Menu

AI infrastructure

Goldman Sachs, JPMorgan and Morgan Stanley capture 80% of equities trading revenue

Broadcom and Intel CEOs cite capital constraints as AI infrastructure risk

, , , , , ,

Two of the chip industry’s most prominent leaders say financing constraints, not semiconductor shortages, may pose the greatest risk to the artificial intelligence infrastructure build-out. As borrowing costs rise, sustaining the massive capital expenditures required for data centers and advanced chips could become the limiting factor for continued AI expansion.

  • Broadcom CEO Hock Tan said capital access could become a bigger constraint than chip, memory or power shortages on AI expansion.
  • Intel CEO Lip-Bu Tan highlighted rising capital costs and investor pressure for near-term returns on heavy AI infrastructure investments.
  • Both executives expect sustained demand through 2029 or 2030, but long-term growth depends on AI delivering measurable business value.
  • 2028-2029 Timeline when wafer production capacity may ease current shortages, according to Broadcom CEO
  • 2029-2030 Period when supply-demand imbalance for compute infrastructure could persist, Broadcom projects

CNBC reported that the chief executives of Intel and Broadcom warned in Singapore on Monday (October 12) that access to capital could emerge as a major constraint on artificial intelligence infrastructure expansion, even as demand for computing capacity continues to exceed available supply. Speaking at the DBS Asia Future Forward Conference, the two executives identified financing challenges as potentially more limiting than semiconductor shortages, memory constraints or power availability. Their remarks reflect a shift in thinking about the AI industry’s primary bottlenecks as massive capital expenditures on data centers and chip production face headwinds from higher borrowing costs and questions about eventual profitability.

Broadcom CEO names capital as the largest potential constraint for AI growth

Hock Tan said capital constraints could overshadow the supply challenges the industry has repeatedly overcome in recent years. The Broadcom chief noted that the sector has navigated multiple production bottlenecks over the past two years, from memory shortages to wafer capacity limits, and expects those particular constraints to ease as manufacturing expands.

[What] I think might be the biggest constraint is capital

Hock Tan, CEO, Broadcom

Tan suggested wafer shortages could ease by 2028 or 2029, with power availability also improving as new infrastructure comes online.

Despite financing concerns, Tan remained confident in near-term demand dynamics, stating that the imbalance between computing infrastructure supply and demand could persist through 2029 and potentially into 2030. He emphasized that the industry continues to encounter more demand for compute capacity than it can currently supply.

Intel CEO points to rising borrowing costs and return-on-investment pressures

Lip-Bu Tan raised parallel concerns about the cost and availability of capital for companies making substantial AI infrastructure bets. He highlighted investor expectations for returns on multibillion-dollar investments as a complicating factor, particularly when many infrastructure projects require years before generating positive cash flows.

Higher bond yields threaten to make funding for data center construction and chip purchases significantly more expensive, potentially challenging project economics.

Intel’s chief expected the industry to avoid a significant slowdown over the next two to three years, citing accelerating adoption of AI applications and growing demand for computing power. However, sustaining growth beyond that window will depend on how rapidly businesses integrate AI and whether the technology delivers tangible economic benefits to justify the investment levels involved.

Long-term growth hinges on AI delivering measurable returns

Both executives cautioned that the trajectory beyond the near term remains uncertain and contingent on adoption patterns and demonstrated value.

Hock Tan warned that demand for computing infrastructure could decelerate if adoption of advanced AI models disappoints relative to expectations or if capability improvements plateau. Lip-Bu Tan emphasized that companies would increasingly need to prove their AI investments generate tangible benefits through improved operational efficiency and new business models that can justify the capital deployed.

The two executives offered competing views on what constitutes the most urgent challenge, with Broadcom emphasizing capital availability and Intel stressing the need for visible returns, yet both agreed the financing environment will shape the industry’s next phase.

The BlockWest read. We note neither executive addressed whether central banks or policymakers might ease capital constraints through policy levers, nor did they propose industry-level coordination to manage demand. The capital crunch framing implicitly places the burden on private markets and individual balance sheets to navigate, leaving open whether government support for foundational AI infrastructure might become part of the political conversation in major markets.

Investors should monitor quarterly earnings results from both Intel and Broadcom over the next two quarters for updated guidance on capital expenditure trends, financing conditions and management confidence in the sustainability of current demand levels beyond 2027.

Goldman Sachs, JPMorgan and Morgan Stanley capture 80% of equities trading revenue

Wall Street data center REITs face political backlash and construction delays

, , , , , , , ,

Wall Street’s push to democratize data center investments through public REITs and funds is colliding with political backlash, construction delays and the concentrated tenant risk that underpins the sector’s economics. Investors weighing exposure to AI infrastructure need to understand both the genuine structural demand and the mounting headwinds reshaping project timelines and valuations.

  • Blackstone’s Digital Infrastructure Trust REIT has declined roughly 16% since its May 2026 debut at $20 per share, now trading under $17.
  • National polling shows 70% of Americans oppose data centers in their area, with New York and Texas imposing moratoriums on new hyperscale facility approvals.
  • Blue Owl plans a public REIT rollup of over 130 data centers valued at $6.5 billion, while Csquare’s July IPO has fallen close to 16% from its $21 debut price.
  • $1 trillion Blackstone’s stated total addressable market for data center business over several years
  • 70% Share of Americans opposing data center construction in their local area
  • $1.05 billion Csquare gross proceeds from its July 2026 initial public offering at $21 per share
  • $18 billion Blue Owl’s estimated total assets across 130 data centers in 32 global markets

Wall Street firms are aggressively marketing data center investments to retail and institutional investors as the physical foundation of the artificial intelligence economy, but the strategy is colliding with political opposition, regulatory delays and the sector’s inherent concentration risk. According to CNBC reporting, Blackstone has led this push through its Blackstone Digital Infrastructure Trust, a newly launched real estate investment trust trading on the NYSE. The firm sold 87.5 million shares at $20 each in May 2026 but the REIT has shed roughly 16 percent of its value, closing below $17 by Friday (October 9). Other major players including Blue Owl and Brookfield Asset Management’s Csquare have launched or are planning similar public vehicles, yet each faces headwinds that were less pronounced when these funds debuted.

Blackstone markets stabilized data centers as lower-risk AI exposure

Blackstone CEO Nick Pell framed the Digital Infrastructure Trust as a way to “capture the whole market of stabilized data centers and build a home in the public market where we think it belongs,” in a CNBC interview in May. The strategy focuses on mature markets such as Northern Virginia and Dallas where data center infrastructure predates the AI boom, avoiding the construction and permitting delays that plague greenfield projects. Pell described the opportunity as “massive, with a total addressable market for our business expected to eclipse $1 trillion over the next several years.”

We saw this as an opportunity to capture the whole market of stabilized data centers and build a home in the public market where we think it belongs.

Nick Pell, CEO, Blackstone

Yet Blackstone’s BXDC has not yet deployed any capital into investments despite its public listing.

Blue Owl and Csquare launch competing public vehicles amid political backlash

Blue Owl, which already manages over 130 data centers representing more than $18 billion in assets across 32 global markets, is planning a public REIT rollup valued as high as $6.5 billion according to reporting by Bloomberg. The firm’s co-CEO Marc Lipschultz called data centers “one of the strongest long-term investment opportunities in decades,” citing “attractive, lower risk-return profile” with contracts structured to “make us whole even if a tenant exits early.” Brookfield Asset Management took a different approach by listing its data center services provider, Csquare, on the NYSE in July at $21 per share, raising approximately $1.05 billion in gross proceeds before underwriting costs.

Both vehicles have underperformed their launches. Csquare shares have declined close to 16 percent since debuting in July. The pullback accelerated as political opposition to data center construction intensified nationally, with Gallup polling showing 70 percent of Americans oppose a data center being built in their area, a position that transcends party lines. New York became the first state to pass a moratorium on new hyperscale data center approvals in July, and Texas followed in August after Governor Greg Abbott, who had called the state the AI “epicenter” just a year earlier, ordered a halt on new approvals.

Oracle’s Project Jupiter delay signals regulatory and political risk to investors

The Oracle announcement in late September crystallized how political delays can derail investor return assumptions. Oracle sent a force majeure notice tied to Project Jupiter, a New Mexico data center campus developed by Blue Owl as part of the broader Stargate AI buildout, seeking to delay payment if the facility is not operational by 2028 due to regulatory hurdles and local opposition. Oracle’s stock fell 4 percent on the news, though the company stated the project remains on track. The incident underscores that for investors, the risk extends beyond whether a data center gets built to whether local and state politics slow or derail projects embedded in a fund’s return model.

An Nvidia and Blackstone-backed Australian data center company also pulled its planned IPO due to underwhelming investor interest.

Sabur Mollah, professor of finance at Gettysburg College, acknowledged that high-quality data center facilities can “generate relatively stable rental income through long-term lease agreements with large technology companies.” However, Mollah cautioned that “the key risks include overvaluation driven by excessive enthusiasm surrounding AI, substantial development and maintenance costs, dependence on a limited number of major technology tenants, and constraints on power availability.” Additional concerns span technological obsolescence, refinancing risk, construction delays and the capital intensity of ongoing operations.

Patrick Datz and Rachel Nixon of IMA, an insurance brokerage and risk-advisory firm, noted that data centers historically concentrated in “Data Center Alley” in Virginia specifically to avoid natural disasters, but the current push to build facilities nationwide exposes projects to elevated physical risk. Both said that data centers held out to investors are typically vetted and underwritten by $3 to $5 billion in insurance, and “a lot of thought is going into the design.” They described data centers overall as “a safe bet,” yet acknowledged that “if the uptime goes down” or promises are broken, contractual problems can cascade.

Retail investors have shown limited appetite so far. Roughly 94 percent of ownership in the Blackstone REIT is institutional, according to market data. That gap may narrow if Securities and Exchange Commission Chair Paul Atkins’ recent proposal to expand retail access to private investments gains traction, potentially bringing hedge fund fee structures to a broader range of allocators.

Jake Falcon, CEO of Falcon Wealth Advisors, said that publicly traded REIT vehicles offer sufficient liquidity to merit consideration by individual investors but cautioned against investing in private data center funds without deep scrutiny. Private market infrastructure funds were designed “for a relatively small number of institutional investors making large commitments and not for millions of retail investors investing smaller amounts,” according to Andrew Tarver, president of Altic at InvestCloud. That structural mismatch means redemption windows and liquidity limits can trap capital, a lesson reinforced by battles over private credit fund redemptions in 2026 as investors fled over concerns of the “SaaSpocalypse.”

The BlockWest read. We’re watching whether the institutional money backing these data center vehicles will sustain confidence as political opposition hardens and project timelines slip. The capital isn’t leaving the sector, but it’s rotating into deals already operational in permitting-friendly markets rather than greenfield builds, compressing returns and concentrating portfolio risk on a narrower set of mature facilities. The tail risk is that tenant concentration among a handful of hyperscalers plus power grid constraints prove larger headwinds than the bull case currently prices in.

Watch for Blue Owl’s REIT filing timing and asset rollup strategy, expected before year-end. Investors should also monitor whether the Oracle-Project Jupiter delay triggers similar force majeure notices from other hyperscalers, which would signal whether political risk is priced into existing fund valuations or remains underestimated.

Marathon Digital transfers 996 bitcoin to Galaxy Digital as it exits mining

Marathon Digital transfers 996 bitcoin to Galaxy Digital as it exits mining

, , , , , , ,

Marathon Digital’s transfer of bitcoin to Galaxy Digital signals an accelerating shift away from pure mining toward AI infrastructure, a strategic pivot that will define how publicly traded crypto firms allocate capital in 2026 and beyond. The sale reflects pressure to strengthen balance sheets while hedging exposure to bitcoin volatility.

  • Marathon Digital transferred approximately 996 BTC worth $81.13 million to Galaxy Digital, according to Lookonchain data flagged Friday.
  • Marathon’s bitcoin holdings fell by 18,245 BTC between February 26 and August 7 as the company reduced debt and diversified into AI.
  • The company sold 15,133 BTC for approximately $1.1 billion in March, primarily to repurchase convertible debt and fund new infrastructure investments.
  • 996 BTC transferred to Galaxy Digital in current transaction worth $81.13 million
  • 18,245 BTC reduction in Marathon holdings between February and August this year
  • $1.1B proceeds from March sale of 15,133 BTC used for debt repurchase
  • $82,000 Bitcoin price recovery over past 24 hours as MARA shares gained

CoinDesk reported that Marathon Digital, the Hallandale Beach, Florida-based bitcoin miner, moved approximately 996 BTC to an address belonging to Galaxy Digital on Friday (October 9). The transfer, flagged by blockchain tracking firm Lookonchain, is valued at $81.13 million based on current bitcoin prices. The move does not by itself confirm that Galaxy Digital executed a sale, though it signals continued liquidation of Marathon’s once-massive bitcoin reserves.

Marathon’s retreat from pure mining accelerates with $1.1 billion debt sale

Marathon has undergone a rapid transformation from a single-purpose bitcoin mining operation into a diversified infrastructure company. The company is now building capacity in artificial intelligence and high-performance computing, a strategic pivot that requires capital beyond what mining operations alone generate. To fund this expansion and shore up its balance sheet, Marathon has begun systematically reducing its bitcoin holdings.

In March, Marathon sold 15,133 BTC for approximately $1.1 billion, with proceeds directed primarily toward repurchasing convertible debt and supporting infrastructure development. The Galaxy Digital transfer represents the continuation of that pattern, converting dormant balance sheet assets into either cash or strategic partnerships that serve the company’s new direction.

Bitcoin treasury declined 34 percent in six months amid strategic repositioning

Marathon held 53,822 BTC as of February 26, when its treasury reached near-peak levels. By August 7, that figure had fallen to 35,577 BTC, representing an 18,245 BTC reduction over roughly five months. At current valuations near $82,000 per coin, Marathon’s remaining holdings represent substantial balance sheet value even after the aggressive sales campaign.

Marathon shares traded 2 percent higher in pre-market on Friday to $10 as bitcoin itself recovered above $82,000 over the previous 24 hours.

The BlockWest read. Marathon’s pivot from bitcoin treasury accumulation to active liquidation reflects a bet that AI infrastructure and high-performance computing will generate higher returns than passive bitcoin holding. Whether that shift proves correct depends on execution in unfamiliar markets and sustained capital availability as the company funds both debt repayment and new asset purchases simultaneously.

Watch for Marathon’s next quarterly filing and earnings call, where management will detail how much capital remains allocated to AI buildout versus bitcoin mining and whether Galaxy Digital or other recipients of these bitcoin transfers become anchor customers for Marathon’s infrastructure services.

Goldman Sachs, JPMorgan and Morgan Stanley capture 80% of equities trading revenue

OpenAI discloses $50 billion revenue, lower than $68 billion figure cited previously

, , , , ,

OpenAI’s actual annualized revenue of $50 billion, disclosed to investors this week, falls $18 billion short of widely circulated figures and has spooked markets betting on AI infrastructure buildout. The correction underscores how valuation expectations for private AI companies remain volatile ahead of anticipated IPOs and funding rounds.

  • OpenAI reported $50 billion in annualized revenue at end of September, lower than previously reported $68 billion figure.
  • The $68 billion figure included gross revenue from OpenAI’s partners, making the true company revenue $18 billion smaller.
  • OpenAI posted 77% total run rate growth in Q3 and 107% run rate growth for enterprise business in same quarter.
  • $50B OpenAI annualized revenue versus $68 billion figure widely reported last month
  • 77% Total run rate growth for OpenAI in third quarter
  • $852B OpenAI valuation investors must justify ahead of expected IPO
  • $122B Size of OpenAI funding round completed in March

Shares of Nvidia, Oracle, CoreWeave and other artificial intelligence infrastructure companies fell sharply on Thursday (October 8) after CNBC confirmed OpenAI’s disclosure that it reached roughly $50 billion in annualized revenue at the end of September. The figure proves substantially smaller than the $68 billion widely circulated late last month, prompting a market reassessment of AI sector expansion and spending forecasts. A person familiar with the matter explained that the higher $68 billion number included gross revenue from OpenAI’s partners, providing investors with a comparison point closer to Anthropic, its chief rival.

OpenAI corrects revenue to $50 billion after months of inflated guidance

OpenAI shared the corrected financial update through an investor presentation, the person said, requesting anonymity to discuss the numbers. The Financial Times first reported the $50 billion figure. The disclosure arrives as the company operates under pressure to justify its $852 billion valuation to investors ahead of what is widely expected to be a blockbuster initial public offering.

The gap between reported and actual revenue signals potential weakness in near-term AI deployment economics, hitting stocks whose valuations rest on assumptions of accelerating cloud infrastructure demand.

OpenAI shows enterprise growth outpacing total company metrics

Beyond the revenue correction, OpenAI reported 77% total run rate growth during its third quarter, while its enterprise business posted 107% run rate growth for the same period. The enterprise figure outpacing overall growth suggests that business customers are adopting OpenAI’s products faster than the consumer base, or that higher-margin business products are driving disproportionate gains. This split underscores where the company sees its strongest tailwinds as it prepares for the public markets.

OpenAI eyes 2027 IPO while pursuing fresh $30 billion raise

OpenAI confidentially filed its prospectus with regulators in June, and executives have signaled that the company is targeting a 2027 debut on public exchanges. Before that milestone, the company is in early stage discussions with investors about a potential new funding round that could raise around $30 billion, CNBC previously reported, though that figure remains subject to change depending on investor demand and term sheet negotiations.

OpenAI closed a historic $122 billion funding round in March, and CFO Sarah Friar told CNBC last week that the company is still “very well capitalized.”

The BlockWest read. The $18 billion gap between reported and actual revenue does not strip the company of its growth narrative, but it does suggest that OpenAI and its backers may have allowed less rigorous accounting to inflate near-term spending assumptions in the market. Investors who sized positions on the higher figure now face a reset of infrastructure spend forecasts that could drag valuations for suppliers through year-end.

Watch for OpenAI’s 2027 IPO prospectus filing and the pricing of its next funding round to reveal whether the market accepts the corrected revenue or demands valuation adjustments ahead of the public offering.

Goldman Sachs, JPMorgan and Morgan Stanley capture 80% of equities trading revenue

SpaceX seeks $40 billion in financing from Apollo and banks for Nvidia GPU purchase

, , , , , ,

SpaceX is working with Apollo Capital Management and a group of banks to finance a $40 billion purchase of Nvidia graphics processing units, deepening Wall Street’s exposure to the AI infrastructure buildout. The talks signal that debt markets, not just equity, are now central to how the biggest AI buyers fund their chip spending.

  • SpaceX is seeking $40 billion in financing to buy Nvidia GPUs, with Apollo in a lead role
  • SpaceX raised $25 billion in bonds in mid-June, two weeks after its record-breaking IPO
  • Credit spreads on AI-related bonds have widened since June, and investors now demand higher yields
  • $40B size of SpaceX’s planned Nvidia GPU financing deal
  • $25B SpaceX bond sale completed in mid-June, four months earlier
  • 7 years expected useful life of GPUs used as loan collateral

SpaceX is in discussions with Apollo Capital Management and several banks to arrange financing for $40 billion worth of Nvidia graphics processing units, according to a person familiar with the talks who spoke to CNBC on condition of anonymity. The Financial Times first reported on the talks, with CNBC reporting additional detail on structure and market context.

The deal is expected to lean primarily on the investment-grade debt market. That is the same market SpaceX tapped just four months ago.

Apollo set to lead a deal built on investment-grade debt

The person familiar with the talks told CNBC that Apollo is playing a leading role in facilitating the $40 billion financing, alongside multiple banks still in discussions with SpaceX. The structure will likely use the Nvidia GPUs themselves as collateral backing the debt, the person said. Apollo declined to comment, and Nvidia did not immediately respond to a request for comment.

SpaceX’s return to debt markets follows its bond sale in mid-June, roughly two weeks after what CNBC described as a record-breaking IPO. That earlier deal raised $25 billion across multiple bond durations and drew what the company was told at the time was unusually strong demand from fixed-income investors.

GPU collateral rests on a seven-year value assumption

A person in the credit industry told CNBC that the market is still pricing GPU-backed debt on the assumption that the chips retain value for about seven years. That timeline underpins how lenders are willing to treat Nvidia hardware as loan collateral.

The person pointed to ongoing scarcity of computing power as support for that assumption, noting the shortage of high-performing GPUs is not expected to ease soon. Persistent demand for Nvidia’s chips across AI data center buildouts has kept pricing and resale value elevated even as broader AI credit conditions shift.

Those conditions have moved against borrowers since SpaceX’s June bond sale. CNBC reported that AI-related bonds have sold off in the months since, and credit spreads tied to the sector have widened.

Meta, Amazon and Google issuances set the pattern SpaceX now follows

SpaceX’s planned $40 billion raise would follow a string of large bond issuances from Meta, Amazon and Google, as the cost of funding AI infrastructure climbs across the industry. Both sources who spoke with CNBC said additional AI-linked debt sales are expected from major buyers in the months ahead.

Fixed-income investors are now demanding higher yields to hold this paper. Both sources described that as a sign investors are becoming more selective about their AI exposure rather than stepping back entirely.

The BlockWest read. Using GPUs as direct collateral pushes AI credit risk onto a depreciating physical asset rather than corporate cash flow, a structure untested at this scale. If Nvidia’s next-generation chips outpace the seven-year value assumption sooner than lenders expect, Apollo and the banks involved, not SpaceX, absorb the first losses. Watch whether other GPU buyers follow this collateral structure or stick to unsecured corporate bonds.

Terms, pricing and the final list of participating banks have not yet been disclosed, and neither Apollo nor Nvidia has commented publicly on the discussions. CNBC and the Financial Times both indicate talks are ongoing, leaving the deal’s structure and timing as the next details to watch for.

Goldman Sachs, JPMorgan and Morgan Stanley capture 80% of equities trading revenue

Finnish regulator orders Google to suspend data center construction in Finland

, , , , , , ,

Finnish regulators have ordered Google to pause construction on two data center sites just weeks after the company pledged $15 billion for AI infrastructure in the country, exposing friction between Europe’s AI buildout and environmental review rules. The halt affects projects in Muhos and Kajaani and must take effect no later than October 23.

  • Finland’s LVV ordered Google subsidiary Tuike Finland Oy to stop construction work by October 23
  • The suspension covers tree removal, topsoil stripping, excavation, quarrying and road building at both sites
  • Google committed $15 billion to Finnish AI infrastructure just one month before the halt order
  • $15B Google’s pledged Finland AI investment, its largest in Europe to date
  • Oct. 23 deadline for Google to suspend construction, set by Finnish regulator LVV
  • 130 ha land Google plans to replant at Muhos under its biodiversity commitment
  • 2028 year by which Google’s $15B Finland capital is to be deployed

Google has been told to stop building two data centers in Finland, a country that has drawn so much hyperscaler investment it is now nicknamed the “Texas of Europe,” according to CNBC reporting. The Finnish Licensing and Supervision Agency, known as LVV, issued the stop-work notice on Tuesday to Tuike Finland Oy, Google’s local subsidiary, covering projects planned for Muhos and Kajaani.

The order lands roughly a month after Google announced a $15 billion AI infrastructure commitment to Finland, its largest single investment in Europe. That capital was slated for deployment through 2028 across data centers and supporting energy projects, the company said at the time.

LVV halts tree removal and excavation pending environmental review

According to a Google translation of the LVV press release, the suspension targets a specific set of site-preparation activities: removal of trees and topsoil, excavation, quarrying, crushing, soil transfer, ditching, and construction of roads and storage areas. The regulator says these actions must stop until environmental impact assessments, known as EIAs, are completed for both the Muhos and Kajaani sites.

The order carves out exceptions for lower-impact work. Planning, measurements, soil surveys and other activities deemed easily reversible are not covered by the halt, LVV said.

Tommi Muilu, head of the environmental department at LVV, framed the issue as a matter of sequencing rather than outright opposition to the projects.

“In our opinion, the measures in question will change the environment and cause impacts, the identification and assessment of which are among the key objectives of the EIA procedure.”

Tommi Muilu, head of the environmental department, Finnish Licensing and Supervision Agency

Google says it acted in good faith under forestry law

A Google spokesperson told CNBC the company accepts the regulator’s findings even as it defends its process. “We understand the concerns and have fallen short of our own high standards in this instance,” the spokesperson said, adding that Google would follow LVV’s guidance.

The spokesperson said Google conducted nature surveys and had measures in place to protect high-value habitat areas under Finland’s Forestry Act before construction began. The company also pointed to a 130-hectare tree-planting plan at Muhos as part of its longer-term landscaping and biodiversity commitments in the region, though it did not dispute LVV’s characterization of the halted activities.

LVV’s own account of the relevant activities, including large-scale soil and vegetation removal ahead of a completed EIA, is the only detailed version of events in the record so far; Finnish environmental groups or local officials who may have raised concerns are not quoted in the reporting.

Finland’s power and land advantage draws hyperscalers despite the pause

Finland has become a magnet for data center development because of available land and electricity capacity that is scarce across much of the rest of Europe. Several projects with potential capacities in the hundreds of megawatts have been announced in the country in recent months, reinforcing comparisons to Texas, where Google, Meta, Microsoft and Anthropic have all announced large AI data center projects.

The stop-work order does not cancel Google’s Finnish buildout. It pushes the Muhos and Kajaani timelines back until the EIA process concludes, with no new completion date disclosed.

The BlockWest read. For companies racing to lock down scarce European power and land, Finland’s stop-work order is a reminder that permitting risk, not just power availability, now sits on the critical path for AI capex. Hyperscalers modeling multibillion-dollar buildouts through 2028 need to price in environmental review timelines the way they already price in grid interconnection queues, or risk repeat delays as other jurisdictions tighten similar rules.

Google must suspend the covered construction activities at Muhos and Kajaani no later than October 23, pending completion of the environmental impact assessments; neither Google nor LVV has given a timeline for when those assessments, or the resumption of work, might be finalized.

Luxor extends hedging services to AI compute operators facing price pressure

Luxor extends hedging services to AI compute operators facing price pressure

, , , , , , ,

Falling prices for GPU computing power are squeezing the data-center operators who borrowed heavily to build AI infrastructure, and a new class of financial contracts is emerging to help them manage that risk. Luxor, known for serving Bitcoin miners, has extended its hedging playbook into AI compute while CME Group prepares to launch exchange-traded futures tied to GPU rental rates.

  • Luxor told CryptoSlate it is already brokering agreements between GPU capacity owners and customers seeking compute access.
  • CME Group’s Aug. 11 announcement targeted Oct. 5 for H100 and B200 rental-index futures, pending regulatory review.
  • Luxor could not provide a customer hedge example or current derivatives trading volumes, saying a liquid market has not yet formed.
  • $69,000 B300 GPU price versus roughly $36,000 for a new H100
  • $36,000 new H100 system price compared with $29,000 for refurbished units
  • $500,000 hypothetical hedge payout on a 50-cent GPU-hour rate drop
  • Oct. 5 CME’s target launch date for GPU rental futures, pending review

Companies building AI applications can rent GPUs rather than buy them, and falling hourly rental prices make those applications cheaper to run. The same price declines can wreck the finances of operators who bought racks of GPUs on the assumption that customers would keep paying a certain rate, according to CryptoSlate. Machines can keep working and AI demand can stay strong even as per-hour earnings slide below what is needed to service debt on the equipment.

Luxor brokers AI compute deals but has no liquid market yet

Luxor built its business providing services and financial products to Bitcoin miners, and it has now folded AI compute derivatives into its expanded AI infrastructure business. The company said it sees an opportunity to apply its mining-hedging experience to AI operators, who also spend heavily on hardware before knowing what it will earn. Luxor told CryptoSlate it is already brokering agreements between owners of computing capacity and customers who want to use it.

The cash-settled derivatives side of that business remains early. Luxor said it could not supply a customer hedge example or current trading volumes because a liquid market has not yet formed.

Cash-settled contracts work by paying money according to a price formula rather than requiring an exchange of GPU capacity. In a hypothetical case described in the reporting, an operator expecting to sell 1 million GPU-hours a month at $2 an hour would earn $2 million. If the benchmark price fell to $1.50, a hedge would pay the 50-cent difference across those hours, or $500,000, restoring the combined total to roughly $2 million before fees. The obligation runs both ways: if the benchmark rose to $2.50, the operator would owe $500,000 instead of keeping the extra revenue.

CME targets Oct. 5 for H100 and B200 rental futures

CME Group is building an exchange-traded version of the same idea. Its Aug. 11 announcement targeted Oct. 5 for H100 and B200 rental-index futures tied to benchmarks from Silicon Data, subject to regulatory review.

Listing a contract does not guarantee enough trading activity to make it easy to use. The success of CME’s futures will depend on whether operators and their customers actually trade them in volume once they launch.

Basis risk and GPU prices complicate the hedge

Even a working hedge depends on the benchmark price tracking what an operator’s actual customers pay, a gap known as basis risk. If an operator’s customers negotiate rates down to $1.25 an hour while the published benchmark falls only to $1.50, the same $500,000 hedge payout would bring $1.25 million in rental income to $1.75 million, leaving a shortfall against the $2 million target despite the contract performing exactly as written.

Luxor has drawn a parallel to its hashprice measure for Bitcoin mining, which gave miners a shared revenue reference even when their operating costs differ. AI rentals are harder to standardize because customers attach different values to access that looks identical on a spec sheet, such as guaranteed uninterrupted use versus capacity that can be reclaimed on short notice. CCIR’s rental-data methodology already treats interruptibility and commitment length as separate pricing characteristics, and it relies on publicly advertised rates that may not capture privately negotiated discounts.

Luxor’s own AI Hardware Price Index tracks advertised equipment prices rather than rental rates. B300 systems have climbed toward $69,000 while new H100s have settled near $36,000 and refurbished units near $29,000, figures useful for assessing a purchase but not for settling a rental hedge.

Luxor’s August announcement described expanded compute spot pricing as forthcoming but did not detail collateral requirements for AI hedges or procedures if a counterparty fails to pay. The company’s reply also left unanswered how it separates its own trading from the deals it arranges for customers, notable because its launch announcement disclosed an internal compute trading fund.

The BlockWest read. For lenders and allocators financing GPU buildouts, the real test is not whether a hedge exists on paper but whether its benchmark tracks an operator’s actual contracts closely enough to matter during a bad month. Until Luxor discloses collateral terms, default procedures, and how its own trading fund interacts with client business, lenders should treat these instruments as a developing risk-management layer, not a substitute for cash reserves.

CME Group’s Oct. 5 target for H100 and B200 rental futures, still subject to regulatory review, will offer the first visible test of whether an exchange-traded GPU compute market can attract enough participants to trade efficiently.

Amazon warns local data center opposition threatens US AI competitiveness

Amazon warns local data center opposition threatens US AI competitiveness

, , , , , ,

Amazon has published a lengthy defense of its AI data center buildout, warning that local opposition could cost the United States its edge in artificial intelligence. The blog post arrives as more than 100 proposed moratoriums on data center construction work their way through communities nationwide, according to The Verge.

  • Amazon Web Services CEO Matt Garman published a blog post exceeding 3,000 words on Friday (October 2)
  • Garman said over 100 data center moratoriums are currently being considered across the country
  • Amazon pledged more than $1 billion to support communities neighboring its data center sites
  • 3,000+ word count of Garman’s blog post defending data centers
  • 100+ data center moratoriums under consideration nationwide, per Garman
  • $1B+ Amazon pledge to communities near its data center sites

Amazon is pressing the public not to stand in the way of its AI infrastructure expansion, framing local resistance as a threat to national competitiveness. Garman’s post, published Friday (October 2), argues that blocking data center projects over concerns about jobs, electricity costs and environmental impact risks handing AI leadership to rival nations.

Garman says the US is in a race it cannot afford to lose

Garman framed the stakes in explicitly geopolitical terms. He wrote that the AI buildout matters so much internationally that “there are widespread reports of various countries intentionally seeding misinformation in the US about data centers to trick us into slowing down.”

The post pushes back on what Garman characterized as myths spreading faster than facts in an era of social media and constant news cycles. He acknowledged that legitimate questions about data centers deserve answers, but argued misinformation is muddying the debate over projects that he says are vital to American economic and national security interests.

“With any change, there will be important questions raised, but there will also be misinformation and outright lies, and in the age of social media and 24/7 news, myths take hold faster than ever before.”

Matt Garman, CEO, Amazon Web Services

Amazon codifies a Data Center Commitment and drops NDAs with officials

The blog post lays out a set of practices Amazon is formalizing as a “Data Center Commitment,” covering job creation, protections against local electricity rate increases, and the billion-dollar-plus community investment pledge. The move comes as public support for data centers has fallen in polling and pushback has intensified in the communities where facilities are being built, driven partly by broader anxiety over AI safety and its societal effects.

Amazon also disclosed that it has stopped using nondisclosure agreements with government agencies. That change follows scrutiny from Democratic lawmaker Jamie Raskin, who questioned tech companies over confidential arrangements struck with local officials during site negotiations.

Amazon’s chief global affairs and legal officer, David Zapolsky, addressed the NDA practice in an interview reported by The Wall Street Journal. He said “the industry comes from a history of secrecy and so some of that’s just reflex,” adding that community leaders need to be able to share information with constituents and that continued secrecy “just didn’t make sense.”

Over 100 local moratoriums remain the open fight

Garman’s post does not name which jurisdictions are weighing the more than 100 moratoriums he cites, nor does it detail how Amazon plans to engage with specific communities beyond the financial pledge. The Verge’s report does not include a response from the local officials or residents pushing those measures, so their side of the dispute is not represented in this account.

The outcome of those local votes, not Amazon’s blog post, will determine how much of the planned buildout actually gets built.

The BlockWest read. Amazon’s billion-dollar community pledge and NDA reversal look less like generosity and more like risk management for a balance sheet tied to AI infrastructure capital spending. Investors in Amazon and its cloud peers should watch moratorium vote counts as a leading indicator of buildout timelines, since every blocked site delays compute capacity that AWS has already priced into its AI growth forecasts.

The more than 100 moratoriums Garman cited are still working through local governments, and no timeline for resolving them has been disclosed; how many pass will determine whether Amazon’s public appeal changes outcomes on the ground.

Hut 8 closes $1.07 billion credit facility with 40% liquidity requirement from 2027

Hut 8 closes $1.07 billion credit facility with 40% liquidity requirement from 2027

, , , , , , , ,

Hut 8 has locked in a $1.07 billion, four-year revolving credit facility at the parent level, giving the Bitcoin miner and AI infrastructure developer fresh non-dilutive capital as it builds out power-hungry data center campuses. The facility comes with a minimum-liquidity covenant that could force the company to hold 40% of total commitments in reserve once the rule takes effect in 2027.

  • Hut 8 closed the $1.07 billion senior secured revolving credit facility on September 24, with nothing drawn at closing.
  • A minimum-liquidity covenant begins with the quarter ending March 31, 2027, requiring 40% of commitments before a defined stabilization date and 25% after.
  • Term SOFR borrowings carry an initial margin of 175 basis points, adjustable between 150 and 200 points based on Hut 8’s debt-to-market-capitalization ratio.
  • $1.07B four-year revolving facility, about 4.6x Hut 8’s June cash
  • $233.6M cash Hut 8 held on its balance sheet as of June 30
  • $7.5B prior non-recourse financing for River Bend and Beacon Point campuses
  • 40% minimum liquidity covenant before stabilization, versus 25% after

Bitcoin miner Hut 8 Corp. announced on September 28 that it had closed a four-year, $1.07 billion senior secured revolving credit facility, according to reporting by CryptoSlate. The deal, disclosed in a press release, gives the Nasdaq- and TSX-listed company committed bank liquidity to draw on during site development or to back construction obligations through letters of credit. A related securities filing confirmed no amounts were outstanding when the facility closed on September 24.

Hut 8 closes $1.07 billion facility with nothing drawn

The facility names Hut 8 Corp. as borrower, with certain restricted subsidiaries guaranteeing its obligations. First-priority liens cover substantially all assets of the borrower and guarantors, subject to standard exclusions, according to an 8-K filing with the Securities and Exchange Commission.

The $1.07 billion letter-of-credit sublimit sits inside the same overall commitment as the cash borrowing capacity, the press release states. In practice, that means Hut 8 can tap one pool of bank capital either to borrow directly or to post letters of credit covering interconnection deposits and obligations to utilities and equipment vendors, cutting the cash collateral it would otherwise need to set aside.

J.P. Morgan acted as Lead Left Arranger, Bookrunner, and Administrative Agent, while Citi, Goldman Sachs and Morgan Stanley served as Joint Lead Arrangers and Joint Bookrunners. The facility was provided by a syndicate of 12 lenders, the company said.

40% liquidity threshold applies before a defined stabilization date

The agreement includes a minimum-liquidity covenant that begins with the quarter ending March 31, 2027. Hut 8 must maintain liquidity, as defined in the credit agreement, equal to 40% of total commitments before a stabilization date it has not disclosed, dropping to 25% afterward, with equity cure rights available if the company falls short.

That threshold sits against a June 30 balance sheet that showed $233.6 million in cash, reported separately from restricted funds, according to Hut 8’s quarterly filing. The new facility is more than four and a half times that cash balance, giving the company substantially more headroom than its unrestricted cash alone would provide.

Term SOFR loans under the facility carry an initial margin of 175 basis points, which can move between 150 and 200 basis points as Hut 8’s debt-to-market-capitalization ratio changes. The agreement also restricts additional debt and liens, subject to qualifications spelled out in the credit documents.

Sean Glennan frames the line as parent-level flexibility, not project debt

Sean Glennan, Hut 8’s chief financial officer, described the facility as part of a broader push toward an investment-grade corporate profile.

We are building a capital structure designed to scale with the business while giving us control over when, where, and how we deploy capital, flexibility that matters given the speed and capital intensity of AI infrastructure development.

Sean Glennan, Chief Financial Officer, Hut 8

Hut 8 has separately arranged $7.5 billion in non-recourse, fully amortizing project financing for its River Bend and Beacon Point AI data center campuses. Any borrowing or letters of credit issued under the new parent-level facility would create obligations at Hut 8 Corp. and its guarantors that sit alongside, rather than inside, those existing project structures.

The BlockWest read. The 40% liquidity covenant matters more to Hut 8’s balance-sheet flexibility than the headline $1.07 billion figure does. It effectively locks up several hundred million dollars of the facility as an untouchable reserve once the covenant starts in the quarter ending March 31, 2027, meaning the real deployable capital for construction and letters of credit is smaller than the topline number suggests until Hut 8 clears its undisclosed stabilization date.

Hut 8 has not disclosed the stabilization date that determines when its liquidity requirement drops from 40% to 25%, nor how much of the facility it intends to draw as River Bend and Beacon Point construction proceeds. Investors will get their next look at usage and covenant compliance when Hut 8 reports results for the quarter ending March 31, 2027, the first period in which the minimum-liquidity test applies.

Anthropic commits $11.6 billion to Akamai for seven-year cloud deal

Anthropic commits $11.6 billion to Akamai for seven-year cloud deal

, , , , , , ,

Anthropic’s commitment to spend $11.6 billion on Akamai’s cloud infrastructure over seven years represents a significant shift in how AI companies structure compute deals, anchoring the arrangement to CPU capacity rather than the GPU-focused arrangements that have dominated the sector. The warrant structure ties Akamai equity to Anthropic’s spending milestones, reversing the typical pattern in which chip suppliers invest directly in AI labs.

  • Anthropic committed $11.6 billion over seven years to Akamai, more than six times a $1.8 billion deal announced in May.
  • Akamai issued Anthropic a warrant convertible into up to 7.7 million common shares, or about 5% of outstanding stock, at $111.33 per share.
  • The deal hinges on Akamai meeting delivery and service-availability requirements, with either party able to terminate under specified conditions.
  • $11.6B Total value of Anthropic’s seven-year compute commitment to Akamai.
  • $1.7B Expected annual revenue pace by end of 2028, starting second half of 2027.
  • $5.5B Capital spending Akamai expects to deploy building out cloud capacity for the deal.
  • 5% Maximum equity stake Anthropic can acquire through warrant exercise if full spending occurs.

TechCrunch reported on Friday (September 25) that Anthropic will pay Akamai $11.6 billion over seven years for dedicated cloud computing capacity and managed support services. The value dwarfs an earlier arrangement between the companies announced in May, when Bloomberg reported an $1.8 billion computing deal. According to Akamai’s securities filing, the commitment depends on Akamai meeting certain delivery and service-availability requirements, and both parties retain termination rights under defined conditions including material breach or change of control.

Akamai’s largest deal focuses on CPU infrastructure, not GPUs

The arrangement marks Akamai’s largest contract in company history and reflects growing demand for CPU capacity as AI agents take on broader operational tasks. Rather than the GPU-centric infrastructure that has dominated AI compute discussions, this deal centers on general-purpose chips that handle code execution, data processing, and web browsing. Akamai declined to specify what tasks Anthropic will assign to the CPU capacity, leaving the actual use case unclear to investors and competitors.

Akamai will not recognize revenue from the deal this year. On an investor call Thursday, executives disclosed expectations of $150 million to $300 million in revenue beginning in the second half of 2027, ramping to an annual pace of roughly $1.7 billion by the end of 2028. To support the infrastructure buildout, Akamai plans to spend approximately $5.5 billion in capital deployment and is adding $1.7 billion to its 2026 capital spending to pre-purchase components such as memory.

Warrant structure ties equity to spending milestones in reverse of typical AI deals

Akamai issued Anthropic a warrant convertible into up to 7.7 million common shares or approximately 5% of Akamai’s outstanding stock at $111.33 per share. This represents the first time Akamai has attached a warrant to a cloud services agreement and inverts the standard pattern in AI infrastructure deals, where chipmakers and cloud providers typically invest equity directly in the AI labs that become their customers.

About 2% is expected to vest, or become available to Anthropic, upon Anthropic’s first payment under the agreement. The rest is tied to Anthropic spending more. Each additional $3 billion it commits to Akamai’s cloud services unlocks roughly another 1%, so the deal could grow by as much as $9 billion, to about $20 billion in total. AMD employed a similar warrant structure with OpenAI last year, anchoring equity grants to chip-purchase milestones.

Anthropic CEO Dario Amodei told the New York Times last December that the company does not participate in such investment arrangements “at the same scale as some other players,” yet Amazon, Google, Microsoft, and AMD have all invested in or committed capital to Anthropic while supplying chips or cloud capacity.

Market reaction and questions around deal enforcement

Akamai shares rose as much as 17% in after-hours trading on Thursday, the Wall Street Journal reported.

The deal contains several provisions that could limit its actual value. The Master Services Agreement allows Anthropic to terminate each project plan upon notice of a material outage, subject to certain conditions, and grants Anthropic a right to exit upon a change of control of Akamai in favor of a direct competitor. Akamai retains the right to terminate if Anthropic materially breaches and fails to cure the breach. Either party may exit if the other enters bankruptcy or if no active project plan remains in effect, raising the question of whether Anthropic will remain committed to full spending as AI infrastructure markets evolve and alternative suppliers emerge.

The BlockWest read. Anthropic’s shift toward CPU capacity signals confidence that general-purpose computing will drive near-term AI economics, yet the arrangement’s escalating warrant structure suggests Akamai views this as a speculative bet on Anthropic’s long-term scale rather than a contract backed by immediate revenue certainty. The deal’s termination clauses and outage-based exit rights leave material performance risk unresolved.

Watch for Akamai’s 2027 revenue recognition from the deal, expected to commence in the second half of that year, which will reveal whether Anthropic’s CPU spending matches the $150 million to $300 million guidance executives provided and whether the deal’s milestones track toward the projected $1.7 billion annual pace by late 2028.

Load More

Back To Top
© BlockWest 2026
BlockWest Media LLC

Equities, AI, and blue-chip crypto, covered as one connected market.

Sections

Company

Legal

Follow

𝕏 ▶ in

© 2026 BlockWest. All rights reserved.