Daily Investment Brief — June 8, 2026

GPU & Cloud Compute Intelligence Brief

June 8, 2026 | Cross-provider live pricing: AWS, Azure, GCP, Oracle

Market Pulse

The dominant story in GPU cloud markets today is a confirmed V-shaped recovery in H100 SPOT pricing after a tariff-shock demand collapse that bottomed on May 11 at $1.90/hr in Montreal — a floor that held for exactly one day before a clean, six-week recovery brought prices back to $4.89/hr (+157.6% over 30 days). This is not a uniform global bounce: the North Atlantic corridor (Montreal, Stockholm +66.9%, Ohio +37.3%) is leading decisively while South Asia (Mumbai -28.1%) and Southeast Asia (Singapore -18.4%) are diverging, pointing to a regional demand resequencing rather than a monolithic supply shock reversal. Simultaneously, the market is bifurcating structurally along a training/inference axis — training-class H100/H200 is re-tightening while inference-class L4 pricing is perfectly flat globally across all 10+ regions, the signature of a catalog-priced commodity with no remaining pricing power. The binding constraint through the forecast horizon is shifting away from silicon availability entirely: Northern Virginia grid interconnection is now quoted at 14-year lead times, 14 US states are considering data center construction bans, and NERC has issued an explicit reliability guideline for large-load interconnection risk — the power grid is now the rate-limiting input to AI infrastructure expansion, not chip supply chains.

Key Movers

Thin-market noise: The V100 32GB Tokyo SPOT (+243% 30d, single provider) reflects idiosyncratic APAC Volta demand — low liquidity, low signal. The GB200 Seoul SPOT_DEV flat line is a pre-release catalog entry with zero actual inventory; disregard for price discovery purposes.

Investable Insights

H1 — H100 SPOT Recovery: The Hopper Cycle Has a Second Act Confidence: 5 / 5

Thesis: The April 9 tariff shock broke a three-month demand equilibrium that had kept H100 SPOT Montreal stable at ~$5.00/hr since January. The subsequent collapse — a near-perfectly linear 32-day descent to $1.90/hr on May 11 — was demand destruction, not fundamental oversupply: the H100 catalog has remained active on AWS, Azure, and GCP throughout, and depreciation data confirms all three providers are maintaining H100 presence 4+ years post-launch, consistent with the 5–7-year historical GPU catalog lifecycle. The V-bottom was genuine because the underlying demand for Hopper-generation training was not destroyed — it was deferred by uncertainty. As hyperscaler capex restarted (the four majors collectively plan ~$725B in 2026 capital expenditure, up 77% YoY), deferred workloads returned to spot markets, clearing the inventory overhang in roughly six weeks. Today's $4.89/hr in Montreal is approaching the pre-tariff equilibrium of ~$5.00/hr, but the spot discount to on-demand has only compressed from 73% to 35.6% — well above the historical steady-state of 65–67% discount. Structural mean-reversion of that spread implies either spot prices need to rise further toward $6–7/hr, or on-demand prices fall to meet them. Neither scenario is bad for current spot buyers. Critically, the recovery is geographically validated across the North Atlantic corridor (Stockholm +66.9%, Ohio +37.3%, Madrid +24.4%) while APAC diverges (Mumbai -28.1%, Singapore -18.4%) — a multi-regional signal with enough independent markets to eliminate noise as a driver.

The H200 pricing architecture reinforces this view. In three US markets, H200 ON-DEMAND is actually cheaper than H100 ON-DEMAND: Ohio H200 at $6.22/hr versus H100 at $7.73/hr, California H200 at $7.20 versus H100 at $10.08, Iowa H200 at $7.06 versus H100 at $8.11. These are single-provider anomalies being corrected in real time — California just repriced upward +18.9% in 7 days. When these anomalies normalize toward the international H200 range of $9–16/hr, H100's relative value proposition at $4–5/hr SPOT sharpens further: you're buying 80GB HBM2e with proven software ecosystems at a 60–85% discount to on-demand, for a chip that has 1–3 more years of active catalog life.

Key Evidence:

1. Montreal H100 SPOT 90-day history: near-perfect V-bottom at $1.90/hr on May 11; six-week monotonic recovery to $4.89/hr today (+157.6% 30d, live ticker Jun 8, 2026)
2. Pre-tariff equilibrium: Montreal H100 SPOT held $4.97–5.01/hr for the entire January 8–April 8 window, confirming the recovery is approaching a known fundamental price level, not overshooting
3. Cross-regional validation: Stockholm +66.9%, Ohio +37.3%, Madrid +24.4% all positive over 30 days, ruling out single-market noise (live tickers, Jun 8, 2026)
4. Spot discount compression: 73% at the May 11 bottom → 35.6% today, still well above historical equilibrium — recovery has further to run
5. H200/H100 inversion in Ohio ($6.22 vs $7.73) and California ($7.20 vs $10.08) confirms H100 OD pricing is sticky and will not fall materially; creates floor for SPOT relative value
6. H100 depreciation data: AWS, Azure, GCP all classified is_active: true at 4.2 years post-launch; catalog survival consistent with 5–7-year industry norm
7. Counterpoint: Mumbai SPOT -28.1% and Singapore -18.4% over 30 days confirm APAC demand cycle is out of phase — not all regions are recovering

Implied Action:

• Buy H100 SPOT for 3–6 month training runs NOW. The current $2.27–4.89/hr range across US regions (Oregon to Montreal) represents peak value relative to recovery trajectory. The spot discount to OD is compressing, so the window of deep discount narrows. Execute before spot discount falls below 25%, which historically has been the entry point for on-demand switching.
• Monitor Montreal for deceleration. Recovery will likely slow as price approaches the $5.00/hr pre-tariff floor. If Montreal hits $5.10–5.20/hr without a corresponding OD price increase, consider rotating from spot to on-demand for continuity — the premium at that point is just 10–15%.
• NVDA equity read-through: NVDA at $205.10 (forward P/E 16.2x, analyst target $298) with H100 catalog still monetizing at 4+ years of age and spot prices recovering is a positive fundamental signal. The H100's continued catalog viability is a near-term revenue defense that short-sellers are underweighting.
• Avoid: Locking long-term H100 reserved capacity at current OD rates ($7.59–10.08/hr). The H200 anomaly correcting upward will not push H100 OD higher; instead it will shift enterprise buyers to H200 reserved, creating an OD H100 overhang.

H2 — AWS ASIC Monetization: Real but Geographically Constrained Confidence: 3 / 5

Thesis: The 280–480% Inferentia/Trainium SPOT price surge over 30 days — Tokyo Inferentia from ~$0.053 to $0.310/hr, Mumbai Inferentia +438%, Ohio Trainium +280% — initially appeared to be AWS deliberately tightening proprietary silicon availability as Jassy publicly claims $20B annualized revenue from custom chips (with an explicit $50B aspirational target for a hypothetical standalone AWS silicon business). The thesis was compelling: AWS uses Trainium/Inferentia spot as a customer-acquisition loss-leader, then ratchets pricing once enterprises build production pipelines on it. However, the full regional picture breaks the "deliberate global ratchet" framing. Hong Kong Inferentia SPOT is down -93.5% over the same 30-day window. London sits at $0.047/hr and Paris at $0.038/hr — unchanged, 87–89% below the Tokyo rate. Brazil is flat at $0.054/hr. A coherent global monetization strategy does not produce this pattern. The more accurate interpretation: AWS is reallocating APAC Inferentia spot capacity into committed enterprise contracts following regional AI announcements in Japan and India, while European and Latin American spot capacity remains plentiful and cheap. The directionality of AWS silicon rent extraction is correct — the mechanism is regional committed-contract growth, not global spot market tightening.

This distinction matters enormously for the actionable call. The risk is concentrated in APAC inference workloads on Inferentia, not globally. And the Jassy $20B ARR claim is independently validated by Broadcom's Q2 FY2026 AI semiconductor revenue of $10.8B (+143% YoY) — AVSO's custom chip business for Google and Meta is a proxy for the structural ASIC margin capture trend that AWS is executing in parallel.

Key Evidence:

1. Tokyo Inferentia SPOT: near-zero through early May → $0.310/hr today, +480% 30d (live ticker, Jun 8, 2026)
2. Mumbai Inferentia SPOT: +438% over 30 days (live ticker, Jun 8, 2026)
3. Ohio Trainium SPOT: +280% over 30 days to $0.082/hr (live ticker, Jun 8, 2026)
4. Directly contradicting evidence: Hong Kong Inferentia -93.5% over same window; London $0.047/hr, Paris $0.038/hr — flat or declining (live tickers, Jun 8, 2026)
5. Jassy statement: $20B ARR from Graviton/Trainium/Nitro, hypothetical $50B standalone value (NAI500, Jun 8, 2026) — confirms intent but not mechanism
6. Broadcom Q2 FY2026: AI semiconductor revenue $10.8B, +143% YoY (AOL, Jun 7, 2026) — corroborates hyperscaler ASIC investment cycle broadly
7. Inferentia gen-1 catalog_survival: launched December 2019, is_active: true at 6.5 years — longest active catalog tenure in the dataset; not a product being wound down

Implied Action:

• Immediate hedge for APAC inference on Inferentia SPOT: Route APAC inference traffic to European Inferentia spot nodes (London at $0.047/hr, Paris at $0.038/hr) before AWS normalizes European pricing. This cross-regional arbitrage window is approximately 60–90 days before AWS notices and rebalances.
• European and LatAm Inferentia SPOT remains a buy: These markets are 87–89% cheaper than Tokyo right now. For workloads without APAC data-residency requirements, this is a structural discount that's real and current.
• Long AVGO (currently $385, strong-buy consensus, analyst target $513): Broadcom is the cleanest equity expression of hyperscaler custom silicon spending — it builds ASICs for Google and Meta, has a 0.88 PEG ratio implying growth is underpriced, and carries none of the customer-internalization risk that clouds NVDA's multiple. The $10.8B Q2 AI revenue run rate will compound.
• Risk for inference neocloud operators on AWS ASIC spot: If your margin depends on arbitraging cheap Inferentia spot in Tokyo or Mumbai, that spread has materially closed. Migrate to L4 SPOT (Oregon $0.365/hr, no monopoly risk, globally available) or lock Inferentia 1-year reserved rates immediately.

H3 — H200 US Anomaly: The Normalization Wave Has Started and the Window Is Weeks, Not Months Confidence: 4 / 5

Thesis: Three US markets — Ohio, Iowa, and California — are pricing H200 ON-DEMAND below H100 ON-DEMAND, an inversion that cannot persist in a rational market. Ohio H200 at $6.22/hr against H100 at $7.73/hr, California at $7.20 versus $10.08, Iowa at $7.06 versus $8.11 — these are single-provider markets where a provider either seeded capacity with promotional pricing or where the residual-decomposition model is capturing a genuinely distorted bundle price. Either way, the economic reality is: buyers can access H200's 141GB HBM3 (vs H100's 80GB) and 4,800 GB/s memory bandwidth (vs 3,350) at a discount to H100 in the same region. That is an explicit compute-per-dollar arbitrage. The correction mechanism is already active. California H200 repriced +18.9% in just 7 days, from $6.06/hr to $7.20/hr — the normalization wave is propagating. Ohio's perfectly flat 90-day history at $6.22/hr is a lagging indicator; expect the same re-pricing impulse to arrive within 30–45 days based on the California lead. The international H200 OD range ($9.23/hr Sydney, $12.57/hr Seoul, $14.35/hr Oslo, $15.70/hr Osaka) establishes the target — Ohio and Iowa prices are 50–60% below international norms with no fundamental cost-of-infrastructure justification for the gap.

The 90-day Ohio history confirms the structural nature: the price has been stable at a single provider's offering since January 8 with zero spread, zero delta. This is a provider that posted a price and has not touched it in 5+ months. The California repricing event is the first crack in the dam — and it happened while Jensen Huang was in Seoul announcing gigawatt-scale partnerships, which reinforces the thesis that Blackwell-era GPU pricing is being actively reset globally.

Key Evidence:

1. H200 OD Ohio: $6.22/hr, single provider, flat since Jan 8 — compared to H100 OD Ohio $7.73/hr; H200 is 19.5% cheaper for 76% more VRAM (live tickers, Jun 8, 2026)
2. H200 OD California: $7.20/hr, +18.9% repricing event in past 7 days (from $6.06/hr) — confirms normalization is active (live ticker, Jun 8, 2026)
3. International H200 OD range: Sydney $9.23, Seoul $12.57, Dubai $14.42, Oslo $14.35, Osaka $15.70 — US anomaly regions are 50–60% below international floor (live tickers, Jun 8, 2026)
4. H200 Virginia OD $12.61/hr — even within the US, Virginia prices at 2x the Ohio level, confirming Ohio/Iowa/California are outliers, not the norm
5. Ohio 90-day history: zero spread, zero delta, single provider from January 8 — structural single-provider price-setting, not dynamic market competition
6. Counterpoint: No second provider has been observed entering these markets yet; the inversion could theoretically persist if the single provider never adjusts. Zero-spread, zero-delta markets have no competitive forcing function until a second entrant appears.

Implied Action:

• Lock H200 3-year reserved capacity in Ohio and Iowa this week. The implied price at current Ohio OD ($6.22/hr) is structurally mispriced against a $10–15/hr international norm. The California repricing event establishes that the correction mechanism is active. A 3-year reserved commitment at even a 20% discount to $6.22/hr (≈$5.00/hr) against a corrected $10–12/hr benchmark is a 50%+ long-duration arbitrage.
• Enterprise buyers with Ohio data-residency requirements: This is a 30-day window. Execute H200 reserved pricing conversations with the current provider before a second entrant arrives and forces the provider to reprice on-demand upward.
• Watch for Ohio repricing as confirmation signal. When Ohio H200 OD moves above $7.73/hr (current H100 OD price there), the inversion closes and the arbitrage window shuts. Set a price alert on the H200|GPU|ON_DEMAND|us-ohio ticker.
• Caution for colocation operators in Ohio: The H200 underpricing may indicate a hyperscaler using on-demand pricing as a loss-leader to win committed agreements, which does not translate into profitable rack-hour revenue for independent operators trying to match that price.

H4 — L4 Oregon New Entrant: Inference Market Formation, Not Price War (Yet) Confidence: 3 / 5

Thesis: The L4 ON-DEMAND Oregon 24-hour apparent +92% jump — from $0.295/hr to $0.567/hr — is a new-entrant arrival event: a second provider listed L4 capacity in US-Oregon, replacing a single-provider artifact price with a more representative two-provider market. The structural significance is this: across all 10 L4 ON-DEMAND regions in the dataset, every single market shows zero spread, zero delta — perfectly stable catalog pricing at $0.59–1.04/hr depending on geography. L4 is not being dynamically competed; it is being sold as a catalog service by Google Cloud and a narrow set of partners at administered prices. The Oregon event creates the first observable two-provider competition in a US-West inference market, but should not be extrapolated into an imminent price war. The real commoditization pressure on L4 pricing will come from two sources that are not yet visible in live spot data: (1) AMD Instinct MI300X-class chips reaching public cloud spot markets (no tickers present today), which would create a three-tier inference stack at L4 < AMD-class < H100; and (2) GB200-derivative inference SKUs reaching spot markets in 2027, at which point L4 gets repriced to its residual value. The Oregon entry is the leading edge of market formation in US-West inference — it signals the region is becoming a competitive inference hub, which over 12–18 months will exert downward pressure on current catalog prices.

The L4 1-year reserved market is already active at $0.43–0.58/hr (a 25–35% discount to on-demand), which is the pricing architecture of a mature market, not an emerging one. The commodity is already here. The price compression is coming.

Key Evidence:

1. L4 OD Oregon: $0.567/hr today vs $0.295/hr prior single-provider reading — new entrant event confirmed (live ticker, Jun 8, 2026)
2. L4 SPOT Oregon: $0.365/hr, paired new-entrant listing (live ticker, Jun 8, 2026)
3. Global L4 OD pricing: 10 regions, zero spread, zero delta across all — Dubai $0.711, Sydney $0.764, Montreal $0.658, Frankfurt $0.729; catalog-administered, not dynamically priced (live tickers, Jun 8, 2026)
4. L4 1-year reserved: $0.43–0.58/hr, 25–35% discount to OD — mature market structure already established
5. AMD: Meta and OpenAI each committing 6GW of AMD Instinct deployments (news feed, Jun 2026) — MI300X scale deployments underway but no cloud spot tickers yet
6. Counterpoint: Zero AMD Instinct tickers in current dataset — the three-tier inference market formation is a 2027+ event, not actionable today
7. Nebius (NBIS) down -12.3% today, market cap ~$57.8B despite $44–50B contracted backlog — market is already pricing inference neocloud margin compression risk (equity data, Jun 8, 2026)

Implied Action:

• Lock L4 1-year reserved in Ohio at $0.479/hr. The 16% discount to OD ($0.569/hr) provides a defensible floor for 12 months before Blackwell-generation inference pressure arrives. Oregon's new-entrant event confirms US-West inference competition is developing, but the Ohio 1-year reserved is a tighter, more liquid market with established volume.
• Structural short bias on inference neocloud unit economics over 12–18 months. Companies selling inference compute with a persistent $0.40–0.60/hr margin over L4 spot are sitting on a spread that will compress as AMD MI300X reaches cloud spot markets and GB200-derivative inference SKUs appear. Evaluate NBIS and similar names on the basis of whether their contracted backlog at fixed prices is realistic against falling market spot rates.
• Monitor for AMD Instinct tickers. The appearance of MI300X or MI400X cloud spot tickers is the single clearest signal that three-tier inference market formation has begun. When that ticker appears, L4 pricing enters a structural compression cycle.
• Do not extrapolate the Oregon L4 event into an immediate price war. The +92% apparent move reflects a new market establishing at a higher, more accurate price — not an incumbent being undercut.

H5 — Korea AI Infrastructure: Sovereign-Grade Buildout With a Powered, Permitted Runway Confidence: 4 / 5

Thesis: South Korea is executing the fastest-maturing sovereign AI compute buildout outside the US hyperscaler corridor, and the market pricing does not yet reflect the demand density forming there. H100 SPOT Seoul at $2.62/hr today is the cheapest among mature APAC spot markets — but the 90-day history tells a more structurally interesting story. Seoul underwent a three-provider market phase from mid-January through March (with a distorting ultra-cheap third entrant generating 160–330% spreads), then consolidated to a cleaner two-provider market in April. Today's 9.3% spread (down from 44.7% thirty days ago) and +13.2% 30-day price recovery are the signature of a market discovering its floor before the demand wave hits. Meanwhile, H200 ON-DEMAND Seoul is already live at $12.57/hr — not a placeholder, but actual Blackwell-era inventory being sold in the market right now, validating that real hardware allocation is flowing into Korea. The GB200 Seoul SPOT_DEV ticker at $12.08/hr remains flat since January (a catalog placeholder), but its price proximity to the live H200 Seoul OD rate suggests realistic eventual pricing, not aspirational positioning.

The macro infrastructure thesis is independently corroborated and unusually concrete: LG Uplus announced a 200MW Paju AIDC with Phase 1 at 20% construction completion targeting June 2027 — a specific, dated, substantially committed build, not a press release (Korea Times, Jun 7, 2026). The 200MW power supply is already confirmed, which is the critical differentiation: against a backdrop of 14-year Northern Virginia grid interconnection queues and 14 US states considering DC construction bans, a pre-committed 200MW connection is among the most valuable power assets in APAC. Jensen Huang's June visit to Seoul for GW-scale deals with SK Telecom, Naver, and SK Hynix on HBM4 (Let's Data Science, Jun 8, 2026) provides the demand anchor. SK Hynix's HBM4 manufacturing partnership with NVIDIA makes Korea not just a compute buyer but a sovereign AI hardware manufacturer, creating a policy-level incentive to maintain domestic cloud capacity as a strategic national asset.

Key Evidence:

1. H100 SPOT Seoul: $2.62/hr, +13.2% 30d, +5.1% 24h; spread compressed 44.7%→9.3% in 30 days — market consolidation/floor discovery (live ticker, Jun 8, 2026)
2. Seoul H100 SPOT 90-day history: 3-provider chaos Jan–Mar (160–330% spreads) → 2-provider clean consolidation Apr onward — classic market formation pre-maturation sequence
3. H200 OD Seoul: $12.57/hr, live (not placeholder), single provider, -2.5% 7d — actual Blackwell hardware allocated to Korea today (live ticker, Jun 8, 2026)
4. GB200 Seoul SPOT_DEV: flat at $12.08/hr since January — placeholder, but priced adjacent to live H200 OD, suggesting realistic positioning
5. LG Uplus Paju AIDC: 200MW Phase 1, 20% construction complete, June 2027 target, 200kW/rack density, Rubin hardware support confirmed (Korea Times, Jun 7, 2026)
6. Nvidia Seoul announcements: GW-scale partnerships with SK Telecom, Naver, SK Hynix on HBM4 (Let's Data Science, Jun 8, 2026)
7. Korea private DC market projection: KRW 11 trillion (~$8B) by 2029 (Digitimes, Jun 7, 2026)
8. Power constraint context: NOVA grid interconnection quoted at 14-year lead times; Paju 200MW committed — structural advantage quantified (Forbes, May 31, 2026)
9. Bear case: Paju Phase 1 is only 20% complete with 12 months to target; construction risk is real and the committed-power advantage disappears if the build slips past 2027

Implied Action:

• Long SK Hynix (KRX: 000660) as the cleanest expression of the Korea AI thesis. HBM4 production for NVIDIA under a contracted supply agreement into 2027+ is a direct revenue stream from the global $725B hyperscaler capex cycle. Unlike cloud spot pricing, this exposure doesn't require Korea's compute market to reach full liquidity — it monetizes the infrastructure buildout itself.
• Watch GB200 Seoul SPOT_DEV for activation. When this ticker shows a non-zero delta — meaning a provider has listed actual GB200 inventory in Seoul — it signals LG Uplus or a comparable first-mover has allocated Rubin-generation capacity. At that point, the Seoul market enters its monetization phase and the $12.08/hr floor becomes validated pricing.
• Geographic diversification for APAC data-residency compute: Korea offers H100 SPOT at $2.62/hr today vs. Japan Osaka at $9.24/hr and Singapore at $2.32/hr. For workloads with APAC residency requirements, Korea's combination of sovereign policy support, committed power, and domestic chip manufacturing makes it the structurally superior location over a 12–36 month horizon.
• Monitor construction progress on Paju Phase 1. A slip past Q3 2027 completion would signal supply-side risk and reduce the near-term Korea spot market development timeline. Quarterly LG Uplus earnings releases are the most accessible data point for this watch.

Risk Flags

Immediate (0–30 Days)

R1 — H200 US Anomaly Correction Creates a Short-Lived but Disruptive Repricing Event The California H200 repricing of +18.9% in one week is the leading edge of a normalization wave that will reach Ohio and Iowa within 30 days. Enterprise buyers who fail to lock reserved H200 capacity before that correction will face a step-change increase in their committed compute cost. The risk is not prolonged — the window is short and the direction is unambiguous — but the magnitude is significant: Ohio H200 OD could reprice from $6.22/hr to $9–12/hr in a single catalog update. Any procurement team with H200 commitments to make in the next 60 days should treat this as a live operational risk.

R2 — H100 SPOT Recovery Approaching Mean-Reversion; Rate of Gain Decelerating Montreal's recovery to $4.89/hr (vs. pre-tariff floor of ~$5.00/hr) and spot-to-OD discount at 35.6% (vs. historical 65–67%) means the mathematical upside in the current recovery is narrowing. The next 5–12% to equilibrium is structurally supported, but buyers entering SPOT positions today are buying a maturing recovery, not a fresh bottom. The risk is that recovery stalls at $5.00–5.20/hr — the old equilibrium — rather than overshooting, leaving spot buyers with a position that's now pricing at full OD value.

R3 — Oregon H100 ON-DEMAND Two-Provider Market is Unstable The April 16 dropout of the cheapest H100 OD provider in Oregon (which was pricing at an anomalously low ~$0.44/hr implied — likely a bundle artifact) pushed the market median from ~$5.96 to $8.37/hr and left an 80%+ spread between the two remaining providers. With only two providers and an 80%+ spread, this market is a single provider exit away from further disruption. If the more expensive provider reprices or exits, Oregon H100 OD could swing materially in either direction with very little warning.

Near-Term (30–90 Days)

R4 — AWS Inferentia APAC Normalization Spreads to European Markets The current geographic arbitrage (European Inferentia at $0.038–0.047/hr vs. Tokyo at $0.310/hr) is a regional capacity allocation artifact. AWS has both the visibility and the incentive to normalize European pricing upward once APAC tightening is established. The mechanism: as committed enterprise Inferentia contracts grow in Europe (following AWS European AI expansion announcements), spare APAC-style capacity gets absorbed, and spot discounts compress. Watch London Inferentia SPOT closely — any move above $0.07/hr (a doubling from today's $0.047/hr) signals the normalization wave is crossing the Atlantic.