Nvidia × IREN: $2.1B equity option + 5 GW + $3.4B GPU cloud — a former bitcoin miner becomes the hot card of AI data centers

$2.1B + 5 GW + $3.4B — IREN's inflection on May 7

Here's the deal. On May 7 Nvidia and Australian data-center operator IREN simultaneously released two announcements. First, a strategic partnership to deploy up to 5 GW of NVIDIA DSX-aligned AI infrastructure across IREN's global pipeline over five years, anchored by the Texas Sweetwater 2 GW campus. Second, IREN granted Nvidia a 5-year right to purchase up to 30 million shares at $70 each — a maximum $2.1B equity investment option. In a separate release, IREN-Nvidia signed a 5-year $3.4B GPU cloud services agreement.

Total deal value: $2.1B (Nvidia → IREN equity) + $3.4B (IREN → Nvidia cloud revenue) = $5.5B cycle flowing in both directions over five years. That's the central mechanism. Capital flows from Nvidia to IREN; GPUs flow from Nvidia to IREN; IREN's GPU cloud revenue cycles back to Nvidia as GPU payments. One round trip benefits both parties.

After the announcement, IREN was up 12–15% and Nvidia rose alongside. A year ago IREN was classified as a bitcoin miner; this deal moves it formally into the "AI data center operator" category. Market cap has $20B → $50B visibility within six months.

Each player — Nvidia, IREN, Sweetwater

Nvidia. Latest card in the four-month accumulation of $40B in AI equity bets. Unlike OpenAI ($30B), Corning ($3.2B), Intel ($5B), IREN is simultaneously a "GPU cloud operator" generating direct revenue for Nvidia AND a recipient of capital. Nvidia capital flows into IREN data center ramp, enabling more GPU purchases — direct feedback loop.

IREN Ltd. Sydney-based, Nasdaq-listed. Founded 2021 as a bitcoin miner; pivoted aggressively to AI data centers over 2024–2025. Q1 2026 revenue $250M (+180% YoY), AI/GPU cloud revenue 65%, bitcoin mining 30%, other 5%. Operates Texas Sweetwater + British Columbia (Canada) + Australia campuses. Differentiation: full-stack data center (power procurement + site + cooling + GPU operations) all in-house.

NVIDIA DSX (Data center Solutions for AI eXcellence). Reference architecture Nvidia built for data center operators. (1) GB300/Rubin GPU clusters, (2) NVLink 6 + Spectrum-X + InfiniBand 800G networking, (3) immersion or direct liquid cooling, (4) 30–50MW power modules, (5) Dynamo inference stack, (6) NVIDIA AI Enterprise software full stack. IREN deploys 5 GW following the DSX standard.

Sweetwater 2 GW campus (Texas). IREN's flagship since 2024. Phased expansion 250MW → 750MW → 2GW. Power: ERCOT (Texas grid) + own wind/solar PPAs. Immersion cooling infrastructure as core differentiator. Anchor 2 GW of the 5 GW total. Remaining 3 GW spread across Canada BC + Australia + additional US states (Arkansas, Wyoming under review).

Core developments — deal structure and capital cycle

Item	Detail
Nvidia option for IREN shares	5 years × 30M shares × $70 = up to $2.1B
GPU cloud services contract	5 years $3.4B (IREN bills Nvidia)
AI infrastructure deployment	5 GW (NVIDIA DSX-based)
Sweetwater anchor	2 GW (Texas)
Canada / Australia / US additional	3 GW
Online	Phased from H2 2026; 5 GW complete by 2030

The capital cycle elegance. Nvidia commits $2.1B capital to IREN → IREN accelerates NVIDIA DSX-based data center ramp → IREN buys massive Nvidia GPUs → IREN generates GPU cloud revenue → 60–70% of that revenue cycles back to Nvidia as GPU payments. Over five years, the $2.1B Nvidia bet returns as GPU revenue scenarios of $10B+.

Risk analysis. (1) Data center ramp delay — Texas ERCOT grid constraints, site permitting, cooling infrastructure material supply, (2) GPU cloud price pressure — CoreWeave, Lambda, Crusoe, Together AI may accelerate price competition, (3) AI training demand cycle — 2027–2028 transition from training to inference may lower cluster utilization, (4) Bitcoin mining revenue decline — IREN's existing 30% revenue is BTC-exposed.

Companion announcements. The May 7 cluster of Nvidia announcements (OpenAI, Corning, IREN) sends a clear message: in four months Nvidia has locked every major card of the AI infrastructure capex cycle with capital. Direct pressure on other GPU cloud operators (CoreWeave, Lambda, Together AI). Without a capital alliance, you can't match IREN's ramp velocity.

Who wins — beneficiary breakdown

Nvidia. GPU revenue stability + capital appreciation upside. IREN's $5–10B incremental GPU purchases over 5 years scenario, while the $2.1B equity option could 5–10x via IREN market cap appreciation. The last puzzle piece in the full cycle, after OpenAI $30B and Corning $3.2B.

IREN. $20B → $50B market cap visibility + lowest-ever cost of capital. Category transition from bitcoin miner to AI data center operator complete. Nvidia capital + DSX standard + GPU cloud revenue lock-in delivers ramp velocity ahead of any competitor over the next five years.

Sweetwater Texas local economy. A 2 GW data center brings 1,500 direct jobs + 5,000 indirect + $5–8B in power/cooling infrastructure investment. Texas state government actively supportive. The combination of energy + data center cements Texas as an "AI infrastructure super hub."

Existing GPU cloud competitors (CoreWeave, Lambda, Crusoe, Together AI): Most pressured. CoreWeave is NYSE-listed at ~$30B and larger than IREN, but likely behind IREN in the Nvidia capital alliance. Lambda Labs is mid-Series D ($5B valuation) — valuation pressure. Crusoe differentiates with natural gas + data center fusion. Together AI counters with model hosting + data center bundling.

Bitcoin miner category (Marathon, Riot, Hut 8): IREN's successful AI pivot enables category-wide re-rating. Marathon is converting some campuses to AI. Hut 8 is ramping GPU data centers in BC.

Korean data center operators (KT, LG U+, SK Telecom, NHN Cloud): Pressure from global AI data center capex acceleration on Korea's domestic capacity shortfall. Concurrent pressure to build own GPU clouds for Korean model training demand (LG AI, KT Mi:dm, SKT A.X).

Past parallels — Equinix, Digital Realty, CoreWeave

Equinix-AWS/MS/Google triangle, 2010s. Cloud Big 3 locked data center operators with capital + contracts to secure stable capacity. Equinix grew to $80B-class REIT. IREN is the AI-era version of this pattern. Differences: (1) Equinix is colocation-centric while IREN bundles GPU operations + cloud services, (2) AI cycle capex explosion runs 3–5x faster than the Equinix cycle.

CoreWeave 2024–2025 IPO + ramp. First mega-case of bitcoin miner → AI data center pivot. Post-NYSE IPO at ~$30B. Limitation: high concentration on OpenAI and Microsoft contracts → single-customer risk. IREN differentiates: (1) Nvidia capital alliance + multi-customer base (OpenAI plus Anthropic, Cohere, Mistral plus enterprise direct), (2) regional diversification across Texas + Canada + Australia.

Digital Realty data center REIT. ~$50B market cap. Global #2 in colocation. Slower to enter own GPU operations for AI ramp. Counter: strengthen site/power leasing to AI operators. Diverges from the IREN model of capital + operations integration.

Counter-case: Compute North 2022 bankruptcy. Started as bitcoin mining + data center but went bankrupt during the BTC price crash. Direct similarity to IREN's risk pattern. Difference: IREN's Nvidia capital lock-in reduces BTC exposure, and AI revenue is already 65%. Even so, a >50% BTC price drop would still pressure IREN revenue by –10–15%.

Competitor counterplays — CoreWeave, Lambda, Crusoe, Together AI

CoreWeave's counter. May 8 announcement of an additional 5-year contract with OpenAI ($11B class) emphasizes own revenue stability. As a NYSE-listed company, capital channels are more diversified than IREN's. But behind IREN in the Nvidia capital alliance. Counter: mega-contracts with OpenAI/MS + global data center expansion (UK, Germany, Japan).

Lambda Labs's counter. Differentiated on GPU cloud price competitiveness + developer-friendly interface. Series D round in progress. Targets ML engineers / startups rather than direct enterprise. Market segmentation away from IREN's enterprise mega-contracts.

Crusoe Energy's counter. Natural gas power + data center bundling for power cost differentiation. AI data center ramp underway. Capital + Nvidia alliance still weaker than IREN.

Together AI's counter. Differentiated via model hosting + inference optimization + data center bundling. Strong open-source model hosting (Llama, Mistral, Qwen). Rather than head-on competition with IREN, climbs up the value chain into the inference optimization layer.

Korea-specific: NHN Cloud, KT Cloud, NAVER Cloud. AI data center capacity shortfall is structural. Counter via own GPU clouds + Japan/Singapore data center expansion. To prevent Korean model companies (Upstage, Naver HyperCLOVA, KT Mi:dm) losing training demand to global operators like IREN/CoreWeave, urgent expansion of Korean GPU capacity is needed.

So what changes — by persona

AI model companies (OpenAI, Anthropic, Cohere, Mistral): GPU cloud option diversification strengthens negotiation leverage. But mega contracts (5-year $3–10B) only viable with operators that have capital scaling capability. Emerging operators like IREN becoming mega-contract candidates strengthens multi-cloud strategies among major model vendors.

Enterprise CIOs: Multi-source AI infrastructure strategy now considers IREN-class emerging operators as options. CoreWeave + AWS + Azure + IREN as a four-vendor strategy. But IREN's market adaptation speed and SLA stability need 6–12 month monitoring.

Investors: IREN $20B → $50B visibility. Concurrent rerating possible for CoreWeave + Crusoe + Together AI. Re-evaluation of AI pivot potential across the entire bitcoin miner category.

Data center engineers / operators: Need to be familiar with NVIDIA DSX. Immersion cooling + 30–50MW modules + Spectrum-X networking become standard skill set. Demand explodes for talent combining bitcoin mining ops experience + AI data center experience.

Korea market implications: Korean data center domestic capacity is among the biggest weak points in the global capex cycle. Power procurement + site permitting + cooling infrastructure all lag US/Japan/Singapore. Kakao, Naver, KT will accelerate Japan/Singapore data center expansion. Korean government urgently needs fast-track policies on data center sites and power.

References

• NVIDIA and IREN Announce Strategic Partnership — NVIDIA Newsroom, 2026-05-07
• Nvidia to Invest Up to $2.1 Billion in Data Center Firm IREN — Bloomberg, 2026-05-07
• IREN inks AI infrastructure deal with Nvidia — CNBC, 2026-05-07
• NVIDIA DSX reference architecture — NVIDIA Developer
• IREN Q1 2026 results — IREN IR
• Texas ERCOT power grid + AI data center — Reuters, 2026

Key signals to track over the next 12 months

Four signals will determine whether the IREN bet validates Nvidia's vendor-financing thesis or becomes its cautionary tale. First, Sweetwater 750MW phase online date — currently targeted for Q3 2026 — every quarter of slip cuts roughly $200M from IREN's GPU cloud revenue ramp and pressures the $3.4B contract trajectory. Second, ERCOT grid response — Texas already runs at 95% peak summer capacity, and adding 2GW concentrated in West Texas requires ERCOT to fast-track new transmission corridors; failure here forces IREN to relocate growth. Third, the customer mix at IREN GPU cloud — if OpenAI dependency exceeds 40%, single-customer concentration risk peaks; if it stays diversified across Anthropic, Cohere, Mistral, and enterprise, IREN's risk profile justifies the $50B market cap scenario. Fourth, BTC price floor — IREN's residual 30% mining revenue is the cushion against AI cycle softness, and a sub-$60K BTC year stresses the P&L meaningfully. Track Sweetwater milestones, ERCOT filings, IREN customer disclosures, and the BTC chart in tandem.

Bottom line

The Nvidia-IREN deal completes the four-month Nvidia capital-cycle play: GPU buyer (OpenAI), GPU supplier (Corning), GPU operator (IREN), GPU sovereign (Intel). What was a $40B portfolio bet becomes a $100B+ revenue scenario over five years if all four execute. IREN is the highest-velocity execution risk — bitcoin miner pivoting to AI operator at full scale — but also the most leveraged upside. Worth watching as the test case for whether vendor financing can manufacture demand instead of just chasing it.