT +22%, MOD +18% over 3 months — lagging the thesis, still room to re-rate). CoolIT Systems is private. For neocloud exposure to this theme: NBIS over CRWV — Nebius has the balance sheet quality to invest in premium cooling infrastructure at scale; CRWV has the first-mover advantage but the leverage to service it. The highest-conviction expression is hyperscalers over neoclouds on a 3-year view: Google, Microsoft, and AWS have the infrastructure budgets to deploy premium cooling at scale across their entire fleets; neoclouds will face capital allocation constraints. The trigger to watch: NVIDIA publishing mandatory DLC specifications for Vera Rubin/GB200 deployment — this would immediately create a compliance timeline for all operators and re-price facilities without DLC infrastructure downward.

H5 — AWS Inferentia APAC Pool Depletion Is a Geographically Isolated Signal, Not a Structural ASIC Tightening Confidence: 3 / 5

Thesis: The Tokyo Inferentia SPOT price rising +772% over 30 days ($0.33/hr, spot discount compressed to 45.8%) and Mumbai +881% ($0.10/hr, discount at 76.1%) are genuine pool depletion signals — but they are regional and specific, not a harbinger of global Inferentia2 reserved instance saturation as initially hypothesized. The critical complication: Ohio Inferentia SPOT is down -38.5% over 7 days (discount 87.5%), London Inferentia SPOT is -2.5% over 7 days (88.4% discount), Singapore is -8% over 7 days (62.4% discount), and Hong Kong is -93.2% over 30 days. If this were a global Inf2 pool being swept into reserved instances, compression would appear across regions simultaneously. Instead, it is a thin-pool AWS spot pricing engine reacting to what is likely a small number of specific customers converting or reserving capacity in two APAC markets. The structurally high-confidence signal here is Trainium obsolescence, not Inferentia tightening. Oregon Trainium SPOT at -70% over 30 days ($0.060/hr, 82.9% discount), Virginia Trainium at -35.8%, and Ohio Trainium collapsing from ~$0.49 in late February to ~$0.025 by early May is an unambiguous first-gen ASIC depreciation event as Trainium2 capacity scales. The broader directional inference: AWS's custom silicon production inference stack is increasingly served from Inferentia2 reserved capacity and diversified across Trainium2, Graviton-4, and Anthropic's committed 5GW Trainium arrangement — making the Inferentia1 spot pool increasingly irrelevant as a market signal.

Key Evidence:

1. Tokyo Inferentia SPOT: +772.6% over 30 days, $0.33/hr, spot discount 45.8% — down from ~88% at 30-day baseline, genuine thin-pool compression (live pricing data, June 5, 2026)
2. Mumbai Inferentia SPOT: +881.3% over 30 days, $0.10/hr, from near-zero baseline of $0.010/hr (live pricing data, June 5, 2026)
3. Contradiction: Ohio Inferentia SPOT -38.5% over 7 days (discount 87.5%), London -2.5% (88.4%), Hong Kong -93.2% over 30 days — global Inf2 tightening thesis is not supported by non-APAC markets (live pricing data, June 5, 2026)
4. Oregon Trainium SPOT: -70% over 30 days, $0.060/hr, 82.9% spot discount — from ~$0.49 in late February; Trainium1 obsolescence fully confirmed (live pricing data, June 5, 2026)
5. Anthropic committed to AWS Trainium infrastructure up to 5GW scale — but simultaneously diversifying across SpaceX Colossus, Google, and CoreWeave (news feed, June 2026)

Implied Action: Do not trade on the Tokyo/Mumbai Inferentia SPOT spike directly — the absolute price levels ($0.33/hr, $0.10/hr) are too low for meaningful exposure, and the thin-pool mechanism makes these prices unreliable for 30+ day duration. The actionable signal from this data is the Trainium obsolescence confirmation: avoid any entity with significant unhedged first-gen Trainium reserved instance commitments above current spot pricing. Monitor Seoul and Oregon Inferentia SPOT as the confirming catalysts — if those two markets begin showing discount compression toward 70% or below, the thesis upgrades to structural and becomes actionable. Until then, this is a watch-and-confirm signal, not a trade.

Risk Flags

Immediate (0–30 days)

R1 — Montreal SPOT Reversion Risk: Single-Buyer Absorption The Montreal H100 SPOT market has only two providers and spread_pct at 140% — the upper provider is at $6.02/hr and the lower at $2.51/hr. If the upper-provider trades are driven by a single large buyer exhausting their spot budget, the bid disappears and the market re-prices sharply to the lower provider's floor. The denial condition for H1 is a retracement below $3.50/hr within 10 days — monitor the spread_pct as the primary warning signal. If spread_pct begins compressing back below 100%, capacity is being added or the single large buyer has exited.

R2 — London H100 SPOT Post-Cycle Softening Accelerates London's -13.5% over 7 days, with spread_pct declining from 165% to 124%, may indicate the H100 spot cycle has already peaked in the main European grid-adjacent market. If London falls below $3.50/hr in the next two weeks while Dublin and Frankfurt remain soft, it would suggest the propagation thesis (H2) is narrower than modeled — confined to pure clean-power markets rather than extending to mixed EU markets.

R3 — AWS Spot Pricing Engine Artifacts in Thin ASIC Markets The Inferentia Tokyo (+772%) and V100 Tokyo (+168%) moves share a structural characteristic: both are ultra-thin single-provider markets where a handful of capacity units can create extreme price dislocations. Positions built on these moves — either directionally or as hedges — carry pricing-engine noise risk. Any retracement within 72 hours should be treated as the null hypothesis (pool artifact) rather than a signal reversal.

Near-Term (30–90 days)

R4 — CoreWeave August 12 Earnings: Capex/Revenue Ratio Is the Key Variable If Q2 2026 capex continues at the Q1 pace (~$7.7B), the quarterly free cash flow burn will approach -$5B for the second consecutive quarter against $2.1–2.5B revenue. The consensus EPS estimate of -$1.24 already prices in losses, but a fourth consecutive miss on cost discipline — particularly if gross margin shows any compression from competitive pricing pressure — could re-test the May 2026 low. The competitive dynamic (Google/Blackstone JV, Nebius NVL72 deployment) will be a visible narrative by August 12. Watch for any forward guidance on customer concentration changes or pricing concessions to anchor tenants.

R5 — H200 SPOT Market Emergence Could Immediately Re-Baseline H100 Pricing H200 on-demand listings are present across 18+ regions ($8.60–$15.74/hr range) but no H200 SPOT market exists yet. The moment AWS, Azure, or GCP initializes H200 SPOT pricing — likely in the next 60–90 days as H200 supply scales — H100 SPOT will immediately be benchmarked against H200 compute-per-dollar. At current H100 SPOT levels of $2.26–$4.26/hr vs. H200 on-demand of $8.60–$15.74/hr, the H200/H100 performance-per-dollar comparison is the binding pricing ceiling for H100. Any acceleration of H200 SPOT availability should be treated as a H100 pricing headwind.

R6 — AMD MI25 Coordinated EU Decommission Has Contagion Risk to A100/V100 Generation The simultaneous -33% to -40% MI25 SPOT repricing across Dublin, Italy, and Stockholm over 7 days suggests a coordinated Oracle cloud decommission of older AMD GPU inventory. This is a depreciation event, not an anomaly — but it sets a precedent for the pace of Gen-3/Gen-4 GPU retirement timelines. If Oracle (or Azure) extends this decommission logic to A100 80GB or V100 32GB inventory in the same quarter, neocloud operators with unhedged legacy GPU exposure at above-market carrying costs will face mark-to-market losses. Watch for A100 SPOT in EU regions to confirm or deny.

Structural (6–24 months)

R7 — Power Grid Regulatory Hardening Becomes a Sustained Drag on US Neocloud CapEx Plans The convergence of NERC's "emerging large loads" regulatory category, 14 states considering data center legislation, Florida's SB 484, PJM's 6.5 GW shortfall by 2027/2028, and 3,300+ projects in Pennsylvania's interconnection queue is not a single event — it is a regulatory regime shift. The 14-year Northern Virginia interconnection timeline is the extreme case, but even 3–5 year timelines break the economics of new US East Coast GPU cluster development. Operators with permits already filed and approved (or Canadian/Nordic jurisdiction) have a structural execution advantage; operators still in queue face both cost escalation (PJM capacity cost additions) and timeline risk. The X-Energy (XE) 10-Q filing documenting non-U.S. graphite supply chain risks for TRISO-X nuclear fuel introduces supply-side risk to the nuclear-power-for-data-centers thesis: the regulatory arbitrage that nuclear was meant to solve may face its own bottlenecks at the fuel supply level.

R8 — Blackwell Transition Creates a Bifurcated Obsolescence Curve for H100-Era Operators The H100 depreciation cycle has approximately 4.2 years of elapsed lifecycle. As H200 SPOT markets emerge and Vera Rubin NVL72 capacity comes online through H2 2026 (CRWV first, Nebius following), the compute-per-dollar premium that H100 commands will compress. Operators with long H100 reserved instance contracts or CapEx-heavy balance sheets built around H100 utilization assumptions will face a structural revenue headwind as enterprise buyers benchmark against Blackwell-class alternatives. This is a 12–18 month transition risk, not immediate — but it is the mechanism underlying the CRWV valuation fragility thesis and will manifest first in contract renewal pricing for non-anchor customers.