DISCLAIMER — Not financial advice. Educational content only, not an offer or solicitation to buy or sell any security. Biotech and small/mid-cap stocks are highly speculative and volatile and can result in a partial or total loss of capital. Do your own research and consult a licensed advisor where appropriate.
Tools I Use
Real-Time Market Scanner
Fast movers • Alerts • Momentum
10% OFF with SAVE10
Recommended Platform
Professional Trading Workflow
Charts • Watchlists • Multi-broker tools
Built for active traders
Merlintrader Trading Pub
Biotech catalyst, news and analysis PDUFA tracker
Merlintrader Trading Pub
Biotech catalyst, news and analysis PDUFA tracker
$WULF, $AVGO, $MU: AI Infrastructure Back in Focus as Data Centers Replace the Old Crypto-Mining Story
The July 6, 2026 AI-infrastructure tape is not only about another chip rebound. The more important shift is structural: compute demand is moving backward through the stack, from models to data centers, from data centers to power, from power to custom silicon, and from accelerators to memory and storage. TeraWulf’s Anthropic lease puts the former crypto-mining-to-AI-infrastructure trade back under the spotlight. Broadcom’s extended Apple relationship reinforces the custom-chip and supply-chain-lock-in layer. Micron’s recent results and new automotive supply agreements show why memory is no longer a boring cyclical afterthought inside AI infrastructure.
Lead catalyst$19B leaseTeraWulf announced a 20-year Anthropic lease at the Justified Data campus.
Critical IT load~401 MWCapacity is expected to be delivered in phases from late 2027 to early 2028.
Abernathy sale~$530MTeraWulf agreed to sell its 50.1% interest in the Abernathy JV to a Fluidstack-led group.
Broadcom / AppleThrough 2031Broadcom and Apple expanded their custom-chip relationship through 2031.
Micron Q3 revenue$41.46BFiscal Q3 2026 revenue surged from $9.30B in the prior-year period.
Micron Q4 guide$50B ± $1BThe company guided to another step-up in fiscal Q4 revenue.
Memory pricingTight supplyDRAM and NAND supply remains structurally tight under AI demand.
Core riskExecutionData-center leases, chip supply and memory pricing all require delivery discipline.
Article map
This report reads the day’s AI-infrastructure news as one connected chain: TeraWulf as the power-and-campus trade, Broadcom as the custom silicon and customer lock-in trade, and Micron as the memory-and-storage bottleneck trade.
Executive summary: the AI trade is moving from “who owns the model?” to “who owns the infrastructure?”
The most important market story today is not simply that AI-related stocks are moving again. The more durable question is whether the AI trade is changing shape. In 2023 and 2024, the market mostly chased model owners, GPU leaders and hyperscaler capex. By mid-2026, the trade has become more complicated. Investors are now asking who controls the power, the land, the interconnection, the data-center campus, the custom silicon roadmap, the memory supply and the storage layer needed to keep AI models running at scale.
That is why the combination of $WULF, $AVGO and $MU is useful. These are not the same kind of company. TeraWulf is a high-volatility infrastructure transition story. Broadcom is a large-cap custom silicon and infrastructure software compounder. Micron is a memory and storage leader riding one of the most extreme supply-demand cycles in modern semiconductor history. Together, they show how AI demand is moving down the stack from software excitement to physical capacity.
The lead catalyst is TeraWulf. On July 6, 2026, the company announced a 20-year lease agreement with Anthropic at the Justified Data campus in Hawesville, Kentucky. The lease is expected to generate approximately $19 billion of contracted revenue over the initial term and cover approximately 401 MW of critical IT load, with initial capacity expected in the second half of 2027 and full ramp expected by early 2028. At the same time, TeraWulf agreed to sell its 50.1% interest in the Abernathy Joint Venture to a Fluidstack-led investor group for approximately $530 million in aggregate consideration.
The TeraWulf announcement is important because it makes the post-crypto-mining infrastructure pivot visible in numbers large enough for the market to care. A former or legacy crypto-mining-linked infrastructure company can try to rebrand around AI. That is common. What matters is whether the company can sign long-term AI customers, secure power, deploy capital, deliver campuses and turn energy infrastructure into durable lease revenue. The Anthropic lease gives TeraWulf a more concrete AI-infrastructure narrative, but it also creates a new execution test.
Broadcom’s role in the same tape is different. Reuters reported that Broadcom and Apple expanded their partnership through 2031 to develop and supply custom chips. Apple has long relied on Broadcom for radio-frequency, Wi-Fi, Bluetooth and other networking semiconductor components, and analysts cited by Reuters said Apple accounts for roughly 20% of Broadcom’s annual revenue. The agreement sits on top of Broadcom’s broader custom silicon opportunity with hyperscalers and AI infrastructure customers.
Micron completes the chain. The AI boom is not only about accelerators. Accelerators need high-bandwidth memory, DRAM, NAND, SSDs and storage architecture. Micron’s fiscal Q3 2026 results were extreme: revenue of $41.46 billion, compared with $9.30 billion in the same quarter a year earlier; non-GAAP EPS of $25.11; and Q4 revenue guidance of $50.0 billion ± $1.0 billion. Micron also described memory as strategically important in the AI era and pointed to tight DRAM and NAND supply conditions.
The simplest framework: $WULF is the power-and-campus bottleneck, $AVGO is the custom silicon and strategic customer lock-in layer, and $MU is the memory and storage bottleneck. The AI trade is no longer just about models and GPUs. It is about whether the physical stack can keep up with demand.
Why this matters now: data centers are becoming the new strategic infrastructure
The AI market is entering a phase where software capability is constrained by physical deployment. Better models need more inference, more inference needs more compute, more compute needs more data centers, more data centers need power and cooling, and every part of that stack needs chips, memory, storage, networking and supply agreements. This is why AI infrastructure has become a market category in its own right.
The old shorthand was simple: buy the GPU winner. That was useful early, but it is no longer enough. A hyperscaler or AI lab cannot serve hundreds of millions of users simply by announcing a better model. It needs racks, power, land, permits, transformers, substations, custom silicon, networking, HBM, DRAM, NAND, SSDs, operations staff, financing and long-term customer contracts. The bottleneck keeps moving.
The TeraWulf news is important because it highlights one of the most severe bottlenecks: power-secured data-center capacity. AI companies do not only need generic space. They need large blocks of power, credible timelines, infrastructure capable of supporting high-performance computing workloads, and counterparties that can build and operate under long-term leases. A 401 MW campus is therefore not just a real-estate story. It is a power-and-compute availability story.
Broadcom’s news fits because custom silicon is becoming a way for major technology companies to reduce dependency, improve efficiency and lock in long-term supply. When Apple extends a chip relationship through 2031, the significance is not only one customer contract. It reflects the broader trend of customers turning silicon supply into strategic infrastructure.
Micron’s role fits because AI systems are hungry for memory bandwidth and storage. A data center full of accelerators becomes bottlenecked if memory supply, HBM packaging, DRAM availability or SSD capacity cannot keep pace. Memory is cyclical, but AI has made it strategically important. The question is whether this cycle is merely a super-cycle or a structural reset in how customers contract for supply.
This is also why the crypto-mining-to-AI-infrastructure pivot is getting attention. Crypto miners built or controlled energy-heavy sites and developed operational knowledge around power procurement, power markets, facility buildout and high-load computing. Some of those assets may be repurposed for AI and HPC. But the market must distinguish between companies that simply say “AI data center” and companies that secure credible customers, financing, capacity and delivery timelines.
$WULF — TeraWulf: from Bitcoin-mining infrastructure to long-duration AI campus revenue
TeraWulf Inc. is the most important stock in this specific theme today because its news is the clearest example of the old mining-infrastructure story being replaced by a data-center infrastructure story. The company describes itself as a developer, owner and operator of large-scale digital infrastructure for AI, high-performance computing and other advanced compute workloads. The market is now being asked to value TeraWulf less like a Bitcoin mining proxy and more like a power-secured AI infrastructure platform.
On July 6, TeraWulf announced a 20-year lease with Anthropic at the Justified Data campus in Hawesville, Kentucky. The lease is expected to generate approximately $19 billion of contracted revenue over the initial term. The campus is expected to accommodate about 401 MW of critical IT load and be developed in phases, with initial capacity expected during the second half of 2027 and full ramp to 401 MW by early 2028.
The company’s 8-K gives important additional detail. The lease is with Anthropic as tenant through TeraWulf subsidiary Raylan Data LLC as landlord. Delivery is expected in phases beginning in late 2027 and concluding in early 2028. Anthropic’s rent obligation begins when the relevant leased premises are delivered and continues for 20 years, with options to extend the term for up to an additional ten years through two successive five-year renewal options. TeraWulf also said Anthropic’s payment obligations are expected to be supported by investment-grade credit.
The second part of the announcement is the Abernathy transaction. TeraWulf agreed to sell its entire 50.1% ownership interest in the Abernathy Joint Venture to an investor group led by Fluidstack. The 8-K states aggregate consideration of approximately $530 million, payable in three installments: $250 million within 14 days after execution, $150 million on or before December 31, 2026, and approximately $130 million, subject to adjustments, on or before April 30, 2027.
This matters because the Abernathy sale is not just an asset sale. Strategically, TeraWulf is arguing that it can crystallize value from a developed JV investment and recycle capital into wholly owned AI infrastructure opportunities where it maintains direct ownership, customer relationships and operational control. In other words, management wants the market to see a capital-recycling model: originate or develop campuses, secure customers, monetize where useful, redeploy capital into higher-control infrastructure platforms.
Why the Anthropic lease changes the TeraWulf narrative
The Anthropic lease gives TeraWulf a long-duration customer story tied to one of the most visible AI labs in the world. For investors, this is more credible than a generic “we are moving into AI” statement. A named customer, a 20-year lease, a 401 MW critical IT load target and a $19 billion contracted revenue figure create a much more concrete framework.
Still, the deal does not eliminate risk. The first important risk is timing. Initial capacity is expected in the second half of 2027, and full ramp is expected by early 2028. That means investors are underwriting construction, delivery and commissioning risk over multiple years. The second risk is capital intensity. Large AI campuses require significant capital, equipment, power infrastructure, grid integration and customer-specific buildout. The third risk is execution quality. A lease becomes financially valuable only when capacity is delivered, accepted and producing rent.
The fourth risk is counterparty and credit structure. The company expects Anthropic’s payment obligations to be supported by investment-grade credit, which is a positive detail, but the market will still care about the exact financing structure, project economics, capex requirements and how much of the $19 billion lease revenue converts into durable free cash flow.
Why $WULF is attractive and dangerous at the same time
The attractive part is obvious: TeraWulf may be repositioning itself into one of the market’s most valued infrastructure categories. Power-secured AI campuses are scarce. If the company can repeatedly source power, sign major customers and deliver campuses, the market may start to value it as an AI infrastructure developer rather than a legacy crypto-mining proxy.
The dangerous part is equally obvious: the stock can move faster than the business can execute. A $19 billion headline is enormous, but it is a 20-year contracted revenue figure over an initial term, not immediate cash. The capacity is not all online today. The economics will depend on capital cost, financing, energy cost, operating efficiency, delivery schedule and contract details. In short, the headline is powerful; the execution burden is large.
For $WULF, the key question is no longer “can a miner pivot to AI?” The key question is “can TeraWulf become a repeatable power-secured AI data-center developer with credible customers, acceptable capital intensity and durable cash-flow conversion?”
$AVGO — Broadcom: custom silicon, Apple lock-in and the infrastructure-chip annuity
Broadcom Inc. is a very different story from TeraWulf. It is not a speculative infrastructure pivot. It is a large-cap semiconductor and infrastructure software company with deep customer relationships, scale and a long record of disciplined execution. In this report, Broadcom represents the custom silicon layer of AI infrastructure.
Reuters reported on July 6 that Broadcom and Apple agreed to expand their partnership through 2031 to develop and supply custom chips. Apple has historically relied on Broadcom for key components including radio-frequency chips used in iPhones, Wi-Fi and Bluetooth connectivity chips, and other networking semiconductors. Analysts cited by Reuters estimate Apple accounts for roughly 20% of Broadcom’s annual revenue.
At first glance, an Apple chip supply extension may look separate from the AI data-center story. It is not. Broadcom’s relevance lies in the broader trend: the world’s largest technology customers are locking in custom chip supply, advanced networking, connectivity and infrastructure silicon over multi-year horizons. Custom silicon has become strategic. Supply agreements are no longer just purchasing contracts; they are infrastructure planning tools.
Broadcom’s fiscal Q2 2026 results help explain why the market cares. The company reported revenue of $22.187 billion, up 48% from the prior-year period, GAAP net income of $9.310 billion, non-GAAP net income of $12.074 billion and adjusted EBITDA of $15.244 billion, or 69% of revenue. Broadcom also stated that Q2 semiconductor revenue from AI of $10.8 billion grew 143% year over year, driven by demand for custom AI accelerators and AI networking.
This is the Broadcom story the market is watching: a combination of long-duration customer relationships, custom silicon, AI networking, infrastructure software and operating leverage. The Apple agreement reinforces customer durability. The AI semiconductor revenue reinforces growth. The EBITDA margin reinforces financial quality.
Why Broadcom matters in the AI infrastructure stack
AI infrastructure is not only a GPU problem. High-performance AI systems require custom accelerators, networking silicon, switching, connectivity, radio-frequency components, storage connectivity and software layers. Broadcom’s value comes from being embedded across multiple parts of that infrastructure stack.
The rise of custom ASICs is especially important. Hyperscalers and large technology companies increasingly want chips tailored to their own workloads. They may still buy GPUs, but they also want internal or custom architectures that optimize performance, cost, power and supply control. Broadcom is one of the most important companies in this custom silicon value chain.
The Apple extension through 2031 also addresses a different concern: customer concentration and replacement risk. Apple has been developing more in-house silicon over time. Any long-term Broadcom investor has to monitor whether Apple internalizes components that Broadcom previously supplied. A five-year extension through 2031 does not eliminate all long-term customer concentration risk, but it eases immediate concerns around the relationship.
What could go wrong for $AVGO
The first risk is valuation and expectation. Broadcom is not a hidden small-cap. It is a large, widely followed AI infrastructure winner. Strong companies can still underperform if expectations become too aggressive.
The second risk is customer concentration. Apple is a major customer. Hyperscaler and AI customer ramps can be lumpy. A delay, design change or internalization decision by a major customer can affect revenue quality and sentiment.
The third risk is cycle exposure. Even a high-quality custom silicon company is not immune to semiconductor cycles, inventory correction, supply-chain friction or capex digestion by customers. Broadcom’s AI growth is powerful, but the market will continue to ask how durable the ramp is and how much has already been priced in.
$MU — Micron: memory is no longer the invisible layer of AI infrastructure
Micron Technology is the memory-and-storage leg of this theme. The old market habit was to treat memory as a brutally cyclical commodity business: DRAM prices rise, companies over-earn, supply expands, prices fall, and the cycle repeats. That framework is still useful, but it is no longer enough. AI has turned memory into a strategic bottleneck.
Micron’s fiscal Q3 2026 results show the scale of the current cycle. The company reported revenue of $41.456 billion, compared with $23.860 billion in the prior quarter and $9.301 billion in the same period last year. GAAP net income was $28.243 billion, and non-GAAP EPS was $25.11. Gross margin reached 84.6% on a GAAP basis and 84.9% on a non-GAAP basis. The company guided fiscal Q4 revenue to $50.0 billion ± $1.0 billion, with gross margin around 86%.
Those numbers are extreme even by semiconductor-cycle standards. They reflect a market where AI systems are consuming memory and storage faster than the industry can supply it. Micron also said data-center revenue exceeded $25 billion in fiscal Q3, an annualized run rate of more than $100 billion, and that data-center SSD revenue exceeded $5 billion, more than doubling sequentially.
Micron’s management framed the cycle as more than temporary pricing. In its investor materials, the company said DRAM and NAND industry demand continues to significantly exceed supply and that tight conditions are expected to persist beyond calendar 2027 because of AI-driven demand and structural supply constraints. It also said it had signed 16 strategic customer agreements, typically with five-year terms for many customers, representing roughly 20% of DRAM volume and about one-third of NAND volume over the covered period.
Why Micron fits the same story as TeraWulf and Broadcom
At first, a TeraWulf data-center lease and a Micron memory cycle may appear unrelated. They are actually two sides of the same physical-stack problem. A data center without memory and storage is not useful. AI accelerators without HBM, DRAM and storage throughput cannot serve models efficiently. Agentic AI, context-heavy workloads, retrieval systems and inference at scale all increase memory and storage demand.
Reuters reported on July 6 that Micron and Ford signed a long-term agreement to secure memory and storage platforms for next-generation vehicle production. The article also noted a similar agreement with General Motors days earlier and cited DRAM prices up about 70% since December amid AI data-center demand. The automotive agreements show that memory shortages are not only a hyperscaler issue. Intelligent vehicles, ADAS and data-heavy infotainment systems are now competing for advanced memory and storage supply.
This is why Micron’s strategic customer agreements matter. They suggest customers want supply assurance, not only spot-market exposure. If large customers are willing to commit to multi-year agreements, the memory business may become less purely spot-driven than in prior cycles. That does not eliminate cyclicality, but it can change the revenue visibility and negotiating power of top suppliers.
The risk: memory super-cycles can become memory downcycles
Micron’s strength is also its risk. Memory is famous for overshooting. When prices rise dramatically, customers push back, governments get involved, competitors invest, and new capacity eventually arrives. Axios highlighted the uncomfortable side of the memory boom: memory chipmakers have made huge gains because AI demand hit a market with fixed supply, but those price increases can become a problem for buyers and may eventually incentivize new supply or alternative sourcing.
For $MU, the market debate is whether AI has permanently improved memory economics or merely created an extraordinary cycle. The right answer may be somewhere in the middle. AI likely raises the strategic importance of memory and storage, but that does not repeal supply response, capacity investment, customer bargaining power or cycle risk.
The red flag is not that Micron is weak. The company’s recent numbers are very strong. The red flag is that expectations may become extreme. When a cyclical business prints extraordinary margins, the market must ask how much of the profit pool is structural and how much is temporary scarcity.
The AI infrastructure stack: why these three tickers belong in one report
The key value of grouping $WULF, $AVGO and $MU is that they represent different layers of the same bottleneck chain. The AI model is the visible product. The physical infrastructure behind the model is the less visible but increasingly scarce system that makes the product usable.
| Layer | Representative ticker | What it controls | Why the market cares |
|---|---|---|---|
| Power-secured data-center campuses | $WULF | Large-scale AI infrastructure sites, power access, customer leases, campus development | AI labs need physical capacity, not only model ambition. Power and interconnection can become bottlenecks. |
| Custom silicon and networking | $AVGO | Custom chips, AI accelerators, networking, connectivity and long-duration customer supply relationships | Hyperscalers and major technology customers increasingly want custom chips and strategic supply lock-ins. |
| Memory and storage | $MU | DRAM, HBM, NAND, SSDs, data-center memory/storage products and strategic customer agreements | AI systems are architecturally dependent on memory bandwidth, capacity and storage performance. |
| Models and AI applications | Private / hyperscalers | Front-end AI products, inference demand, model training and agentic workloads | Application demand pulls the entire physical stack tighter. |
The market’s mistake is often to treat these layers as separate themes. In reality, they are one system. Anthropic needs capacity. TeraWulf needs to deliver the campus. The campus needs power, equipment, networking and compute. Compute needs custom chips and accelerators. Accelerators need memory and storage. Memory companies need long-term customer commitments to invest in capacity. The loop is circular.
This is why the AI trade can broaden beyond the biggest GPU names. Not because GPUs are unimportant, but because the industry’s constraints are multiplying. When one bottleneck is solved, another appears. In 2026, the bottlenecks increasingly include data-center power, custom silicon supply, memory availability, storage throughput, power electronics, transformers, cooling and grid infrastructure.
For readers, the operational takeaway is simple: follow where the bottleneck is moving. If AI demand keeps expanding, the next winners may not be only the companies with the best models. They may be companies with scarce power, long-term leases, custom silicon design wins, HBM capacity, memory supply agreements or the ability to finance and build infrastructure on time.
Risks and red flags: why this is not a simple “AI up, stocks up” story
Red flag 1: contracted revenue is not immediate free cash flow
The $19 billion TeraWulf headline is powerful, but investors must remember that it is expected contracted lease revenue over an initial 20-year term. It is not cash received today, and it is not automatically free cash flow. The campus must be delivered, the customer must occupy the capacity, rent must commence, operating costs must be managed and financing must be structured.
Red flag 2: AI campuses are capital intensive
Large AI data centers require major capital investment. Power infrastructure, cooling, interconnection, site development and customer-specific requirements can be expensive and complicated. If capital costs rise, timelines slip or financing becomes less favorable, headline lease values can be less attractive than they appear.
Red flag 3: custom chip relationships can be concentrated
Broadcom’s Apple relationship is valuable, but large customer concentration cuts both ways. A long-term agreement can support revenue visibility; it can also expose the supplier to customer bargaining power and future internalization risk. Broadcom is higher quality than a typical single-customer vendor, but customer concentration still belongs in the risk framework.
Red flag 4: memory pricing can overshoot
Micron’s current numbers are extraordinary. That is positive, but it also raises the bar. Memory markets can swing from shortage to oversupply if capacity catches up or if demand disappoints. AI may change the cycle, but investors should not assume cyclicality has disappeared.
Red flag 5: the AI trade can rotate violently
AI infrastructure stocks can move together during momentum phases, but the underlying businesses are not equally durable. A data-center developer, a custom silicon leader and a memory supplier should not receive the same valuation multiple or risk discount. When momentum reverses, the market usually becomes more selective very quickly.
The core risk in this trade is extrapolation. TeraWulf is not automatically a data-center REIT because it signed a major lease. Broadcom is not risk-free because it has elite customers. Micron is not immune to cycles because AI demand is strong. The opportunity is real, but each layer has its own failure mode.
Bull case, base case and bear case
| Scenario | What has to happen | What would support it | What could break it |
|---|---|---|---|
| Bull case | AI infrastructure demand remains strong, data-center capacity stays scarce, custom chip agreements deepen and memory supply remains tight through 2027. | TeraWulf delivers Justified Data capacity on schedule, Broadcom continues custom silicon growth, and Micron’s strategic customer agreements make memory earnings more durable. | Construction delays, capex overruns, AI capex digestion, customer pushback on memory prices or valuation compression across the AI trade. |
| Base case | The AI infrastructure trend remains intact, but stocks become more selective. High-quality large caps hold up better than high-beta transition stories. | Broadcom and Micron continue to print strong numbers, while TeraWulf trades around project milestones and financing details. | Market realizes that some headline AI contracts are long-dated and capital-intensive rather than immediately cash-generative. |
| Bear case | AI infrastructure expectations overshoot. Memory pricing draws customer resistance, data-center buildouts become more expensive and investors start discounting long-term contracted revenue more heavily. | Weak guidance, delayed capacity, higher project costs, inventory correction, looser memory supply or evidence that AI capex growth is slowing. | A renewed wave of long-term customer commitments and visible infrastructure delivery could challenge the bear case quickly. |
Merlintrader bottom line
The July 6 AI infrastructure tape is important because it shows the AI trade moving deeper into the physical economy. TeraWulf is no longer being discussed only as a mining-linked power user. With the Anthropic lease, it is being evaluated as a large-scale AI campus developer. Broadcom is not only a semiconductor supplier; it is a custom silicon and strategic customer lock-in platform. Micron is not only a memory-cycle stock; it is a supplier of one of the key bottlenecks in AI compute.
The theme is powerful because it is practical. AI models do not run on press releases. They run on power, chips, memory, storage, networking, cooling and data-center campuses. Every time AI usage grows, the market has to ask which part of the stack becomes scarce next. Today’s news points to three answers: power-secured data centers, custom silicon and memory.
For $WULF, the opportunity is dramatic but high-risk. The Anthropic lease gives TeraWulf a credible long-duration AI infrastructure anchor, but the business must still deliver capacity, manage capex, structure financing and convert a 20-year revenue headline into attractive economics.
For $AVGO, the story is quality and lock-in. The Apple extension through 2031 reinforces Broadcom’s role as a critical supplier to elite customers, while AI custom silicon and networking continue to support the broader growth case. The risk is less about survival and more about valuation, customer concentration and cycle expectations.
For $MU, the story is scarcity. Micron’s numbers show that memory has become strategic in the AI era. The company is benefiting from tight supply, HBM and data-center demand, strategic customer agreements and strong pricing. The risk is that memory markets have historically corrected hard when supply response catches up or demand expectations reset.
The right reading is not “all AI infrastructure stocks are buys.” The right reading is more disciplined: AI infrastructure is becoming a real investment category, but each layer must be judged on its own economics. TeraWulf must execute. Broadcom must sustain design wins and margins. Micron must prove that the memory cycle has become more durable, not simply more extreme.
As an editorial radar theme, this is one of the stronger non-biotech stories of the day. It connects small-cap/high-beta infrastructure, large-cap custom silicon and memory-cycle scarcity into one clean market narrative: the AI boom is no longer only about software or GPUs. It is becoming an infrastructure buildout, and the market is starting to price the companies that control scarce physical capacity.
Primary and reference sources
TeraWulf July 6, 2026 press release: TeraWulf — Anthropic lease at Justified Data Campus and Abernathy JV sale
TeraWulf Form 8-K: 8-K summary — lease terms, phased delivery, extension options and Abernathy transaction consideration
Reuters — Broadcom / Apple: Broadcom and Apple expand partnership through 2031
Broadcom Q2 FY2026 results: Broadcom — Q2 fiscal 2026 revenue, adjusted EBITDA and AI semiconductor revenue
Micron fiscal Q3 2026 results: Micron — fiscal Q3 2026 revenue, margins, EPS, product highlights and Q4 guidance
Micron fiscal Q3 2026 investor materials: Micron — data-center revenue, strategic customer agreements and supply-demand commentary
Reuters — Micron / Ford: Micron and Ford long-term memory and storage supply agreement
Axios — memory chip boom context: Memory chip pricing, supply constraints and buyer pushback risk
Informational and educational content only. This article is not financial advice, investment advice, personalized advice, or a recommendation to buy or sell any security. $WULF, $AVGO and $MU carry materially different risk profiles. TeraWulf is a high-volatility infrastructure transition story exposed to construction, financing, customer, energy, data-center and execution risks. Broadcom is exposed to customer concentration, semiconductor-cycle, valuation, custom silicon and supply-chain risks. Micron is exposed to memory-cycle, pricing, capacity, supply-demand, capital-expenditure and valuation risks. Contracted revenue, AI demand, customer agreements, lease terms, supply conditions, pricing, margins and guidance can change. Always verify company-specific data with official filings, company releases and primary sources.