The Engine Behind the Model: Anthropic IPO Infrastructure & Supply Chain Stocks 2026
- Hawkmont Research

- Jul 3
- 27 min read
Updated: Jul 30
The Hyperscalers, Chipmakers, and Hidden Suppliers Wired Into Claude
Hawkmont Research | Equity Research | AI Infrastructure Economy
July 3, 2026 | INITIATING COVERAGE
This report is produced independently by Hawkmont Research. No compensation was received from any company mentioned herein. All prices and market data as of July 2, 2026 unless otherwise noted. This is not investment advice. Please read full disclosures at the end of this report.

Which Stocks Benefit from the Anthropic IPO?
You cannot buy Anthropic today, and when its S-1 does go public, you will be buying a company that spends roughly as much on compute as it earns in revenue. The more accessible, arguably more attractive trade sits one layer down: the confirmed cloud, chip, memory, and infrastructure partners that are contractually locked into Claude's build-out regardless of where Anthropic's own valuation lands.
Unlike the SpaceX supply chain, Anthropic's is not a story of undiscovered small caps. It runs through five of the largest companies on earth, plus a handful of second-order beneficiaries that remain comparatively under-discussed. That changes the framework: this report is less about finding hidden gems and more about correctly weighting exposure across mega-cap names whose Anthropic revenue contribution is real, disclosed, and growing, but is buried inside much larger consolidated financials.
The names Hawkmont Research has identified with confirmed, contracted Anthropic infrastructure relationships are:
Amazon.com Inc (NASDAQ: AMZN): Primary cloud and training partner since 2023, $8B+ invested, Project Rainier Trainium2/3 cluster, up to $50B in Series H linked infrastructure commitments
Alphabet Inc (NASDAQ: GOOGL): Early investor with roughly a 14% stake, up to 1 million TPUs in a "tens of billions of dollars" deal, multi-gigawatt Broadcom TPU expansion through 2027+
Microsoft Corp (NASDAQ: MSFT): Up to $5B equity investment, $30B Azure compute commitment plus 1 GW of Nvidia powered capacity, Claude now on all three hyperscalers
NVIDIA Corp (NASDAQ: NVDA): Up to $10B equity investment, 1 GW Grace Blackwell/Vera Rubin compute commitment, deep co-engineering partnership
Broadcom Inc (NASDAQ: AVGO): Confirmed custom silicon design partner on Anthropic's "Titan" accelerator program, multi-gigawatt TPU packaging and networking exposure via Google
Marvell Technology (NASDAQ: MRVL): Co-design partner on AWS Trainium2/3, the chip family carrying the bulk of Claude's training and inference workload
Micron Technology (NASDAQ: MU): New Series H strategic investor and multi-year HBM/DRAM/SSD supply partner, joining Samsung and SK Hynix as the third global HBM supplier now aligned with Anthropic
The full analysis, financial context, deal verification, valuation framework, and a bull/bear case for each name follows.
Table of Contents
Executive Summary
Anthropic: Where the Company Actually Stands Ahead of an IPO
Investment Framework: How Infrastructure Partners Win
Company Deep-Dives: Amazon, Alphabet, Microsoft, NVIDIA, Broadcom, Marvell, Micron
Notable Mentions: Samsung Electronics, SK Hynix, Fluidstack, and AMD
Portfolio Recommendation and Conviction Ranking
Risks, Sensitivities, and Multi-Year Durability
Appendix: Comparables Table, Valuation Snapshot, Key Sources, Deal Timeline
Section 1: Executive Summary
Anthropic confidentially submitted a draft Form S-1 registration statement to the SEC on June 1, 2026, a proposed IPO that would follow a run of private financings almost without precedent in corporate history. The company's post-money valuation moved from $61.5 billion in March 2025 to $183 billion in September 2025, to $380 billion in February 2026, to $965 billion in its $65 billion Series H round in late May 2026. Annualized revenue crossed $47 billion in May, up from roughly $1 billion at the end of 2024. No enterprise software or infrastructure company has compounded revenue at this rate, at this scale, over three consecutive years.
That growth is not free. Anthropic has told investors it expects to spend approximately $19 billion on training and inference compute in 2026, close to matching its full year revenue, and the company is not expected to be profitable before 2028. The practical consequence for public market investors is that the more durable, better understood way to underwrite Anthropic's trajectory right now is through the infrastructure companies that are contractually supplying the compute, memory, and silicon that Claude runs on. Every one of those relationships is disclosed, dated, and denominated in real dollars, unlike Anthropic's own eventual float, price range, and exchange listing, none of which has been set as of this writing.
The three highest conviction ideas from this report:
Amazon.com Inc (NASDAQ: AMZN). Anthropic's primary cloud provider since 2023 and primary training partner since 2024. Amazon has invested roughly $8 billion into Anthropic and operates Project Rainier, an Indiana based data center campus built around nearly one million AWS Trainium2 chips, with a planned buildout to 2.3 gigawatts of capacity. Anthropic committed up to $50 billion of new Series H linked infrastructure spend that flows disproportionately toward AWS. Rothschild & Co Redburn estimated in October 2025 that Anthropic had already added one to two points of growth to AWS in Q4 2024 and Q1 2025, with that contribution expected to exceed five points in the second half of 2025.
Alphabet Inc (NASDAQ: GOOGL). Google was an early Anthropic investor (roughly $3 billion for an approximate 14 percent stake) and is now both shareholder and critical infrastructure supplier, having agreed in October 2025 to expand Anthropic's access to Google Cloud TPUs to as many as one million chips in a deal reported as worth "tens of billions of dollars," bringing well over a gigawatt of capacity online in 2026. In April 2026, Google, Anthropic, and Broadcom expanded that relationship again for multiple additional gigawatts of TPU capacity starting in 2027, sited predominantly in the United States. Google Cloud's own revenue growth (63 percent year over year in Q1 2026) is running well ahead of the broader cloud market, and Anthropic is one of the disclosed reasons why.
NVIDIA Corp (NASDAQ: NVDA). Alongside Microsoft, Nvidia announced in November 2025 that it will invest up to $10 billion into Anthropic and establish, for the first time, a direct technology partnership under which Anthropic will deploy up to 1 gigawatt of Nvidia Grace Blackwell and Vera Rubin systems, with the two companies collaborating on design and engineering. Anthropic remains a comparatively diversified buyer across Trainium, TPUs, and Nvidia GPUs, but the sheer scale of the compute build means even a minority share of Anthropic's roadmap represents a large absolute number of GPUs.
Summary risk note: These are mega cap, diversified businesses, so Anthropic specific risk is diluted relative to the SpaceX supply chain report's small caps. But that dilution cuts both ways: Anthropic's contribution to any single hyperscaler's growth, while real and analyst quantified, is one input among many. Compute overbuild, a slower than expected Anthropic IPO, memory pricing volatility, and antitrust or DOJ scrutiny of the circular hyperscaler investor supplier structure are all live considerations.
Section 2: Anthropic, Where the Company Actually Stands Ahead of an IPO
2.1 What Has Been Confirmed
On June 1, 2026, Anthropic, PBC confidentially submitted a draft registration statement on Form S-1 to the SEC under Rule 135 of the Securities Act of 1933. The company has not disclosed share count, price range, exchange, or ticker. A confidential filing gives Anthropic the option to go public once SEC review concludes; it is not a commitment to a specific listing date. Reported market chatter points to a possible listing window around October 2026, though this is unconfirmed and, as with the SpaceX and OpenAI processes running in parallel, subject to slippage.
Key confirmed details from the run up to the filing:
Series H: $65 billion raised in late May 2026, co-led by Altimeter Capital, Dragoneer, Greenoaks, Sequoia Capital, Capital Group, Coatue, D1 Capital Partners, GIC, ICONIQ, and XN, at a $965 billion post-money valuation
Revenue run-rate: $47 billion as of late May 2026, up from a $30 billion run-rate earlier in the year and roughly $10 billion in full year 2025 revenue
Total capital raised to date: approximately $125 to $132 billion across 18 rounds
Strategic infrastructure investors in Series H: Micron, Samsung, and SK Hynix, alongside $15 billion of previously committed hyperscaler investment including $5 billion from Amazon
Structure: Anthropic is a Public Benefit Corporation (Anthropic, PBC), which legally binds the company to balance shareholder returns against its stated mission, a governance feature public investors typically price as a discount factor
Anthropic's June 2026 IPO filing landed the same week that Anthropic disclosed it would pay SpaceX $1.25 billion per month through May 2029 for compute capacity at SpaceX's Colossus 1 data center in Memphis, an unusual cross company compute leasing arrangement that surfaced in SpaceX's own IPO prospectus and underscores just how compute constrained the frontier labs remain even at hundred billion dollar funding scale.
2.2 The Growth Engine: Enterprise, Coding, and Consumer
Anthropic's revenue is not a single product line. Three distinct growth vectors are driving the run-rate:
Enterprise API and Claude Code: Claude Code, launched broadly in May 2025, reached a $2.5 billion annualized run-rate by February 2026, more than doubling since the start of the year, with an estimated 4 percent of all public GitHub commits worldwide now authored by Claude Code, roughly double the share from a month earlier. Business subscriptions to Claude Code quadrupled in the first six weeks of 2026, and enterprise now represents more than half of all Claude Code revenue. Eight of the Fortune 10 are disclosed Anthropic customers, and the company has said more than 1,000 enterprise accounts now spend over $1 million a year on Claude, up from roughly a dozen two years earlier.
Consumer: Claude jumped to the number one slot on Apple's U.S. free app chart in late February 2026, a period that coincided with a public dispute between Anthropic and the U.S. Department of Defense, whose blacklisting of Anthropic's models in the defense contractor channel did not appear to slow broader commercial and consumer momentum. Anthropic is suing to reverse that blacklisting, and litigation is ongoing.
Frontier model cadence: Anthropic has shipped a rapid sequence of model releases through the period covered by this report, including Claude Opus 4.7, Opus 4.8, Sonnet 4.5, Sonnet 5, and a limited access cybersecurity focused model line under Project Glasswing (Claude Mythos and Claude Fable), the latter of which briefly had its access suspended in June 2026 to comply with U.S. Commerce Department export controls before being restored on July 1, 2026. That cadence is central to the bull case for infrastructure suppliers: each generation requires a step change in training compute, and Anthropic has shown no evidence of slowing its release schedule to conserve capital.
2.3 The Infrastructure Bill
Anthropic's compute spending is the central variable for every name in this report. Forbes reported in May 2026 that Anthropic planned to spend approximately $19 billion on training and inference compute in 2026, a figure that roughly matched projected full year revenue at the time of reporting, with gross margins compressed to around 40 percent after inference costs ran 23 percent over internal projections. That is a materially thinner margin profile than most enterprise SaaS comparables, and it is the single best argument for owning the infrastructure suppliers rather than waiting for Anthropic's own public listing: the dollars are flowing to Amazon, Google, Microsoft, Nvidia, Broadcom, Marvell, and the memory makers today, in contracted, disclosed amounts, regardless of how Anthropic's own IPO ultimately prices.
Section 3: Investment Framework, How Infrastructure Partners Win
3.1 Criteria for a High-Quality Anthropic-Linked Position
Contract confirmation and disclosure quality. Every name in this report has a publicly confirmed, named relationship with Anthropic, either through joint press releases, regulatory disclosure, or direct commentary from company executives on earnings calls. Unlike the SpaceX small cap supply chain, there is no rumored relationship tier here; the surrounding chatter simply does not exist at this scale of counterparty.
Revenue materiality and disaggregation. This is the central analytical challenge with this report relative to a small cap supply chain piece: none of the seven companies below break out "Anthropic revenue" as a separate line item. Analysts (Redburn, SemiAnalysis, Bloomberg Intelligence) have produced estimates of Anthropic's contribution to AWS and Google Cloud growth rates, and those estimates are cited throughout, but investors should treat them as informed approximations, not disclosed fact.
Contract duration and structural lock-in. The strongest theses here (AWS/Trainium, Google/TPU, Micron/Samsung/SK Hynix memory) involve multi-year capacity commitments and, in several cases, direct co-engineering. Anthropic has structured its silicon relationships with AWS's Annapurna Labs and with Nvidia in ways that go beyond a simple purchase order and create real switching costs on both sides.
Diversification of the underlying business. Every name in this report, unlike the SpaceX small caps, has revenue streams entirely independent of Anthropic. That cuts both ways: Anthropic specific upside is diluted across a much larger revenue base, but so is Anthropic specific downside. Investors seeking concentrated, high beta Anthropic exposure will not find it in mega cap infrastructure names; they will have to wait for the S-1 itself.
3.2 The Supply-Chain Segments
Cloud capacity and training partnerships: AWS, Google Cloud, and Microsoft Azure, the three hyperscalers now all host Claude, an unusual multi cloud posture that Anthropic has deliberately maintained (and which insulated Claude from an October 2025 AWS outage that affected many single cloud competitors).
Compute silicon: AWS Trainium (Marvell co-design), Google TPU (Broadcom co-design), and Nvidia GPUs (Grace Blackwell, Vera Rubin). Anthropic runs a genuinely multi chip fleet rather than standardizing on one vendor, a strategic choice management has described as a resilience and price performance decision.
Memory and storage: HBM, DRAM, and SSD supply from the three global HBM suppliers, SK Hynix (roughly 50 to 57 percent share), Samsung (roughly 22 to 28 percent), and Micron (roughly 21 to 22 percent), all three of which are now Anthropic Series H strategic investors as well as suppliers, a rare "supplier as shareholder" structure across an entire concentrated oligopoly.
Physical infrastructure and power: Anthropic's own $50 billion Fluidstack partnered data center buildout in Texas and New York, alongside AWS's Project Rainier campus and Google/Broadcom's expanding TPU footprint, all sited predominantly in the United States.
3.3 How to Read the Valuation Data in This Report
Every name below carries a short valuation and return framework alongside the deal specifics. These are not price targets. Hawkmont Research does not publish 12-month price targets for mega-cap names whose Anthropic exposure is a single, undisclosed line item inside a much larger business; a formal target would imply a precision this report cannot support. Instead, each name gets a plain read on what has to happen, on the multiple and on the Anthropic-specific catalyst, for the position to work, and what breaks the thesis if it does not.
Section 4: Company Deep-Dives
4.1 Amazon.com Inc (NASDAQ: AMZN)
Conviction: HIGH | Segment: AWS (cloud/compute) Relationship: Confirmed, Deepest, and Structurally Central
Amazon has been Anthropic's primary cloud provider since 2023 and primary training partner since 2024. Amazon has invested approximately $8 billion into Anthropic across multiple tranches, with an additional $5 billion previously committed as part of the Series H hyperscaler bloc disclosed in May 2026.
The physical manifestation of the partnership is Project Rainier, an AWS data center campus in St. Joseph County, Indiana (with additional capacity in Pennsylvania and Mississippi), built specifically for Anthropic's training and inference workloads. AWS invested $11 billion in the Indiana site alone. As of late 2025, Project Rainier hosted nearly 500,000 AWS Trainium2 chips across seven buildings on roughly 1,200 acres, with a stated plan to exceed one million Trainium2 chips by the end of 2025 and a longer term buildout to 23 additional buildings and 2.3 gigawatts of capacity. Anthropic's Claude now runs on more than one million Trainium2 chips, and Anthropic is described by industry analysts as effectively using AWS's Annapurna Labs as a custom silicon design partner, a level of hardware software co-design SemiAnalysis has called unique among AI labs alongside Google DeepMind's TPU relationship.
Financial framing: Redburn analyst Alex Haissl estimated in an October 2025 note that Anthropic had already added one to two percentage points to AWS's growth in Q4 2024 and Q1 2025, with that contribution expected to exceed five points in the second half of 2025. AWS CEO Andy Jassy has described Trainium2 as "fully subscribed" and a "multibillion dollar business" that grew 150 percent quarter over quarter, with Trainium3 supply expected to be fully committed by mid-2026. Amazon's overall capex guidance for 2026 runs well above $100 billion, with AI infrastructure, much of it Anthropic linked, the primary driver.
Valuation and entry point (as of July 2, 2026, verify before publish): AMZN closed near $242.67, roughly 12 percent below its 2026 high of $278.56 and comfortably above its February low of $196.00. The stock trades around 33 to 35x trailing earnings, a premium to the broader market that assumes continued AWS growth acceleration. The return case here is not that Anthropic alone moves the stock; it is that Anthropic is one of two or three visible reasons AWS growth reaccelerates off a multi-year low, and the market has not fully priced that reacceleration into AWS's segment multiple. Downside case: if AWS growth disappoints even with Trainium fully subscribed, the market will conclude Anthropic's contribution was already in the price.
Key risks: AWS is a large, diversified business, and Anthropic is one demand driver among many (Amazon's own retail AI, Bedrock third party hosting, and a separate, growing AWS-OpenAI relationship). Trainium yield and ramp have faced normal new product bumps. Any acceleration of Anthropic's multi cloud diversification toward Google or Microsoft would dilute AWS's share of incremental spend, though Series H commitments suggest absolute exposure keeps growing regardless.
4.2 Alphabet Inc (NASDAQ: GOOGL)
Conviction: HIGH | Segment: Google Cloud / TPU Relationship: Confirmed, Investor and Critical Supplier Simultaneously
Google is both one of Anthropic's earliest and largest investors (an initial roughly $3 billion investment for what has been reported as close to a 14 percent stake, since added to) and the supplier of what may be Anthropic's single largest compute commitment. In October 2025, the two companies announced an expansion of Anthropic's Google Cloud TPU access to as many as one million chips, in a deal reported as worth "tens of billions of dollars" and expected to bring well over a gigawatt of capacity online in 2026, Anthropic's largest TPU commitment to date.
In April 2026, Anthropic, Google, and Broadcom expanded the partnership again for multiple additional gigawatts of next generation compute starting in 2027, which Anthropic described as a major expansion of its November 2025 commitment to invest $50 billion in U.S. computing infrastructure, sited predominantly domestically.
Financial framing: Google Cloud reported $20 billion in Q1 2026 revenue, up 63 percent year over year, a growth rate meaningfully ahead of AWS and Azure's cloud segments in percentage terms, even as it remains the smallest of the three hyperscale clouds in absolute revenue. Anthropic is one of the disclosed drivers of that acceleration. Industry reporting has highlighted a strategic tension worth noting: Google Cloud is renting its most scarce strategic asset, TPU capacity, to the company that competes most directly with its own Gemini model family, a dynamic that has not, to date, slowed the pace of deal expansion.
Valuation and entry point (as of July 2, 2026, verify before publish): GOOGL traded near $362, off from a May 2026 high above $408, at roughly 27x trailing earnings and 29x forward, the cheapest multiple of the three hyperscalers in this report despite Google Cloud posting the fastest growth rate of the three. That gap between growth rate and multiple is the core of the Alphabet return case: the market is still pricing Google Cloud closer to a legacy third-place cloud than a business now growing faster than AWS or Azure, with Anthropic's TPU commitment as one visible reason why. The counterargument, and the reason the multiple stays discounted, is the same competitive tension noted above: investors are reluctant to fully credit a business for supplying the compute that powers a direct competitor to its own flagship product.
Key risks: Alphabet is a large, diversified company where Anthropic linked TPU revenue is a single line within a much larger cloud and advertising business. The competitive tension between supplying Anthropic's compute and competing with Claude via Gemini is a structural oddity worth watching. Regulatory overhang is also live and current: Alphabet lost a EU antitrust appeal in early July 2026 tied to a roughly 4.1 billion euro fine, a reminder that Alphabet carries meaningful regulatory risk unrelated to the Anthropic thesis.
4.3 Microsoft Corp (NASDAQ: MSFT)
Conviction: MEDIUM-HIGH | Segment: Azure / equity stake Relationship: Confirmed, Newest of the Big Three Hyperscalers
Announced in November 2025 alongside Nvidia, Microsoft committed to invest up to $5 billion into Anthropic (some reporting has cited figures up to $10 billion), while Anthropic committed to purchase $30 billion of Microsoft Azure compute capacity and to contract for additional capacity of up to 1 gigawatt using Nvidia Grace Blackwell and Vera Rubin systems. The deal made Claude the only frontier model family available on all three major hyperscale clouds simultaneously, and extended Claude's distribution into Microsoft Foundry.
The deal pushed Anthropic's valuation at the time to a reported $350 billion range, up from $183 billion two months earlier, the clearest single data point of how directly infrastructure commitments and equity valuation moved together in this cycle.
Financial framing: Microsoft's own AI infrastructure narrative has historically centered on OpenAI (a reported 27 percent stake and 20 percent revenue share). The Anthropic deal is explicitly diversifying: Microsoft is now a strategic investor and infrastructure partner to both leading frontier labs simultaneously.
Valuation and entry point (as of July 2, 2026, verify before publish): MSFT traded near $386, down sharply from a 52-week high above $555 set in late 2025, a drawdown driven mainly by concern over AI capex outrunning near-term monetization rather than by anything Anthropic-specific. At roughly 21 to 22x forward earnings, this is the cheapest forward multiple Microsoft has carried since 2023. The return case is a re-rating trade: Azure growth reacceleration, partly OpenAI-driven and partly Anthropic-driven, closing the gap between a AAA-rated compounder and a stock currently priced for disappointment. This is the most contrarian entry point of the three hyperscalers in this report precisely because the drawdown has been the steepest.
Key risks: Microsoft's Anthropic exposure is the newest and smallest of the three hyperscaler ties in this report, and the $30 billion Azure commitment still trails Anthropic's cumulative AWS and Google commitments. Microsoft's dual OpenAI/Anthropic positioning could create allocation tension for Nvidia GPU supply. The stock's recent weakness also reflects company-specific concerns (reported layoffs, capex-versus-monetization skepticism) that are independent of the Anthropic thesis and should not be conflated with it.
4.4 NVIDIA Corp (NASDAQ: NVDA)
Conviction: HIGH | Segment: GPU compute / equity stake Relationship: Confirmed, Newest Direct Technology Partnership
Nvidia's relationship with Anthropic was, until November 2025, notably thinner than its relationships with most other frontier labs; Anthropic had built its compute strategy primarily around AWS Trainium and Google TPUs. That changed with the joint Microsoft-Nvidia-Anthropic announcement: Nvidia committed up to $10 billion in Anthropic and, for the first time, established a direct engineering partnership under which the two companies will collaborate on design and price performance optimization, with Anthropic committing to deploy up to 1 gigawatt of Grace Blackwell and Vera Rubin systems. Jensen Huang called the partnership "a dream come true," language that reflected how long Anthropic had remained a comparative Nvidia holdout.
Financial framing: One gigawatt of AI compute capacity is commonly estimated at roughly $50 billion in total infrastructure cost, of which roughly $35 billion is typically GPU or chip spend, meaning Anthropic's Nvidia commitment alone represents a multi billion dollar order book addition. Nvidia's willingness to directly finance a customer mirrors the structure it has used with OpenAI at a larger scale ($100 billion committed) and is part of a broader "circular economy" pattern several analysts have flagged as worth monitoring.
Valuation and entry point (as of July 2, 2026, verify before publish): NVDA closed near $193, well off its May 2026 closing high of $235, at roughly 33x trailing earnings, a multiple that has compressed materially even as the AI buildout narrative has not slowed. Nvidia is the largest company in this report by market capitalization and the one where Anthropic exposure is the smallest fraction of the overall investment case; Nvidia works or does not work on the broader hyperscaler capex cycle, with Anthropic as a real but incremental contributor. The entry point is more attractive after the pullback than it was in May, but the return case here rests almost entirely on total AI capex, not on Anthropic specifically.
Key risks: Nvidia remains more exposed to OpenAI, Meta, and broader hyperscaler capex than to Anthropic specifically. Anthropic's continued use of Trainium and TPU capacity alongside Nvidia GPUs means Nvidia is one of three chip suppliers rather than the primary one. Semiconductor stocks broadly sold off in the days ahead of this report's publication, a reminder that this name carries sector-wide beta well beyond anything Anthropic-related.
4.5 Broadcom Inc (NASDAQ: AVGO)
Conviction: MEDIUM-HIGH | Segment: Custom silicon design (ASIC) Relationship: Confirmed Custom Silicon Design Partner
Broadcom is the design partner behind Google's TPU program, and multiple 2026 industry reports have confirmed Broadcom is also the design partner on a custom Anthropic accelerator program reported under the name "Titan," alongside a parallel OpenAI custom chip program. Broadcom's disclosed AI ASIC customer list as of Q1 FY2026 earnings included Google, Meta, OpenAI, Anthropic, and Apple. Broadcom is also central to Anthropic's exposure indirectly through Google; the April 2026 multi gigawatt TPU expansion explicitly names Broadcom as a partner.
Financial framing: Broadcom's custom AI silicon franchise is estimated to command roughly 60 to 70+ percent of the hyperscaler custom ASIC design services market, a share that has been expanding through 2026 as its confirmed customer list has grown. Broadcom's non-GAAP operating margin (reported around 60 percent including VMware) is structurally higher than pure GPU or memory manufacturers, reflecting its royalty and design services model.
Valuation and entry point (as of July 2, 2026, verify before publish): AVGO traded in the high $360s, holding up better through the recent semiconductor pullback than either Marvell or Micron, consistent with a business model (design royalties, multi-customer diversification) that carries lower single-customer concentration risk than its peers in this report. The return case is the least Anthropic-specific of any name here: Broadcom's five-plus-customer ASIC book means the stock works if the custom silicon shift away from merchant GPUs continues broadly, with Anthropic as one confirmed name among several rather than a swing factor.
Key risks: Broadcom's Anthropic specific exposure is not separately disclosed and sits inside a much larger, diversified AI ASIC business spanning five plus major customers. The custom ASIC design business is also a target for in-house design ambitions at several hyperscalers over a multi year horizon.
4.6 Marvell Technology (NASDAQ: MRVL)
Conviction: MEDIUM | Segment: Custom silicon design (ASIC) / networking Relationship: Confirmed Co-Design Partner on AWS Trainium
Marvell is AWS's primary co-design partner for the Trainium chip family, the silicon carrying the largest single share of Claude's current training and inference workload. Marvell and Amazon formalized a five year supply agreement in late 2024, and Marvell's Trainium linked revenue has scaled alongside Project Rainier's buildout; Trainium3, manufactured on TSMC's 3-nanometer node with additional fabrication support from Marvell, began ramping in 2026. Marvell has separately been reported to anchor Microsoft's Maia custom accelerator program, giving it a second hyperscaler anchor, and holds an estimated 20 to 25 percent share of the custom ASIC design services market versus Broadcom's larger share.
Financial framing: Morgan Stanley has guided to custom silicon revenue growth at Marvell of roughly 20 percent in 2026 and 100 percent in 2027, driven substantially by the Trainium ramp. Marvell shares have historically traded at more modest multiples than Broadcom, reflecting both its smaller scale and periodic investor concern about the durability of its hyperscaler relationships.
Valuation and entry point (as of July 2, 2026, verify before publish): This is the name in this report requiring the most caution on entry timing. MRVL fell more than 10 percent on July 2, 2026 alone, part of a broader unwind following the stock's inclusion in the S&P 500 on June 22 and a subsequent analyst downgrade citing gross margin softness and a trailing P/E north of 90x. The stock is still up several hundred percent over the trailing year on the Trainium and custom silicon narrative, but the July 2 move is a reminder that this is the highest-beta, most binary name in the report: Marvell's growth forecast leans heavily on a small number of hyperscaler relationships, and roughly 82 percent of its revenue is concentrated among its top ten cloud customers. Investors sizing into this name should expect volatility measured in double-digit daily percentage moves, not the steadier drift of Amazon or Broadcom.
Key risks: Marvell's stock has, at points in 2026, struggled to hold gains on reports questioning the durability of its Amazon and Microsoft positioning as those hyperscalers expand in-house design capability. Marvell's Anthropic exposure is entirely indirect, flowing through Amazon's infrastructure decisions rather than a direct Marvell-Anthropic relationship, the most levered but least controllable name in this report. The valuation, even after the July pullback, prices in a great deal of execution that has not yet happened.
4.7 Micron Technology (NASDAQ: MU)
Conviction: MEDIUM-HIGH | Segment: HBM / DRAM / SSD memory Relationship: Confirmed, Newest Strategic Investor and Supplier
Micron's relationship with Anthropic was formalized on June 22, 2026, the most recent deal in this report, bundling a multi year supply agreement spanning HBM, DRAM, and SSDs; joint memory and storage co-design work; internal Claude deployment inside Micron; and a Micron equity investment as part of Anthropic's Series H. The deal made Micron the third of the world's three HBM suppliers, alongside Samsung and SK Hynix, to become both an Anthropic investor and a named supplier, an unusual structure in which an entire concentrated oligopoly has aligned itself with a single AI lab's capital raise.
Financial framing: HBM remains the tightest bottleneck in the AI hardware stack. SK Hynix holds an estimated 50 to 57 percent global HBM share, Samsung roughly 22 to 28 percent, and Micron roughly 21 to 22 percent, and as of Q1 2026 all three suppliers' annual HBM capacity was reported fully booked, with an estimated 20 to 50 percent supply demand gap persisting into 2028. DRAM pricing has risen sharply through 2026 as capacity shifts toward higher margin HBM production.
Valuation and entry point (as of July 2, 2026, verify before publish): MU traded near $1,035, down roughly 6 percent on July 2 alongside the broader chip sell-off but still reflecting an extraordinary run built on the memory supercycle narrative; Micron, Intel, and AMD added a combined $2 trillion in value in Q2 2026 alone. The return case here is the cleanest of any name in this report: Micron is the only U.S.-listed pure play on a genuine, multi-year, physically constrained supply shortage, and it is now both a supplier to and shareholder alongside the two companies that control the rest of global HBM capacity. The risk is that a run this steep has already priced in several years of favorable memory pricing, leaving less room for error than the underlying supply-demand story would suggest on its own.
Key risks: The three memory makers, including Micron, are named defendants in a June 25, 2026 U.S. antitrust class action alleging coordinated restriction of conventional DRAM supply disguised as an HBM capacity shift. Memory pricing is also cyclical by nature; the current shortage narrative has reversed before in prior semiconductor cycles.
Section 5: Notable Mentions
5.1 Samsung Electronics (KRX: 005930) and SK Hynix (KRX: 000660)
Assessment: CONFIRMED STRATEGIC PARTNERS. Not Easily Accessible to Retail via US Brokerages
Both Korean memory giants joined Micron as named "strategic infrastructure partners" in Anthropic's Series H round, and both saw share price gains in Seoul trading following the late May 2026 announcement. SK Hynix's roughly 50 to 57 percent global HBM share makes it, by volume, the single most important memory supplier to the entire AI industry. Neither trades as a standard U.S. equity; access for most retail investors runs through OTC ADR-equivalent listings or international brokerage access. Note for timing: both stocks fell sharply in Seoul trading in the days ahead of this report's publication as the global chip sell-off spread internationally, a reminder that this exposure carries currency risk on top of sector volatility.
5.2 Fluidstack
Assessment: PRIVATE, NOT INVESTABLE. Included for Context
Fluidstack, a UK-based "neocloud" provider, is the counterparty on Anthropic's $50 billion, November 2025 announced independent data center buildout across Texas and New York, Anthropic's first major move to own physical infrastructure rather than lease it entirely from the big three hyperscalers. Fluidstack is privately held and not a public equity; it is included here purely so investors understand where a meaningful share of Anthropic's own infrastructure capital is being spent.
5.3 Advanced Micro Devices (NASDAQ: AMD): A Notable Absence
AMD does not appear in the conviction ranking in Section 6, and readers familiar with the AI chip landscape may wonder why. As of this report's publication, there is no confirmed Anthropic-AMD supply relationship. A job listing on Anthropic's Reinforcement Learning team in April 2026 referenced familiarity with AMD's ROCm software ecosystem alongside Nvidia's CUDA and Triton, and SemiAnalysis has speculated Anthropic may be evaluating AMD's forthcoming MI450 accelerators, but this remains industry chatter rather than a disclosed deal on the order of the seven relationships covered in Section 4. Hawkmont Research will revisit AMD's inclusion if and when a confirmed commercial relationship is announced.
Section 6: Portfolio Recommendation and Conviction Ranking
6.1 Conviction Ranking
A note on terminology: "sleeve weight" below refers to the suggested allocation within a dedicated Anthropic-infrastructure thematic bucket, not a recommendation for what share of an investor's total portfolio this theme should occupy. An investor might reasonably run this entire thematic sleeve at 5 to 15 percent of a diversified portfolio, with the percentages below describing how to weight names inside that sleeve, not against total net worth.
Rank | Company | Ticker | Conviction | Suggested Sleeve Weight |
1 | Amazon.com Inc | AMZN | High | 20 to 25% |
2 | Alphabet Inc | GOOGL | High | 20 to 25% |
3 | NVIDIA Corp | NVDA | High | 15 to 20% |
4 | Microsoft Corp | MSFT | Medium-High | 10 to 15% |
5 | Broadcom Inc | AVGO | Medium-High | 10 to 15% |
6 | Micron Technology | MU | Medium-High | 5 to 10% |
7 | Marvell Technology | MRVL | Medium | 5 to 10% |
6.2 Entry Strategy
Pre-public-S-1 window: Anthropic's confidential S-1 does not yet disclose financials publicly. This is not a "get in before the crowd" trade the way the SpaceX supply chain was; it is an "own the picks and shovels regardless of Anthropic's own IPO outcome" trade.
Public S-1 filing: When the prospectus becomes public, expect detailed disclosure of compute commitments by counterparty, the first chance to verify, rather than estimate, each partner's actual dollar exposure.
Anthropic IPO pricing and listing: Given the scale of this offering relative to Anthropic's compute counterparties, Hawkmont Research would not expect a material standalone re-rating of any single mega cap name purely on Anthropic's listing.
Post-listing: The more durable catalyst is Anthropic's continued compute spend as a public company with quarterly disclosure, making Anthropic's own capex guidance a trackable, recurring input into estimates for these names going forward.
6.3 Position Sizing Principles
Every name in this report is a mega cap, diversified business. Anthropic specific concentration risk is structurally lower than in a small cap supply chain report, which also means Anthropic specific upside is more diluted. Investors seeking genuinely concentrated Anthropic exposure will not find it in any of these seven names; that exposure will only exist once Anthropic itself lists. Samsung and SK Hynix access constraints should be factored into position sizing for U.S.-based retail investors; Micron is the most accessible pure U.S.-listed way to express the HBM oligopoly thesis. Marvell's July 2 volatility is a useful live example of why the "Medium" conviction names in this report warrant smaller position sizes and staged entry rather than a single lump-sum purchase.
Section 7: Risks, Sensitivities, and Multi-Year Durability
7.1 Anthropic-Specific Risks
IPO timeline and pricing uncertainty: Anthropic's confidential filing does not commit the company to a listing date. If market conditions deteriorate, or the parallel SpaceX and OpenAI IPO processes absorb investor appetite, Anthropic's listing could slip into 2027.
Compute spend versus revenue: Anthropic's compute spend has been reported as roughly matching its full year revenue, with margins compressed by inference costs running well over internal projections. A slowdown in Anthropic's growth would flow through to reduced orders across every name in this report.
Governance structure: Anthropic's PBC structure legally requires balancing shareholder return against its stated mission, a feature public investors typically price as a discount factor.
Regulatory and defense-channel disruption: Anthropic's models were blacklisted from the DOD contractor channel earlier in 2026 amid a dispute now in litigation; a similar disruption elsewhere would be a headline risk.
7.2 Infrastructure Partner Risks
Compute overbuild: The scale of committed gigawatt capacity across Anthropic's cloud partners raises the standard industry question of whether AI infrastructure investment is outrunning near-term demand.
Circular financing structure: Nvidia and Microsoft investing equity into Anthropic while simultaneously selling it tens of billions in compute is part of a broader "circular economy" pattern drawing analyst and regulatory attention.
Antitrust exposure in memory: Micron, Samsung, and SK Hynix face an active June 2026 U.S. class action antitrust suit alleging coordinated restriction of conventional DRAM supply under cover of the HBM capacity shift.
Customer concentration at the design services layer: Broadcom's and Marvell's custom silicon businesses depend on continued hyperscaler willingness to outsource chip design rather than build fully in-house capability.
Currency and geopolitical exposure: Samsung and SK Hynix carry KRW currency exposure and are subject to Korea-U.S. trade and export control dynamics. The June 2026 Commerce Department export control suspension and July 1 restoration of Anthropic's own Mythos/Fable model line is a live example of how quickly policy can move in this sector.
7.3 What Has to Stay True Through 2027-2028
This report is built on 2026 disclosures. The thesis is durable, not permanent, and investors should track four things over the next 18 to 24 months to know whether the thesis is still working:
Anthropic's revenue growth has to keep outrunning its compute cost growth. Gross margins compressed to roughly 40 percent in 2026 as inference costs ran over projections. If that gap widens rather than narrows heading into a 2027 or 2028 profitability target, capex commitments across every name in this report get renegotiated downward, not just Anthropic's.
The multi-cloud, multi-chip posture has to hold. Anthropic's decision to run Trainium, TPUs, and Nvidia GPUs simultaneously is what spreads this thesis across seven names instead of one. If Anthropic consolidates toward a single primary chip supplier over the next two years, whether by choice or supply constraint, the conviction ranking in Section 6 would need to be rebuilt around whichever supplier wins that consolidation.
The HBM shortage has to remain a shortage. The Micron thesis in particular depends on the 20 to 50 percent supply-demand gap persisting into 2028. Memory is a historically cyclical business, and every prior shortage narrative in this sector has eventually resolved through capacity additions. A faster-than-expected resolution would be the single biggest risk to the Micron position specifically.
Anthropic's own IPO has to actually happen on a reasonable timeline. A confidential S-1 is an option, not a commitment. If Anthropic's listing slips materially, or if a down-round or valuation reset accompanies eventual pricing, the infrastructure names in this report should hold up better than Anthropic's own private-market valuation, since their revenue is diversified. But a genuinely disorderly Anthropic outcome, rather than a simple delay, would be a signal worth reassessing the entire thesis against.
Section 8: Appendix
Appendix A: Key Sources
Anthropic Series H and valuation: Anthropic.com (May 28, 2026); TechCrunch; CNBC; Forbes (May 4, 2026); Sacra; Tracxn. Anthropic IPO filing: CNBC, TechCrunch, Anthropic.com, NPR (June 1, 2026). Microsoft/Nvidia/Anthropic: CNBC, joint press release, Data Center Dynamics, Constellation Research, The Motley Fool (Nov 2025). Google/Anthropic TPU deal: Google Cloud, CNBC, Bloomberg (Oct 2025); Google-Broadcom expansion (Apr 2026). AWS/Project Rainier: Data Center Knowledge, Data Center Dynamics, SemiAnalysis, Introl Blog. Fluidstack: SiliconANGLE, Constellation Research. Micron/Samsung/SK Hynix: Blocks and Files, Digital Applied, Eastern Herald, TradingKey, HTX Insights, Tom's Hardware, Tech Times (Jun 2026). Marvell/Broadcom: Trefis, TheStreet, The Motley Fool, Hashrate Index, Oplexa, HeyGoTrade, TechCrunch. Equity pricing and market data: Yahoo Finance, Morningstar, CNBC, CNN Markets, TradingView, TradingKey, Kraken, MacroTrends, digrin.com (week of July 2, 2026).
Appendix B: Comparable Companies Table
Company | Anthropic Relationship | Contract Type | Confirmed? |
Amazon (AMZN) | Primary cloud/training partner | Cloud + Trainium silicon | Yes |
Alphabet (GOOGL) | Early investor + TPU supplier | Up to 1M TPUs | Yes |
Microsoft (MSFT) | Investor + Azure supplier | $30B Azure commitment | Yes |
NVIDIA (NVDA) | Investor + GPU supplier | 1 GW commitment | Yes |
Broadcom (AVGO) | Custom silicon design partner | "Titan" ASIC program | Yes (reported) |
Marvell (MRVL) | AWS Trainium co-design partner | Indirect via AWS | Yes (indirect) |
Micron (MU) | Series H investor + supplier | HBM/DRAM/SSD | Yes |
Appendix C: Valuation Snapshot (Week of July 2, 2026, VERIFY BEFORE PUBLICATION)
Company | Ticker | Approx. Price | Approx. P/E (TTM) | Note |
Amazon | AMZN | ~$242.67 | ~33x | 12% off 2026 high |
Alphabet | GOOGL | ~$362 | ~27x (29x fwd) | Cheapest of the three hyperscalers |
Microsoft | MSFT | ~$386 | ~21-22x fwd | Steepest 2026 drawdown of the three |
NVIDIA | NVDA | ~$193 | ~33x | Off May 2026 high of $235 |
Broadcom | AVGO | ~$369 | N/A, verify | Held up best in July 2 sell-off |
Marvell | MRVL | ~$244 | ~90x+ TTM, ~66x fwd | Fell over 10% on July 2 alone |
Micron | MU | ~$1,035 | N/A, verify | Down ~6% on July 2, still up sharply YTD |
These figures were pulled from multiple public sources during the week of publication and carry normal cross-source discrepancies of a few percent. Confirm against a live terminal or broker feed immediately before this report goes out, particularly for Marvell and Micron given the volatility both names showed on July 2.
Appendix D: Compute Buildout Timeline
2023: Amazon named primary cloud provider; Google's early roughly $3B investment
2024: Amazon named primary training partner; Marvell-AWS 5-year deal
Oct 2025: Google TPU expansion to roughly 1M chips; Project Rainier reaches roughly 500K Trainium2 chips
Nov 2025: Fluidstack $50B buildout; Microsoft/Nvidia $30B Azure plus investment deal
Feb 2026: Series G, $30B at $380B valuation
Apr 2026: Google-Anthropic-Broadcom multi-gigawatt TPU expansion
May 2026: Series H, $65B at $965B valuation; Micron/Samsung/SK Hynix join
Jun 2026: Confidential S-1 filed; Micron memory deal formalized
Jul 2026: Report publication date; broad semiconductor sell-off in the days ahead of publication (MRVL -10.3%, MU -5.7%, Philadelphia Semiconductor Index -5% on July 2 alone)
Hawkmont Research is free. Every note, every report, no paywall. Join the desk. Join 100,000+ readers.
No spam.
Disclosures and Important Notices
Hawkmont Research is an independent research firm. The authors of this report do not hold positions in any of the securities mentioned at the time of publication. This report is produced for informational and educational purposes only and does not constitute investment advice, a solicitation, or an offer to buy or sell securities. Past performance does not guarantee future results. Investing involves risk including the potential loss of principal.
International securities carry additional risks including currency fluctuation, political risk, and differences in financial reporting standards. Large cap securities are not immune to volatility, and single customer estimates cited in this report (for example, Anthropic's contribution to hyperscaler growth) are third party analyst approximations, not company disclosed figures.
All equity pricing and valuation data in this report was gathered from public sources during the week of July 2, 2026, and should be independently verified before publication or use in any investment decision. Readers should conduct their own due diligence and consult a qualified financial advisor before making investment decisions.
No compensation was received from any company mentioned in this report. Hawkmont Research has no investment banking relationships with any company mentioned herein.
Hawkmont Research | July 3, 2026 Research is conflict-free. No sell-side affiliations. No advertiser relationships.




