Analysis

Cerebras IPO: The Wafer-Scale AI Challenger That Just Priced at $185 — and Why the Market Is Betting It Can Crack Nvidia’s Fortress

Published

2 months ago

May 14, 2026

Cerebras Systems (CBRS) priced its IPO at $185/share on May 13, 2026, raising $5.55 billion at a $56B+ valuation. Here’s a deep analytical dive into the Cerebras wafer-scale chip, WSE-3 vs. Nvidia, the OpenAI deal, financials, risks, and whether CBRS stock is worth buying.

There is a dinner-plate-sized piece of silicon sitting inside a data center in Sunnyvale, California, that Wall Street just valued at more than $56 billion. On the evening of May 13, 2026, Cerebras Systems priced its initial public offering at $185 per share — well above a revised range of $150 to $160, which was itself a sharp upgrade from the original $115 to $125 estimate floated just days earlier.

When trading opened on the Nasdaq under the ticker symbol CBRS on Thursday morning, the question hanging in the air was not whether artificial intelligence infrastructure had become the most consequential capital formation story of the decade. That debate is long settled. The real question is whether Cerebras Systems — a ten-year-old chip startup built around a radical idea so counterintuitive it initially drew more skepticism than funding — has genuinely broken open a new chapter in AI hardware, or whether it is riding a wave of irrational exuberance that will eventually meet the immovable reef of Nvidia’s dominance.

Key Takeaways

Cerebras IPO priced at $185/share on May 13, 2026, raising $5.55 billion — one of the largest US tech IPOs in recent years, with the book approximately 20x oversubscribed at the original range.
Market cap exceeds $56 billion at IPO price, implying a trailing revenue multiple of ~100x on $510 million of 2025 revenue that grew 76% year-over-year.
The WSE-3 wafer-scale chip is 57x larger than Nvidia’s H100, delivering claimed inference speeds up to 15x faster on leading open-source models.
The OpenAI deal — worth over $20 billion for 750MW of contracted compute — provides significant revenue visibility but also creates future customer concentration risk.
UAE concentration (MBZUAI at 62%, G42 at 24% of 2025 revenue) remains the key near-term risk; AWS partnership and enterprise channel development are the most important de-risking catalysts.
CBRS stock trades on Nasdaq; investors seeking positions are advised to monitor post-IPO earnings for revenue diversification evidence before making significant commitments.

The numbers arriving into the open market are, by any measure, arresting. Cerebras sold 30 million Class A shares, with underwriters holding a 30-day option to purchase up to 4.5 million additional shares, generating gross proceeds of $5.55 billion — making it one of the largest technology IPOs in recent American history. The order book, according to sources familiar with the offering, was oversubscribed roughly 20 times at the original price range. Lead underwriters Morgan Stanley, Citigroup, Barclays, and UBS Investment Bank ran a process that had the hallmarks less of a standard IPO and more of a controlled release of a scarce commodity. The company’s market capitalization at pricing exceeded $56 billion. Its 2025 revenue was $510 million.

Do the arithmetic, and you arrive at a trailing revenue multiple north of 100 times — the kind of valuation that demands either a ferociously compelling growth narrative or a willingness to suspend financial gravity altogether. Cerebras is making the case for the former. The market, for now, appears persuaded.

From a Garage Bet to a Dinner-Plate Chip: The Cerebras Origin Story

To understand why any of this matters, it helps to go back to April 2016, when Andrew Feldman, a serial entrepreneur who had previously sold a chip company to AMD, co-founded Cerebras Systems in Sunnyvale with a team of computer architects and AI researchers. The founding insight was simple to articulate and fiendishly difficult to execute: the central bottleneck in AI computation was not raw processing power but memory bandwidth. Graphics processing units, the Nvidia chips that power virtually every major AI workload in existence, are small silicon dies. Data must constantly travel between the GPU’s on-chip cache, external high-bandwidth memory, and network interconnects linking dozens or hundreds of GPUs together. Each hop consumes energy, introduces latency, and creates coordination overhead that compounds at scale.

Cerebras proposed eliminating those hops entirely by manufacturing a chip the size of an entire silicon wafer — a single monolithic die containing everything a neural network could need, on one continuous piece of silicon. The company calls it the Wafer Scale Engine. The current generation, the WSE-3, is fabricated on TSMC’s 5-nanometer process node and measures 46,225 square millimetres — making it 57 times larger than Nvidia’s H100 GPU by surface area. It packs 4 trillion transistors, 900,000 AI-optimized cores, and 44 gigabytes of on-chip SRAM with a memory bandwidth of 21 petabytes per second. By keeping all that memory directly on the wafer, Cerebras achieves bandwidth that the company claims is orders of magnitude higher than competing GPU-based architectures.

The practical implication, particularly for AI inference — the task of running a trained model to generate responses, code, or analysis — is speed. Cerebras claims its systems deliver inference up to 15 times faster than leading GPU-based solutions on leading open-source models. CEO Andrew Feldman has been characteristically blunt about what that means for competitive dynamics. “Obviously,” he told Yahoo Finance earlier this year, “[Nvidia] didn’t want to lose the fast inference business at OpenAI, and we took that from them.”

It is a remarkable claim, backed by a remarkable contract. But before exploring the OpenAI relationship, it is worth acknowledging that Cerebras’s path to this IPO was anything but linear.

The Rocky Road to Nasdaq: CFIUS, G42, and a Second Attempt

The Cerebras IPO story is, in many ways, two stories separated by an uncomfortable year in regulatory purgatory. The company first filed to go public in September 2024, only to withdraw its submission months later as regulators at the Committee on Foreign Investment in the United States (CFIUS) trained their scrutiny on the company’s relationship with G42, a UAE-based artificial intelligence conglomerate that was backed in part by Microsoft and had, at certain points, contributed the overwhelming majority of Cerebras’s revenue.

The optics were fraught. At the time of its initial filing, a single UAE-affiliated company — G42 — had accounted for 87% of Cerebras’s revenue in the first half of 2024. In an era of heightened concern about AI technology transfer to Gulf states with complicated relationships to both Washington and Beijing, CFIUS moved slowly. The review concluded in October 2025, after G42’s stake was restructured to non-voting shares, clearing the path for Cerebras to refile its S-1 with the SEC on April 17, 2026.

The second filing revealed a company that had not merely survived the delay but had fundamentally transformed its customer base. By 2025, G42’s share of Cerebras revenue had fallen from 87% to 24%. The Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), another UAE-affiliated institution, contributed 62%. Cerebras had also secured a binding deal with Amazon Web Services in March 2026, integrating its inference chips into AWS data centres, and had signed — most consequentially — a multi-year Master Relationship Agreement with OpenAI.

These developments did not eliminate concentration risk. Combined, UAE-affiliated entities still accounted for roughly 86% of 2025 revenue. But the strategic trajectory, and the credibility lent by the OpenAI relationship, proved sufficient to satisfy institutional investors and, eventually, regulators.

In a footnote worth savouring for its sheer drama, Bloomberg reported earlier this week that both Arm Holdings and SoftBank Group had approached Cerebras with acquisition overtures in the weeks before the IPO. Cerebras declined to comment. The company chose independence — and, at $56 billion, it is easy to see why.

The $20 Billion OpenAI Deal: Circular Economics and Strategic Validation

The centerpiece of the Cerebras investment thesis — and its most complex structural element — is the relationship with OpenAI. In January 2026, the two companies announced a deal worth more than $20 billion, under which OpenAI will consume 750 megawatts of Cerebras computing capacity, potentially expandable to 2 gigawatts. Cerebras supplies OpenAI with cloud-based computing power to operate an AI-assisted coding tool, making Cerebras the infrastructure layer beneath one of OpenAI’s most commercially important products.

The arrangement has an ingenious and somewhat vertiginous circularity. Cerebras is granting OpenAI warrants worth up to 10% of the company — approximately $5 billion at the IPO midpoint, representing roughly half the gross profit Cerebras stands to make on the deal, according to Financial Times calculations. It is architecturally similar to the circular arrangement OpenAI struck with Advanced Micro Devices, whose shares tripled following that announcement. For Cerebras, the warrant structure aligns OpenAI’s financial interests with Cerebras’s market capitalisation while simultaneously providing the kind of tier-one customer validation that transforms a niche chip company into a credible platform challenger.

There is also a historical curiosity worth noting. Court testimony in Elon Musk’s lawsuit against OpenAI revealed that in 2017, OpenAI considered merging with Cerebras, with Musk said to have been open to such a deal. OpenAI co-founder Greg Brockman stated in court that Cerebras’s planned chips represented “the compute we thought we were going to need.” A decade later, that assessment appears vindicated by contract.

WSE-3 vs. Nvidia: The Architecture Battle at the Heart of AI Infrastructure

To evaluate the Cerebras IPO investment case, one must grapple seriously with the technology differentiation. The artificial intelligence chip market is, in 2026, functionally a Nvidia hegemony. Nvidia’s quarterly revenue runs at approximately $51 billion — a figure that dwarfs Cerebras’s entire annual revenue by a factor of roughly 100. The CUDA software ecosystem, Nvidia’s parallel computing platform, has accumulated 15 years of developer familiarity, optimised libraries, and institutional inertia that represent perhaps the most formidable moat in modern technology.

Cerebras’s challenge to this dominance is narrow, deliberate, and — on the evidence — commercially real. Rather than attempting to compete across the full AI compute stack (training, fine-tuning, inference), Cerebras has concentrated its pitch on inference at ultra-low latency. The reasoning is architectural: inference tasks tend to be memory-bandwidth-constrained rather than compute-constrained. When a language model generates a response token by token, it must repeatedly load model weights from memory. On a GPU cluster, this means traversing the memory hierarchy — HBM, NVLink, InfiniBand — thousands of times per second. The WSE-3’s 44GB of on-chip SRAM, directly accessible by 900,000 cores without off-chip traversal, eliminates that bottleneck almost entirely.

For workloads where speed of response is the primary commercial differentiator — customer-facing AI assistants, coding tools, real-time translation, medical triage — the 15x inference speed advantage Cerebras claims is not an incremental improvement. It is a category-defining capability.

The architecture is not, however, without vulnerabilities. Manufacturing a chip the size of a dinner plate on a single TSMC wafer means defect rates are inherently higher than for conventional die-sized chips. Cerebras has developed proprietary redundancy and yield-optimisation techniques, but scaling production to meet the OpenAI contract will test these systems at unprecedented volumes. The monolithic design also means that unlike modular GPU clusters, Cerebras systems cannot easily scale horizontally by simply adding more nodes; the architecture’s advantages are indivisible.

Nvidia, meanwhile, is not standing still. The company’s Vera Rubin heterogeneous rack architecture and its recently reported acquisition of inference specialist Groq for approximately $20 billion signal that Nvidia understands the inference bottleneck and is aggressively engineering solutions. The AI chip landscape of 2027 may look substantially different from 2026. Cerebras investors are, in effect, betting that the company can establish sufficient revenue scale, customer stickiness, and software maturity before Nvidia closes the performance gap.

Financials: Spectacular Growth, Complex Profitability

The Cerebras S-1 presents a financial profile that rewards careful reading. Headline figures are impressive: revenue grew from $24.6 million in 2022 to $78.7 million in 2023, $290.3 million in 2024, and $510 million in 2025 — a 76% year-over-year acceleration. The 2025 revenue comprised $358 million in hardware sales and $152 million in cloud and managed services, reflecting the company’s strategic pivot toward recurring cloud revenues that began several years ago.

Profitability figures require more nuanced interpretation. Cerebras reported GAAP net income of $87.9 million for 2025 — a dramatic reversal from the $484.8 million GAAP loss in 2024. The reality, however, is that this headline profit was substantially manufactured by a one-time, non-cash accounting gain of approximately $363.3 million from extinguishing a forward contract liability related to the G42 restructuring. Strip that out, and the underlying picture is of a company with widening non-GAAP operating losses of $75.7 million.

On a non-GAAP basis, Cerebras reported net income of approximately $237.8 million — a figure that multiple analysts have cited as reflecting a 47% net margin on $510 million of revenue. This is genuinely unusual for an IPO-stage technology company. CoreWeave, the GPU cloud provider that went public in March 2026 at a $23 billion valuation, was not profitable at a comparable scale. The margin, however, is somewhat inflated by the high concentration of UAE customers who may have received pricing terms that do not reflect arm’s-length commercial rates.

Cerebras Financial Snapshot (FY 2025)

Metric	2025	2024	YoY Change
Total Revenue	$510M	$290.3M	+76%
Hardware Revenue	$358M	$212M	+69%
Cloud & Services Revenue	$152M	$78.3M	+94%
GAAP Net Income / (Loss)	$87.9M	($484.8M)	—
Non-GAAP Net Income	$237.8M	—	—
Non-GAAP Operating Loss	($75.7M)	—	—

The IPO valuation — at $185 per share, implying a market cap above $56 billion on a fully diluted basis — represents a trailing revenue multiple that, depending on methodology, ranges from approximately 100 to 110 times. By any traditional semiconductor valuation framework, this is exceptional. By the standards of AI infrastructure companies with contracted hyper-scaler revenues and demonstrated growth trajectories, the institutional community appears willing to pay it.

The Competitive Landscape: Nvidia, AMD, and the Inference Arms Race

Cerebras is not the only company to have identified Nvidia’s inference bottleneck. The AI chip challenger landscape has broadened substantially since 2023:

Groq — now acquired by Nvidia in a deal reportedly valued at approximately $20 billion — built its Language Processing Unit architecture around a similar memory-bandwidth thesis. Its acquisition by Nvidia simultaneously validates the inference-speed market opportunity and removes one significant independent competitor.

AMD has made meaningful inroads with its MI300 series, which offers competitive memory bandwidth through stacked HBM configurations. AMD’s deal with OpenAI, announced in late 2025, injected strategic momentum and a stock price catalyst.

Google’s TPU infrastructure remains formidable for internal workloads, though it is not commercially available in the same way.

Custom silicon efforts from Microsoft (Maia), Amazon (Trainium/Inferentia), and Meta remain largely captive — serving those companies’ internal demand rather than the open market.

What distinguishes Cerebras is the combination of architectural extremity (wafer-scale is still unique in commercial deployment), demonstrated inference speed leadership, and a $20 billion contracted revenue pipeline with OpenAI that provides a backstop against demand uncertainty. The AWS partnership provides an additional distribution channel that transforms Cerebras from a direct-sale hardware company into something resembling an infrastructure platform.

None of this neutralises the fundamental Nvidia risk. But it meaningfully narrows the scenario in which Cerebras becomes an irrelevance.

CBRS Stock: The Investment Thesis and Its Honest Limits

For investors evaluating whether to participate in the Cerebras IPO or accumulate CBRS stock in after-market trading, the intellectual framework is straightforward — even if the answer is not.

The bull case rests on three pillars. First, the $20 billion OpenAI contract provides revenue visibility over a multi-year horizon that few IPO-stage companies can offer; 750 megawatts of contracted compute at commercial cloud rates represents a significant revenue floor. Second, the AWS partnership opens an enterprise distribution channel that could systematically broaden the customer base beyond UAE-affiliated entities — the single most important de-risking factor the market wanted to see. Third, the inference-speed advantage, if it persists through competitive responses from Nvidia and others, positions Cerebras as a structurally differentiated supplier in the fastest-growing segment of AI infrastructure.

The bear case is equally coherent. Customer concentration remains extreme: even with the OpenAI deal, the near-term revenue base is dominated by two or three relationships, any one of which could prove unstable. The underlying operating business was loss-making on a non-GAAP basis in 2025, meaning the profitability narrative depends heavily on achieving scale that the company has not yet demonstrated. Manufacturing risk at wafer scale is non-trivial; production disruptions at TSMC or yield deterioration could impair the OpenAI delivery timeline with severe contractual and reputational consequences. And Nvidia’s response — whether through Groq integration, Vera Rubin architecture advances, or pure pricing aggression — may prove more rapid than current market assumptions imply.

The valuation multiple also raises uncomfortable questions about what “success” must look like to justify the entry price. At $56 billion and growing revenues at 76% annually, Cerebras would need to sustain extraordinary growth and dramatically improve its unit economics over the next three to five years to produce compelling returns at IPO pricing. Prediction markets have been modestly more sanguine: a Polymarket contract placed the probability of a day-one market cap between $50 billion and $60 billion as the most likely outcome at 33%, with $60 to $70 billion at 25% — suggesting the broader market expected a meaningful first-day pop.

For retail investors, the conventional wisdom applies with particular force: IPOs of high-growth companies with extreme valuations are rarely cheapest on the first day of trading. The signal-to-noise ratio in the first weeks of post-IPO trading is poor, driven more by momentum and lock-up dynamics than fundamental reassessment. The considered view — as expressed by senior investment editors at publications including Kiplinger — is to wait for one or two quarterly earnings reports before sizing a significant position.

Sovereign AI, Geopolitics, and the Deeper Stakes

There is a broader framing for the Cerebras story that transcends quarterly earnings and valuation multiples. The company’s early revenues came predominantly from the Gulf, where UAE-affiliated institutions were building sovereign AI capabilities — large-scale inference and training infrastructure that nations wary of dependence on American hyperscalers sought to control domestically. This is not a peripheral market. It is, increasingly, the central geopolitical ambition of every mid-sized nation with the resources to pursue it.

Cerebras’s CS-3 systems, housing WSE-3 processors, are physically deployable on-premises — a critical capability for government customers who cannot or will not route sensitive workloads through US cloud providers. The company has been explicit that its sovereign AI addressable market extends across four continents. As the global AI infrastructure investment cycle accelerates — driven by the AI capital expenditure boom that has seen hyperscalers collectively commit hundreds of billions in annual data centre spending — the demand for differentiated, deployable, privacy-preserving AI infrastructure is substantial and growing.

The geopolitical dimension, however, cuts both ways. US export controls on advanced AI chips are an expanding and unpredictable policy instrument. The CFIUS process that delayed the original Cerebras IPO by more than a year illustrates the regulatory surface area that any company serving Gulf, Asian, or other geopolitically complex customers must navigate. Post-IPO, Cerebras will face ongoing compliance obligations and potential policy changes that could constrain its most important historical customer relationships.

Arm Holdings and SoftBank’s reported acquisition interest underscores how the wafer-scale architecture, particularly in inference, is now viewed as genuinely strategic rather than merely technically interesting. That Cerebras chose to remain independent — and is now public with a balance sheet strengthened by $5.55 billion in IPO proceeds — gives it the firepower to invest in manufacturing scale, software ecosystem development, and geographic expansion without the encumbrances of a corporate parent.

The Road Ahead: What the Next 18 Months Will Reveal

The Cerebras IPO is, in many respects, the opening movement of a longer and more complicated composition. The $5.55 billion in gross proceeds will fund manufacturing scale-up at TSMC, software and SDK development to reduce the friction of migrating workloads from GPU-based systems to WSE-3, and the international expansion that the sovereign AI opportunity demands.

Three data points will define the trajectory of CBRS stock in the near to medium term. First, the pace at which AWS and other enterprise channels generate revenue diversification away from UAE-concentrated customers. If the next two or three earnings reports show MBZUAI and G42 declining as a share of total revenue, the concentration discount should compress substantially. Second, the delivery trajectory of the OpenAI contract. A 750-megawatt compute deployment is an enormous logistical undertaking; any slippage or renegotiation would be seized upon by short sellers as evidence of execution risk. Third, the competitive response from Nvidia — specifically, whether Groq’s inference capabilities, once integrated into Nvidia’s data centre stack, offer enterprise customers a credible GPU-based alternative to Cerebras’s speed advantage.

The broader context matters too. The IPO market in 2026 is on the cusp of something arguably unprecedented. SpaceX and OpenAI are both reportedly preparing listings that could together raise a combined $135 billion — offerings so large that, by comparison, Cerebras’s $5.55 billion will seem almost modest. Anthropic’s IPO preparations are also reportedly advanced. This wave of marquee AI company listings will reset market expectations, competitive benchmarks, and institutional portfolio allocations in ways that are genuinely difficult to model.

Cerebras enters public markets at a moment of maximum AI infrastructure enthusiasm and, simultaneously, maximum competitive intensity. Its wafer-scale bet was heretical when it was conceived a decade ago. It is now vindicated by contracts worth tens of billions of dollars, endorsed by the world’s most prominent AI laboratory, and priced by the market at a valuation that would have seemed fantastical when Andrew Feldman first sketched out the WSE concept on a whiteboard.

Whether that price proves prophetic or premature will depend on Cerebras’s ability to execute at a scale and speed that the semiconductor industry has rarely seen. What is not in doubt is that the company has already done the hardest thing: it has made the world take the dinner-plate chip seriously.

Discover more from The Economy

Subscribe to get the latest posts sent to your email.

Up Next

Malaysia Navigates a 5.4% Q1 Expansion as Global Clouds Gather

Don't Miss

Walmart Corporate Layoffs 2026: 1,000 Tech Jobs Cut in Major AI Restructuring

Click to comment

Analysis

Strait of Hormuz 2026: Why Markets Still Don’t Trust It’s Open

Published

5 hours ago

July 11, 2026

Abdul Rahman

If you’ve followed headlines about the Strait of Hormuz over the past several months, you’d be forgiven for losing track of whether it’s actually open. That confusion isn’t a media failure — it genuinely has opened, closed, and reopened multiple times since the conflict began, and the pattern itself is the real story markets need to understand, far more than any single day’s price move.

A Timeline That Explains the Market’s Persistent Skepticism

The crisis began February 28, 2026, when US and Israeli military operations against Iran triggered Iranian retaliation, including drone, ballistic missile, and small-boat attacks on vessels attempting to transit the Strait (Brookings). By March 4, Iranian forces formally declared the Strait “closed.” Insurance for transiting vessels became unavailable or prohibitively expensive, and seafarers largely refused the journey — meaning the Strait was effectively shut even without a formal blockade in the technical sense (Brookings).

What followed was a genuinely chaotic sequence that explains why traders remain reluctant to fully price in a resolution even now. On April 9, there was no sign an earlier agreement to lift the blockade was actually being implemented — ships were once again prevented from passing. Abu Dhabi National Oil Company’s CEO confirmed the Strait remained closed despite an announced ceasefire, noting 230 loaded oil tankers were waiting inside the Gulf (Wikipedia — 2026 Strait of Hormuz crisis). On April 17, Iran’s foreign minister announced the Strait was open to all shipping — oil prices dropped 11% immediately following the announcement. The very next day, April 18, Iran closed it again, citing the US refusal to lift its own naval blockade in response.

Even the June 17 memorandum of understanding between Trump and Iranian President Masoud Pezeshkian to formally end the war and the blockades didn’t hold cleanly: on June 20, Iran said it had closed the Strait again, citing continued Israeli strikes in southern Lebanon as a violation of the broader ceasefire agreement — a claim the US military denied (Wikipedia). By June 27, the US Navy’s Joint Maritime Information Center announced a widened shipping route through the Strait near Oman, an action explicitly framed as challenging Iran’s control over the waterway rather than a clean bilateral resolution.

Why This Chokepoint Matters More Than Any Other Piece of Global Infrastructure

Approximately 20 million barrels of oil per day move through the Strait of Hormuz — roughly 20% of global seaborne oil trade and about 27% of the world’s maritime crude oil and petroleum product trade combined (Congressional Research Service). At its narrowest point, the Strait is just 33-34 kilometers wide, split into two unidirectional two-mile-wide shipping lanes separated by a two-mile buffer zone sitting entirely within Iranian and Omani territorial waters (Congressional Research Service).

Critically, no rerouting option exists that can replace this volume at comparable cost. An extended full closure would remove 17-21 million barrels from daily global supply against total world consumption of roughly 100 million barrels per day — a supply shock with no readily available substitute (Ziro Market).

The Damage Already Done, Even With Partial Reopening

The International Energy Agency characterized the disruption as the largest supply disruption in the history of the global oil market (Wikipedia — Economic impact of the 2026 Iran war). At peak conflict intensity in February-March 2026, Brent crude surged well above $120 per barrel. As ceasefire talks progressed through May and June, prices retreated significantly — falling to around $95-100 per barrel by early June, and briefly dipping to $78.24 per barrel by mid-June, the lowest level since March 3, before the framework agreement was formally signed (Al Jazeera).

But the ripple effects extend well beyond crude oil pricing. The Strait closure disrupted roughly 45% of global sulfur supply — critical for fertilizer production, copper industry metal leaching, and sulfuric acid manufacturing — and constrained helium supply, a commodity essential to semiconductor manufacturing (Wikipedia — Economic impact). Shipping companies including Maersk, CMA CGM, and Hapag-Lloyd suspended transits through the Strait and related routes like the Red Sea entirely, forcing rerouting around the Cape of Good Hope that added two to three weeks to journey times and increased per-shipment costs by 30-50% (Ziro Market).

Europe’s Quieter But Deeper Crisis

While oil price headlines dominated coverage, Europe faced an arguably more severe parallel crisis through the suspension of Qatari liquefied natural gas exports combined with the Strait closure — hitting at the worst possible moment, with European gas storage sitting at just 30% capacity following a harsh 2025-2026 winter. Dutch TTF gas benchmarks nearly doubled to over €60/MWh by mid-March (Wikipedia — Economic impact).

The European Central Bank responded by postponing planned interest rate reductions on March 19, simultaneously raising its 2026 inflation forecast and cutting GDP growth projections, with UK inflation specifically projected to breach 5% during 2026. Chemical and steel manufacturers across the UK and EU imposed surcharges of up to 30% to offset surging electricity costs, and the ECB explicitly warned that a prolonged conflict risked pushing major energy-dependent economies, including Germany and Italy, into technical recession by year-end.

Why OPEC+ Couldn’t Simply Fill the Gap

A natural question is why Saudi Arabia and the UAE — the two largest Gulf Cooperation Council producers with meaningful spare capacity — didn’t simply increase output to compensate. The answer is logistical rather than a lack of willingness: the Strait closure itself limited their ability to actually export any increased production volumes, even when pumping more oil, because the export bottleneck was the same chokepoint causing the broader crisis (Ziro Market). Total OPEC country production fell more than 30% since the start of the war, and the region’s spare capacity — the traditional shock absorber for global oil markets — proved largely irrelevant when the actual export route itself was under attack (Brookings).

US shale producers, meanwhile, responded more slowly to the price signal than historical patterns would predict. Rig counts stayed largely steady through April 2026, though well-completion activity in the Permian Basin did rise roughly 20% over several weeks as previously drilled wells came into production — still below pre-pandemic activity levels overall (Brookings).

The Market Is Still Pricing a Discount for Uncertainty, and Analysts Say That’s Correct

Vandana Hari, founder of Singapore-based Vanda Insights, offered perhaps the most useful framing for understanding current market behavior: crude’s slide following the memorandum of understanding is “entirely sentiment-driven,” with markets front-running the prospective reopening and likely pricing in a best-case scenario for normalized flows — meaning potential hiccups, from logistics to renewed geopolitical tensions, aren’t being adequately factored in (Al Jazeera).

Given the actual track record — multiple announced reopenings followed by renewed closures throughout April and June — that skepticism looks well-founded rather than excessive.

What This Means for Businesses and Investors Going Forward

For companies with Gulf-dependent supply chains: Treat any single reopening announcement as provisional rather than a genuine all-clear, given the pattern of reversals throughout the spring. Maintaining rerouting contingency plans and insurance flexibility remains prudent even after formal ceasefire signings.

For inflation-sensitive investors and central bank watchers: The relationship Ziro Market’s analysis highlights is worth internalizing directly: whether oil settles near $80-85 (supporting rate cuts, lower CPI, stronger oil-importing currencies) or spikes back toward $120 (elevated inflation, delayed rate cuts) functions as a genuine macro regime switch — not a marginal input, but potentially the single largest swing factor for 2026 global monetary policy.

For commodity-exposed sectors beyond energy: The sulfur, fertilizer, and helium supply disruptions are underappreciated second-order effects that specifically hit agriculture and semiconductor manufacturing — sectors not typically associated with Middle East conflict risk but directly exposed through this specific chokepoint.

The Bottom Line

The Strait of Hormuz crisis of 2026 has been less a single supply shock than a recurring pattern of partial resolutions and renewed disruptions, and that pattern itself is the most important thing for markets and businesses to understand going forward. Prices have retreated substantially from their conflict-peak highs, and the June 17 memorandum of understanding represents genuine diplomatic progress. But given that the Strait has been declared “open” and then closed again multiple times within the same several-week windows, treating the current relative calm as a durable resolution — rather than the latest phase in an ongoing negotiation — would be a mistake that both markets and policymakers seem determined not to repeat.

Discover more from The Economy

Subscribe to get the latest posts sent to your email.

AI

AI Capex Bubble 2026: The Hidden $662B Debt Nobody Reports

Published

5 hours ago

July 11, 2026

Abdul Rahman

Every earnings season now brings a fresh wave of headlines about hyperscaler AI capital expenditure hitting a new record. The “big four” — Amazon, Microsoft, Alphabet, and Meta — are on track to spend roughly $725 billion combined in 2026, a 77% jump from the $410 billion deployed in 2025 (UnboxFuture). That number gets reported constantly. What almost nobody is reporting with the same prominence is a separate figure that may matter more: roughly $662 billion in data center lease commitments that hyperscalers have already signed but not yet begun — obligations that currently sit entirely off balance sheet.

Why the Off-Balance-Sheet Number Changes the Whole Picture

Under GAAP accounting rules governing when a lease “commences,” these signed-but-not-started commitments don’t appear in the capital expenditure figures analysts and investors typically scrutinize when assessing hyperscaler financial health. According to reporting citing Moody’s early-2026 analysis, this shadow liability is larger than the combined on-balance-sheet debt of the same companies (Anomaly Investments).

That detail matters enormously for one specific argument AI infrastructure bulls have relied on: the claim that this buildout is being conservatively self-funded from operating cash flow rather than risky leverage. Once the full picture of committed-but-unrecognized obligations is accounted for, that defense becomes much harder to sustain.

The Debt Is Already Showing Up, Not Just Theoretical

This isn’t a purely hypothetical concern about future liabilities. Big tech companies have already issued more than $100 billion of bonds in 2026 specifically to help fund AI capital expenditure, and investors have responded by demanding record levels of protection against potential defaults through credit default swaps — essentially insurance policies against bond default (IEEE ComSoc).

Individual company examples illustrate the shift toward leverage: Oracle issued an $18 billion bond specifically tied to its data center expansion; CoreWeave secured a $2.6 billion loan alongside a $1.75 billion bond package; and OpenAI and Oracle reportedly entered into a $100 billion vendor financing arrangement (Anomaly Investments). At Amazon specifically, capital expenditure over the trailing twelve months has reached $151 billion — a figure that now exceeds the company’s entire operating cash flow, pushing free cash flow into negative territory.

The Depreciation Assumption Almost No Coverage Questions

Here’s an angle genuinely underexplored across most financial media: the depreciation schedules hyperscalers use for AI hardware assume a five-to-six-year useful life. But given how rapidly GPU generations are turning over and how intensively AI workloads are pushing hardware utilization, critics argue the real economic life of this equipment is closer to two to three years. That gap between assumed and actual depreciation is estimated to understate true asset depletion by roughly $176 billion between 2026 and 2028 alone — a figure that grows as accelerating token consumption pushes hardware utilization beyond the assumptions built into current depreciation schedules (Anomaly Investments).

Layered on top of that is the energy cost curve: running the current roughly 30-gigawatt installed base of AI infrastructure costs approximately $27 billion annually today, but that figure is projected to climb to between $45 and $90 billion per year as capacity scales toward 2029 — and crucially, these are first charges against revenue, not optional or deferrable costs.

The Revenue Gap: Who’s Actually Paying for All This?

The most commonly cited justification for the capex surge is that the pure-play AI vendors — OpenAI, Anthropic, and others — represent a massive and rapidly growing revenue opportunity. The reality is more nuanced. OpenAI’s roughly $20 billion annualized revenue run rate, while genuinely impressive for a company with barely any consumer products three years ago, represents only about 3% of projected 2026 hyperscaler capex. Anthropic’s roughly $9 billion run rate, despite showing 9x year-over-year growth, occupies a similarly small share. The entire cohort of pure-play AI vendors combined — including Cohere, Mistral, Perplexity, and others — likely accounts for less than $35 billion in projected combined 2026 revenue against a hyperscaler capex figure exceeding $700 billion (Futurum Group).

That gap is the crux of the bubble debate: hyperscalers are betting the infrastructure will ultimately serve enterprise adoption and their own AI services broadly, not just third-party AI vendor revenue — but that bet requires enterprise AI monetization to arrive at a scale that, as of mid-2026, remains largely unproven outside of code generation and basic customer service automation.

The Skeptic’s Case, From Inside Goldman Sachs Itself

The most prominent voice of institutional skepticism doesn’t come from an outside critic — it comes from within Goldman Sachs itself. Jim Covello, the bank’s Head of Global Equity Research, has consistently argued the economics of the generative AI transition are fundamentally flawed, stating in mid-2026 that the industry has moved “further away” from justifying the scale of capital expenditure compared to two years prior (UnboxFuture). Covello has specifically flagged circular capital flows between cloud providers and AI startups — where hyperscalers invest in AI companies that then spend that same capital purchasing compute from those same hyperscalers — as a red flag reminiscent of vendor financing patterns seen in the dot-com era.

The valuation comparison to that era is explicit and increasingly common among strategists: US technology and AI equities carry EV/EBITDA multiples near 25x, close to historical extremes and above the telecom valuations that preceded the 2000 dot-com peak. More specifically, capex is currently expanding roughly 46 percentage points faster than revenue growth — a gap that exceeds the 32-point divergence observed during the 2001 telecom excess cycle (Allianz Research). Separately, Bank of America strategists have pointed out that AI stock concentration has reached levels matching prior bubble peaks, with the “AI Big 10” (Nvidia, Microsoft, Alphabet, Amazon, Meta, Apple, Tesla, Broadcom, Micron, and AMD) now making up 41% of the S&P 500 — comparable to the concentration of tech and telecom stocks during the actual dot-com bubble (Yahoo Finance).

The Bull Case Isn’t Naive Either

It would be inaccurate to frame this purely as informed skeptics versus blind enthusiasm. Goldman Sachs’ own broader research (distinct from Covello’s individual view) models roughly $7.6 trillion in cumulative AI capital expenditure between 2026 and 2031, built on the expectation that token consumption will increase 24-fold by 2030, driven largely by enterprise AI agents becoming embedded in production workflows rather than remaining experimental (Sesame Disk / Goldman commentary). Microsoft has disclosed an $80 billion backlog of Azure orders it currently cannot fulfill due to power constraints — genuine evidence that demand, at least for existing capacity, is outpacing even the current aggressive build-out pace (Futurum Group).

Leverage levels also remain more conservative than headlines suggest in absolute terms: the top five US capex providers reported a combined $385 billion in debt at the end of 2025, with leverage ratios still roughly 20% below the “high spender” cohort from the 2000 dot-com peak, according to Allianz Research analysis — meaning rising debt levels are a trend worth monitoring closely, not yet an acute crisis.

What Happens If the Bubble Skeptics Are Right

Historical infrastructure cycles offer a specific and somewhat counterintuitive lesson: the investors who fund the initial frenzied build-out phase rarely capture the long-term rewards. If the AI capex cycle follows the pattern of the 1998-2001 fiber optic buildout, hyperscalers may eventually be forced to write down the value of data centers and GPUs purchased at today’s prices and utilization assumptions. But that collapse in computing costs, paradoxically, could pave the way for a new generation of leaner, genuinely profitable software companies to build on top of the resulting cheap, overbuilt infrastructure — much as fiber-optic overbuild eventually enabled the 2000s streaming and cloud computing boom, even after the original telecom investors were wiped out.

What This Means for Investors and Businesses

For equity investors, the practical signal to watch isn’t the headline capex number — it’s the widening gap between capex growth and revenue growth, and whether that gap begins narrowing through 2027 as enterprise adoption either accelerates or disappoints. For businesses evaluating AI vendor relationships, the circular-financing pattern flagged by Covello is worth diligence: understanding whether an AI vendor’s revenue depends partly on capital originally supplied by the same hyperscaler providing its compute is a legitimate red flag for assessing that vendor’s underlying financial independence. For fixed-income investors, the rising credit default swap pricing on hyperscaler-linked debt is itself a market signal worth tracking as an early indicator of shifting sentiment, independent of equity price action.

The Bottom Line

The AI infrastructure buildout genuinely is the largest corporate capital expenditure cycle in recorded history, and it’s happening for real, defensible reasons tied to a genuine technology shift. But the debate over whether it constitutes a bubble isn’t really about whether AI technology is useful — it’s about whether the timing of returns can keep pace with public equity markets’ patience, and whether the $662 billion in off-balance-sheet lease commitments, aggressive depreciation assumptions, and circular vendor financing arrangements represent manageable financial engineering or the early architecture of a genuinely serious correction. Both cases have real evidence behind them. What’s clear is that the headline capex figure everyone quotes is no longer the most important number in this story.

Discover more from The Economy

Subscribe to get the latest posts sent to your email.

Markets & Finance

Gold Overtakes US Treasuries in Reserves: What It Means

Published

5 hours ago

July 11, 2026

Abdul Rahman

Most gold coverage in 2026 has fixated on the price chart — the spectacular run from roughly $2,633 an ounce at the start of the year to fresh record highs above $5,400 by mid-year (Intellectia). That’s a legitimate story. But it’s not the most important one. The more consequential shift is structural, not seasonal: gold has overtaken US Treasuries as the largest share of global central bank reserves for the first time in three decades (BlackRock).

That’s not a headline about a commodity rally. It’s a headline about the architecture of the global monetary system quietly shifting under everyone’s feet.

The Trigger Most Coverage Undersells

The pivotal moment behind this shift traces back to 2022, when roughly $300 billion of Russian central bank foreign exchange reserves were frozen as part of international sanctions following the invasion of Ukraine (ISA Bullion). For reserve managers around the world — not just in Russia — that event functioned as a wake-up call: dollar-denominated assets held abroad are not unconditionally safe from geopolitical sanctions risk. Gold, by contrast, carries no counterparty risk; nobody can freeze a gold bar sitting in a country’s own vault.

That single realization has reshaped reserve management strategy globally. Central bank gold purchases averaged 225 tonnes per quarter between 2021 and 2025 — roughly double the pace seen from 2016 to 2020 (J.P. Morgan Global Research). BRICS+ nations now hold 17.4% of global gold reserves, up sharply from just 11.2% in 2019 (ISA Bullion).

Who’s Actually Buying, and Why the List Matters

Poland has been the standout accumulator, adding 20.2 tonnes in February 2026 alone, another 11.2 tonnes in March, and 14 tonnes in April — extending a rapid buildup that has added more than 360 tonnes to its reserves since 2023 (BestBrokers). China’s central bank maintained consecutive monthly gold purchases for 19 straight months through May 2026, even though much of this buying goes officially unreported to the IMF — analysts widely believe the People’s Bank of China continues accumulating gold “off the books” (ISA Bullion).

China’s motivation appears explicitly strategic rather than opportunistic. Chinese net gold imports jumped to 317 tonnes in the first quarter of 2026 alone — nearly triple the prior quarter — while the People’s Bank of China’s own reported purchases accelerated from roughly one tonne per month through February to eight tonnes in April (J.P. Morgan Global Research). J.P. Morgan’s own analysts frame this as part of a long-term Chinese project to build gold reserves as a foundation for establishing the renminbi as a credible alternative reserve currency.

A World Gold Council survey found a striking 95% of central banks expect to increase their gold holdings in 2026, up from 81% in 2024 and just 52% in 2021 — a trajectory showing accelerating, not plateauing, institutional conviction (BlackRock).

The Part of the Story Most Coverage Misses: Not Everyone Is Buying

Here’s an angle that gets consistently underplayed: this isn’t a uniform global stampede into gold. Several countries, including Singapore, Jordan, Mexico, and the Solomon Islands, actually reduced their gold reserves in 2025 — Singapore in particular emerged as a notable seller, likely driven by portfolio rebalancing decisions and a desire to realize gains after gold’s historic surge, rather than any lack of confidence in the metal (BestBrokers). Germany, for its part, has reduced its gold holdings every year since at least 2002, though its 2024 sale of just 1.1 tonnes was the smallest annual reduction on record.

This nuance matters for anyone trying to build a genuinely accurate picture: the de-dollarization and gold-accumulation trend is heavily concentrated among specific emerging-market and non-aligned economies — not a universal central bank consensus. Understanding which countries are buying and why is more analytically useful than simply citing an aggregate global purchasing figure.

Where Forecasts Diverge — And Why the Spread Is So Wide

Institutional price forecasts for gold currently show a genuinely unusual spread. J.P. Morgan projects gold reaching $6,000 an ounce by the end of 2026, and potentially $6,300 by the end of 2027 (J.P. Morgan Global Research). Morgan Stanley’s more conservative 2026 forecast sits at $4,400 an ounce (Morgan Stanley), while State Street projects a range of $4,750 to $5,500, and DWS targets $5,400 by mid-2027 (Discovery Alert).

A spread exceeding $1,500 per ounce between the most bullish and most conservative institutional forecasts reflects a genuine, unresolved analytical disagreement — not just differing house styles. The bull case rests on the idea that central bank reserve diversification represents a structural, policy-level shift rather than opportunistic market timing, making it fundamentally different from prior gold cycles driven mainly by retail or momentum investors. The more cautious case notes that gold’s roughly 245% rally from September 2022 to January 2026 is the largest percentage advance in modern gold market history — and historically, rallies of that magnitude have eventually triggered significant, multi-year corrections (Discovery Alert).

The Under-Discussed New Buyer: Stablecoin Issuers

One of the least-covered developments in this entire gold story is the emergence of stablecoin issuers as a genuinely new category of gold demand. As crypto markets have matured, some stablecoin issuers have begun holding gold as part of their reserve backing strategy — a development BlackRock specifically flags as part of the “early stages” of a new demand wave that also includes central banks and the broader AI infrastructure buildout’s effect on institutional portfolio hedging behavior (BlackRock).

What This Means for Different Audiences

For everyday investors: Gold ETPs still make up only about 0.17% of total US private financial assets, remaining well below prior peaks seen in the early 2010s, while private wealth gold allocations globally sit roughly 50% below levels seen a decade ago (BlackRock). That suggests meaningful room for incremental Western retail and institutional demand to grow, even after the current rally, if the structural de-dollarization narrative continues to gain mainstream acceptance.

For businesses managing currency exposure: The scale and persistence of central bank gold buying is one of several signals (alongside Fed communication policy changes and fiscal deficit concerns) suggesting continued structural pressure on the US dollar’s long-term reserve currency dominance — a trend worth factoring into multi-year currency hedging strategies rather than treating as a short-term news cycle.

For portfolio allocators: The unusually wide spread between institutional forecasts is itself useful information — it suggests treating any single gold price target as a scenario input rather than a confident base case, and sizing gold allocations based on its role as a portfolio diversifier and inflation/geopolitical hedge rather than as a directional price bet.

The Bottom Line

The gold price chart is the story most people are watching. The reserve-composition shift is the story that actually matters for the long-term structure of global finance. Gold surpassing US Treasuries as the largest share of central bank reserves for the first time since 1996 is a genuinely historic threshold — one triggered specifically by the 2022 Russian asset freeze and now sustained by a broad, if uneven, cohort of emerging-market central banks pursuing deliberate de-dollarization strategies. Whether the price keeps climbing toward J.P. Morgan’s $6,000 target or cools toward Morgan Stanley’s more conservative range matters less, in the long run, than the structural fact that the world’s reserve managers have permanently changed how they think about gold’s role in the global financial system.