Anthropic (NASDAQ: ANT) Initiation: Frontier Leader, Low Margin of Error; Initiate with HOLD

Anthropic leads the frontier and is capturing a large, early-stage enterprise market, but price cuts, thin margins and rising share supply leave little room for error at the USD83 IPO price. We initiate with HOLD and a target price of USD75, and prefer semiconductor ETFs for exposure to AI spending.

iFAST Research Team
iFAST Research Team19 Nov 2026 82 Views
Anthropic (NASDAQ: ANT) Initiation: Frontier Leader, Low Margin of Error; Initiate with HOLD

Key Points

  • We initiate Anthropic at HOLD with a USD75 target price, implying 10% downside from the USD83 IPO price, as its leadership is already priced in.
  • Anthropic’s share of enterprise LLM API spend rose to 55% from 40%, and Fortune 500 customers still spend under 0.1% of total expenses on it.
  • Competition caps upside: price cuts took about 3.5% off quarterly revenue, while the adjusted operating margin fell to 14.5% in 3Q26 from 19.3%.
  • Diluted shares sit 40% above basic, and lock-up releases lift shares available for sale to over three times the IPO float before day 180.
  • For AI exposure, we prefer semiconductor ETFs such as SMH and 3119, which benefit from the labs’ compute spending regardless of which lab leads.

Executive Summary

We initiate coverage on Anthropic with a HOLD rating and a target price of USD75, based on 9.0x FY27E EV/sales, implying -10% downside from the USD83 IPO price. While we see Anthropic as the leading frontier lab, we believe the IPO price already reflects this position, with growth tied to each model cycle and margins yet to benefit from its scale.

Our positive view is centred on an expanding market that Anthropic is equipped to capture. Fortune 500 customers still spend less than 0.1% of their total expenses on Anthropic, while its share of enterprise LLM API spend rose to 55% in 2026 from 40% in 2025. Moreover, with an accelerating development loop, multi-year compute agreements across five chip platforms and about USD232bn of cash after the IPO, we see little risk that capacity or funding will hold back its growth.

That said, we expect upside to be capped by three factors. First, competition is intense, with the lead at the top changing hands within weeks and Anthropic's recent gains supported by price cuts that took about 3.5% off quarterly revenue, pointing to weaker pricing power. Second, competition and committed compute costs limit how much of its growth reaches its margin, with the adjusted operating margin declining to 14.5% in 3Q26 from 19.3% in 2Q26, and we expect it to remain in the low-to-mid teens through FY27. Third, public shareholders own less than the headline share count suggests, with diluted shares 40% above basic shares and lock-up releases adding supply through 2027.

Overall, at 10.2x FY27E EV/sales on a fully diluted basis, Anthropic trades above Microsoft at 8.9x, which we believe already prices in much of its growth. For investors seeking exposure to AI labs, we prefer semiconductor ETFs such as VanEck Semiconductor ETF (SMH) and Global X Asia Semiconductor ETF (3119), which benefit from the labs’ compute spending regardless of which lab leads.


Company Overview

Revenue scales with the work customers delegate to Claude

Anthropic, the developer of Claude, is the first standalone frontier AI lab to seek a public listing, in an offering that ranks among the largest on record. It has also become one of the fastest-growing companies in technology, with run-rate revenue reaching USD81bn in October from USD9bn at end-2025. Anthropic earns its revenue mainly from the work Claude processes for customers, from answering a single question to running an agent that writes and tests software over several hours. This work is sold through its API, its own applications such as Claude Code and Cowork, and packaged solutions for regulated industries.

Consumption-based fees accounted for 79% of 9M26 revenue, with subscriptions making up the remaining 21%. Unlike seat-based software, where revenue is capped by the number of licences sold, consumption-based pricing ties Anthropic’s revenue to the volume of work customers delegate to Claude, and the company attributes much of its expansion to agentic products such as Claude Code. In 3Q26, Claude Enterprise users without Claude Code spent less than USD75 a month, while the top 5% of users with Claude Code spent about 20 times as much, as an agent working over several hours consumes far more tokens than a chat query.

However, consumption-based revenue offers limited visibility, as customers can scale usage down as quickly as they scaled it up. For the same reason, we read Anthropic’s headline run-rate revenue, which annualises the last four weeks of consumption alongside current subscriptions, as a measure of current usage more than of recurring revenue.

Table 1: Anthropic’s product layers.

Layer

Products

Monetisation

Models

Mythos and Fable (frontier), Opus, Sonnet, Haiku

Per token, priced by tier

Platform

API, tool use, memory, Agent SDK, Managed Agents, Model Context Protocol

Per token consumed by customers’ agents

Applications

Claude Code, Cowork, Security, Design, Science; Claude in Microsoft 365, Chrome and Slack

Subscriptions (Pro, Max, Team, Enterprise) plus usage

Vertical solutions

Financial services, healthcare, life sciences, legal, government

Enterprise contracts

Source: Company, iFAST compilations.

Figure 1: Consumption still drives about four-fifths of revenue.

Source: Company, iFAST compilations.

Figure 2: Customers spending more than USD100k a year rose to over 11,900.

Source: Company, iFAST compilations.


Products sold through multi-channel distribution

About 80% of Anthropic’s revenue comes from enterprises, which it reaches through three channels. It sells directly, supported by Applied AI teams that help customers deploy Claude, and through partners such as Accenture, Deloitte, Salesforce and Snowflake. Moreover, Claude is sold through the three major cloud marketplaces, where enterprises can generally buy it against their existing cloud commitments, while Anthropic keeps the customer contract and sets the price. As such, close to half of Anthropic’s revenue reaches customers through the cloud providers, with the marketplaces billing 48% of 1H26 revenue.


A development loop that sustains product leadership

Claude now writes more than 90% of the code contributed to Anthropic’s codebase, and Anthropic estimates that the value of its internal use of Claude has exceeded its monthly cash personnel costs since April 2026. As such, each generation of Claude increasingly does the work of building the next, allowing Anthropic to release five frontier models in 2026, while only a five-fold rise in headcount since 3Q24. Moreover, wider use generates feedback that improves each new model, while faster growth funds the compute needed to train it. That said, the pace also depends on Anthropic’s Responsible Scaling Policy, which has already led it to restrict access to its most capable model, Mythos, at a cost to near-term revenue.

Figure 3: Claude now contributes to building the next generation of Claude.

Source: Company, iFAST compilations.


9M26 Financial Performance: Business Is Scaling, but Growth Is Slowing and Margins Remain Thin

In 3Q26, revenue grew 47% q/q to USD17.0bn, more than 12 times the level a year earlier, although growth slowed from 144% q/q in 2Q26 as usage growth eased in the months between major model releases. Notably, Anthropic has been profitable on an adjusted basis since 2Q26 and generated operating cash flow of USD6.6bn in 9M26. That said, the adjusted operating margin declined to 14.5% in 3Q26 from 19.3% in 2Q26, as compute costs rose 77% q/q against revenue growth of 47%, and the company remained loss-making at the operating level after SBC and non-cash charitable contributions.

Below the operating line, the 9M26 net loss of USD104bn mostly reflects non-cash fair-value changes on convertible notes and SAFEs that convert into equity at the IPO, and a further USD10.3bn on other financial instruments. Excluding these, Anthropic would have reported a positive net profit of about USD0.6bn in 9M26. Meanwhile, the listing will trigger about USD14.2bn of cumulative SBC on RSUs in 4Q26, keeping reported losses large in the near term.”

Table 2: Quarterly results (USD million).

1Q26

2Q26

3Q26

Revenue

4,731

11,555

16,956

Growth, q/q

132%

144%

47%

Compute and infrastructure

3,387

5,558

9,863

Compute, % of revenue

71.6%

48.1%

58.2%

Gross margin after all compute

28.4%

51.9%

41.8%

Adjusted operating income

(617)

2,235

2,463

Adjusted operating margin

(13.0%)

19.3%

14.5%

Operating income (GAAP)

(1,667)

974

(567)

Source: Company, iFAST compilations. Adjusted operating income excludes SBC and non-cash charitable contributions.

Figure 4: Revenue reached USD17.0bn in 3Q26, with q/q growth slowing to 47%.


Industry Overview

A large market, contested by a growing number of labs

The market for AI models remains large and at an early stage. Knowledge work, the activity that AI models aim to assist or automate, was worth more than USD40trn globally in 2025, while cloud computing, the closest precedent for a new technology category, grew from zero to about USD1.9trn of annual spending between 2006 and 2025. Moreover, demand is currently limited more by the supply of compute than by customer appetite, as agentic workloads require several times the compute of a chatbot query. That said, the number of labs competing for this demand continues to grow, from frontier labs funded by profitable parent companies or by capital markets to Chinese developers releasing open-weight models at a fraction of the price. As such, we believe the labs’ earnings will depend mainly on how long a capability lead lasts, what each lab pays for compute and how far prices fall.

Table 3: Addressable market.

Measure

Size

Economic value of knowledge work, 2025

>USD40trn

Cumulative economic impact of AI solutions by 2031 (IDC)

USD22.5trn

Global IT spending, 2006 to 2025

USD1.2trn to USD4.3trn

Cloud computing spending, 2006 to 2025

Zero to USD1.9trn

Anthropic run-rate revenue, October 2026

USD81bn (~0.2% of knowledge work)

Source: Company, IDC, iFAST compilations.


Capability leads change hands quickly

Among peers, Claude Fable 5.2, released on 9 November, leads the Artificial Analysis Intelligence Index with a score of 65, ahead of Google’s Gemini 4 Carbon and OpenAI’s GPT-6 Astra, and Anthropic prices it accordingly at USD40 per million output tokens, four times Gemini 4 Carbon’s introductory price. That said, the lead has changed hands twice in two weeks. When Artificial Analysis relaunched its index on 27 October, Opus 5.5 led with 58, before Gemini 4 Carbon, initially restricted to selected users, took the top spot the following day and held it for 12 days until Fable 5.2's release. Moreover, at maximum effort Fable 5.2 costs about USD27 per task against USD8 for Carbon and USD6 for Astra. . Hence, we believe Anthropic’s leadership is narrow and comes at a premium price that rivals can undercut within weeks.

Table 4: Frontier lab comparison.

Lab

Top model (index score)

Output price, USD/m tokens

Funding

Compute

Anthropic

Fable 5.2 (65); Opus 5.5 (59)

40 / 20

Revenue, public equity, supplier facilities

Rented or leased

Google

Gemini 4 Carbon (60)

10 (introductory)

Search and cloud profits

Own TPUs

OpenAI

GPT-6 Astra (57); GPT-6.1 Sol (55)

50 / 10

Private capital; IPO targeted for 2027

Mostly contracted; own inference chip in development

SpaceXAI

Grok 4.7 (35)

6

SpaceX

Own data centres

Moonshot

Kimi K3.1 Pro (46)

n/a

Domestic market

Domestic chips

Mistral

Mistral Large 4 (26)

2.10

n/a

n/a

Source: Artificial Analysis Intelligence Index (10 November 2026), company pricing, iFAST compilations.

Figure 5: Anthropic leads the index, at the top of the price range.


A handful of frontier labs, funded and equipped differently

The labs also differ in how they secure compute. Google trains and serves its models on its own TPUs, and OpenAI is developing its own inference chip alongside its purchase commitments, whereas Anthropic rents or leases all of its capacity from Google, Amazon, Microsoft, SpaceXAI, Broadcom and smaller providers. While the multi-chip approach gives Anthropic flexibility and bargaining options, it also means Anthropic pays its suppliers' margins on each unit of compute. As such, we see Anthropic at a cost disadvantage to Google should competition at the frontier turn into a sustained price war.


Open-weight models are taking the price-sensitive end of the market while OpenAI is competing on cost

Chinese open-weight models now account for more than 40% of traffic on OpenRouter, and while OpenRouter’s users tend to be more cost-conscious than large enterprises, these models’ capability is converging, with several matching closed models on coding benchmarks at a fraction of the price. Once a model's weights are public, any host can serve it, pushing pricing towards the cost of compute plus a thin margin. Hence, we expect open-weight models to take the price-sensitive end of demand, leaving the closed labs to compete for the top tier, where Anthropic is strongest but where its premium is most exposed.

Similarly, OpenAI has moved towards cost leadership, adding cheaper model tiers and discounting to win enterprise accounts. As such, we expect competition for large enterprise accounts, Anthropic's core market, to intensify.

Overall, Anthropic holds the premium position among the labs on capability, price and enterprise mix. That said, it is one of the few labs that both rents all of its compute and relies entirely on its models for revenue, which leaves its margins more exposed should the industry compete on price.


Investment Theses

An expanding market for Anthropic to capture (Positive)

In an industry where demand is still at an early stage, the key question for Anthropic is how much of that growth it can capture. Enterprise adoption has further to run, with Fortune 500 customers still spending less than 0.1% of their total expenses on Anthropic, while agentic use is starting to extend beyond coding into general knowledge work, where Anthropic notes that Cowork has been adopted faster than Claude Code at the same point after launch. As such, we see a large room for Anthropic to grow its customer base as agents take on longer and more complex tasks.

Anthropic’s share of enterprise LLM API spend rose to 55% in 2026 from 40% in 2025, according to a Menlo Ventures survey of US enterprise buyers, reflecting its strength in coding and other complex tasks, where model quality determines whether work can be delegated to AI. At the same time, its development loop is accelerating, with Anthropic reporting that Claude’s success rate on hard internal engineering tasks rose to about 90% in September, up about 50ppt in nine months. Hence, we expect Anthropic to keep releasing frontier models roughly every quarter, keeping it among the leaders as enterprise spending scales. On top of that, Anthropic has secured the compute and capital to sustain this pace. Its multi-year compute agreements span five chip platforms and are backed by its largest suppliers, several of which have also invested in Anthropic or committed financing to it. Together, these agreements would give Anthropic a total of about 15 GW of computing capacity by the end of 2028, based on estimates by MBI Deep Dives.

Figure 6: Anthropic's share of enterprise LLM API spend rose to 55% in 2026.

Source: Menlo Ventures (2026 enterprise AI survey), iFAST compilations. Data as of 3 November 2026.

Figure 7: Compute secured across five chip platforms.


Growth depends on leading each model cycle, increasingly through price cuts (Negative)

Anthropic's recent gains have increasingly been won on price. By August, spending on Fable 5, Anthropic's most expensive model, had plateaued at about 11% of its total as customers moved to cheaper models and OpenAI competed on cost, prompting Anthropic to launch Opus 5.5 at about 40% lower cost per task than Opus 5 and to price Fable 5.2 20% below Fable 5.1. Price reductions and discounts took about 3.5% off quarterly revenue in both 4Q25 and 3Q26, equivalent to roughly 14% a year, hence Anthropic now needs usage to grow by about that much merely to keep revenue flat. This aligns with the trend we highlighted in our previous report, “AI’s Shift from Optimisation to Maximisation: Implications for AI Models”, in which AI spending is shifting from maximising token usage to optimising usage for each task.

As such, we believe the reacceleration in October reflects weaker pricing power as much as stronger demand, with each unit of work earning less than before.

On top of that, switching costs at the model layer remain low. Unlike systems of record such as Salesforce, where years of customer data and workflows sit within the platform, Anthropic’s models sit on top of its customers’ systems, and moving to another model largely involves changing an API call and retesting prompts. Developers increasingly choose a model for each task through coding tools and model routers, allowing customers such as Rippling and Replit to move workloads between providers within months, while remaining performance obligations cover around two months of the October run-rate. On top of that, Amazon, Google and Microsoft, which bill nearly half of Anthropic’s revenue through their marketplaces, can see its pricing and commercial terms and may favour their own models.

We expect revenue growth to slow further to about 25% q/q in 4Q26, from 47% in 3Q26, as Fable 5.2 lifts November and December revenue but the usual holiday dip in usage offsets part of the gain. With our forecasts already assuming a major model roughly every three months and price cuts of about 14% a year, which translate into revenue growth of around 47% a year to FY31, we see a lengthening of the release cycle or deeper price cuts as the main risk to our revenue estimates.

Figure 8: Run-rate growth has tracked the release cycle.


Margins squeezed between falling prices and committed costs (Negative)

Meanwhile, even as Anthropic sustains its lead, we expect competition to limit how much of its growth reaches its margin, as efficiency gains are passed on to customers while costs are largely committed. In 3Q26, compute and infrastructure costs rose 77% q/q against revenue growth of 47%, partly from about USD5bn of short-term capacity bought at market rates, as rental prices for Nvidia’s B300 chips have risen by more than half since late May while token prices have fallen by a similar amount. Hence, the Jevons paradox explains why Anthropic’s revenue keeps growing, but not why its margin should expand.

Moreover, rising competition pressures Anthropic’s ability to maintain premium pricing. Rivals continue to price aggressively, with Gemini 4 Carbon introduced at a quarter of Fable 5.2's price and open-weight models matching closed models on coding benchmarks at a fraction of the cost. As such, as the capability gap between competing models narrow, this will limit the company’s ability to charge premium pricing, as shown via the recent price cuts on Opus 5.5 and Fable 5.2.

On top of that, the reported gross margin of 86% in 9M26 excludes the compute used to train new models, and treating training as a recurring cost, as most revenue comes from models less than six months old, lowers the 3Q26 gross margin to 42%. Most of this compute is also contracted years ahead, with non-cancellable contracts covering 73-87% of the compute we forecast for FY27-FY30; hence, a slower release cycle would weigh on earnings far more than on revenue.

In summary, we expect the adjusted operating margin to remain in the low-to-mid teens through FY27, before rising towards 30% by FY31 as the cost of training each model is spread over a larger revenue base, and levelling off in the low 30s as serving costs rise with usage. However, compute costs could fall faster than we assume once supply catches up. Even so, with switching costs at the model layer still low, we expect Anthropic to face difficulties in maintaining high gross margin as capability gap between models narrows.

Figure 9: In 3Q26, compute costs grew faster than revenue, lowering the adjusted operating margin.

Figure 10: Token prices fell by more than half while GPU rental rates rose.

Source: Silicon Data, iFAST compilations.

Figure 11: Training and research compute absorbs most of the gross profit.

Figure 12: Non-cancellable contracts cover 73-87% of the compute we forecast for FY27-FY30.


Public shareholders own less than the share count suggests, with share supply rising through 2027 (Negative)

Beyond the business, investors at the IPO price own a smaller share of Anthropic’s future earnings than the headline share count suggests, while share supply rises steadily through 2027. Once employee equity, Amazon’s convertible notes and the donation reserve are included, the diluted share count is 40% above the basic count, hence the USD83 midpoint values Anthropic at about USD1.54trn, against roughly USD1.10trn on basic shares. While our target price is already set on the diluted count, we believe investors who value Anthropic on basic shares will underestimate what they are paying.

Moreover, dilution could rise further, and most heavily when the business is weakest. While supplier investment secures Anthropic’s capacity, part of its compute bill is effectively paid in equity, with Google, Nvidia, Microsoft and other suppliers buying USD26bn of stock alongside the IPO, about 31% of net proceeds. On top of that, Broadcom can lend Anthropic up to USD42bn to fund its chip lease and be repaid in shares at a 10% discount to the market price, widening to 15% if Anthropic misses its revenue targets. Hence, the weaker the share price and growth, the more shares Broadcom would receive, leaving shareholders most diluted when the investment case is already under pressure.

Lastly, share supply rises steadily after listing. While the small float may support the shares in the early weeks of trading, non-founder directors, officers and employees can sell up to 10% of their locked-up shares from the first trading day, with further releases following each results announcement through 2027, lifting the shares available for sale to more than three times the IPO float before day 180. As such, we expect the rising supply to weigh on the shares through 2027, as many holders acquired their shares at a fraction of the IPO price, with the pressure likely greatest after results announcements, when lock-up releases occur and a weaker quarter would meet fresh supply.

Figure 13: Fully diluted shares are 40% above the basic count.

Table 5: Lock-up release schedule.

Holder group

Shares released

Trigger

Approximate timing

Private placement

All

No lock-up

From listing

Non-founder directors, officers and employees

Up to 10%

First trading day

From listing

A further 35%

Two trading days after FY26 results

Early 2027

The rest

Earlier of 1Q27 results or day 180

Around May 2027

Founders and holders above 0.5%

Staggered

Earlier of 3Q27 results or day 365

From around November 2027

Source: Company, iFAST compilations.


Valuation: Fairly Priced Against Software Peers; HOLD With a TP of USD75

We value Anthropic on EV/forward sales against listed software companies, as its adjusted operating margin remains volatile, with compute costs running ahead of revenue, which makes earnings-based multiples less meaningful at this stage. At the USD83 IPO price, Anthropic trades at 10.2x FY27E EV/sales on a fully diluted basis, above Microsoft and other established application companies such as Salesforce, ServiceNow and Adobe at 3-5x. That said, it trades below high-growth names such as Snowflake and Datadog at 10-12x, and well below Palantir and Cloudflare at 27-29x, despite our forecast of 135% revenue growth in FY27, far above that of these companies.

On that basis, we believe the current valuation is broadly justified. We apply 9.0x FY27E EV/sales, in line with Microsoft, as we see Anthropic’s far faster growth largely offset by its thinner margins and lower revenue visibility, with remaining performance obligations covering around two months of its October run-rate.

Hence, we issue a HOLD call on Anthropic with a target price of USD75, implying -10% downside from the USD83 IPO price. With the offering upsized and priced above its marketed range of USD70-78, and a small share of the company floated, the shares may trade above the range at listing, which we would see as running ahead of fundamentals. Moving forward, we would turn more positive should run-rate growth hold up in months without a major release, indicating that Anthropic’s revenue is becoming less dependent on each model cycle, or should the adjusted operating margin track above our FY28E forecast of 22.5%. Conversely, a release that falls behind rivals, deeper price cuts, or a further decline in margin ahead of the main lock-up release after the 1Q27 results would weigh on the shares.

For investors seeking AI exposure, we prefer semiconductors over a single AI lab. While competition among the labs passes efficiency gains on to customers, compute suppliers retain pricing power as demand outstrips supply. As such, we see VanEck Semiconductor ETF (SMH) and Global X Asia Semiconductor ETF (3119) as a more diversified way to capture AI spending, although both remain exposed should the labs’ revenue fall short of their compute commitments.

Figure 15: Anthropic trades in line with Microsoft on FY27E EV/sales.

Source: Bloomberg Finance L.P., iFAST estimates. Anthropic at the USD83 IPO price on FY27E revenue. Data as of 18 November 2026.

Table 6: Target price derivation.

USD bn

FY27E revenue

128.1

Target EV/sales (x)

9.0

Enterprise value

1,152.9

Cash and short-term investments, 30 Sep 2026

131.8

IPO and private placement, net cash at USD83

100.2

Option exercise proceeds

5.2

Equity value

1,390.1

Diluted shares (bn)

18.55

Target price (USD)

75

Source: iFAST estimates. Net cash of USD237.3bn comprises cash and short-term investments at 30 September 2026, net IPO and private placement proceeds at USD83 (956.4m primary shares and the USD26bn placement) and option exercise proceeds; Anthropic has no drawn debt. The underwriters' option over 147m shares is not assumed exercised. Target price on 18.55bn fully diluted shares.

Table 7: Valuation summary.

USD million

FY25

FY26E

FY27E

FY28E

Revenue

4,585

54,402

128,095

193,844

Growth, y/y

1,088%

1,087%

135%

51%

Adjusted operating margin

(105.9%)

11.3%

15.5%

22.5%

EPS (USD)

0.53

1.47

2.38

EV/sales at USD83

23.9x

10.2x

6.7x

EV/sales at target price

21.2x

9.0x

6.0x

Target price (USD)

75

Upside potential

-10%

Source: Company, iFAST estimates. EPS on diluted shares (18.5bn in FY26E, rising with new employee grants), excluding SBC and non-cash fair-value changes on convertible notes and SAFEs; EPS includes interest income on the IPO cash.


Upside risks

  1. Each release adds more revenue than we assume. Our forecasts already assume a major launch every three months. Should each launch keep adding run-rate revenue at October’s pace, revenue would run well ahead of our forecasts.
  2. Switching costs rise faster than we expect. Should Managed Agents and Cowork embed Claude in customers’ workflows, revenue would hold up better between model releases, supporting a higher multiple.
  3. Compute costs fall faster than we expect. A larger share of capacity on Google’s TPUs, Amazon’s Trainium and Broadcom’s custom chips, or a fall in GPU rents once supply catches up, could let Anthropic keep more of the savings and lift margins above our forecasts.
  4. A limited free float supports the shares. With a small share of the company floated, strong demand for the offering could support the share price in the near term even if fundamentals soften.


Downside risks

  1. A model release falls behind rivals. With switching costs at the model layer low, customers could move workloads within months, as Google’s loss of enterprise API share in 2026 showed
  2. Price competition intensifies. OpenAI’s shift towards cost leadership and the rise of open-weight models could force deeper price cuts than the roughly 14% a year seen so far, weighing on both revenue and margins.
  3. Policy and safety constraints slow growth. Anthropic’s designation as a supply-chain risk by the US Department of Defense, which it is challenging in court, and its own Responsible Scaling Policy, which has already restricted access to Mythos, could limit its customer base or slow its release pace.
  4. Part of the revenue base proves less durable. About 5-7% of 9M26 revenue came from customers later found breaching usage terms, and the top five customers accounted for about a fifth of 3Q26 revenue.
  5. Share supply and dilution rise after weak results. With lock-up releases timed after results announcements through 2027, and Broadcom’s facility repayable in shares at a wider discount if revenue targets are missed, a disappointing quarter could meet both fresh insider supply and further dilution.


Disclaimers:

This research report was prepared with the assistance of artificial intelligence (AI) tools. iFAST Financial Pte Ltd does not rely exclusively on AI for content generation; the content of this report – including all investment theses, ratings, price targets and conclusions – has been independently reviewed and verified by the research analyst(s) to ensure accuracy and professional integrity.

For specific disclosure, at the time of publication of this report, IFPL (via its connected and associated entities) and the analyst who produced this report hold a NIL position in the abovementioned securities.


All materials and contents found in this site are strictly for general circulation and informational purposes only and should not be considered as an offer, or solicitation, to deal in any of the funds or products found/identified in this site. While iFAST Financial Pte Ltd ("IFPL") has tried to provide accurate and timely information, there may be inadvertent delays, omissions, technical or factual inaccuracies and typographical errors. Any opinion or estimate contained in this report is made on a general basis and neither IFPL nor any of its servants or agents have given any consideration to nor have they or any of them made any investigation of the investment objective, financial situation or particular need of any user or reader, any specific person or group of persons. You should consider carefully if the products you are going to purchase are suitable for your investment objective, investment experience, risk tolerance and other personal circumstances. If you are uncertain about the suitability of the investment product, please seek advice from a financial adviser, before making a decision to purchase the investment product. Past performance is not indicative of future performance. The value of the investment products and the income from them may fall as well as rise. Opinions expressed herein are subject to change without notice. In respect of any matters arising from, or in connection with the said research analyses or research reports, recipients of the report are to contact IFPL at 10 Collyer Quay, #26-01 Ocean Financial Centre Building, Singapore 049315, or by telephone at +65 6557 2853. Where the report contains research analyses or research reports from a foreign research house and if the recipient of such research analyses or research reports is not an accredited investor, expert investor, institutional investor or an ex-accredited investor, IFPL accepts legal responsibility for the contents of such analyses or reports to such persons only to the extent as required by law. Please note that only certain security(ies) herein are available to all investors, while the rest are only available for certain persons to invest in, such as Accredited Investors (as defined in the Securities and Futures Act) or one who invests at least S$200,000 (or its equivalent currency) per transaction. To qualify as an Accredited Investor, one needs to submit a declaration form and certain relevant supporting documents, according to iFAST’s prevailing policies and procedures.

Please read our full disclaimers on the website at ( https://fsm.global/sg/policies/328125/investment-account-terms-&-conditions).

iFAST Financial Pte Ltd (IFPL) (registered address: 10 Collyer Quay #26-01 Ocean Financial Centre Singapore 049315, Telephone: 6557 2000) holds the Financial Advisers Licence issued by the Monetary Authority of Singapore ('MAS') to conduct regulated activities of advising on securities, marketing of collective investment schemes and arranging of any contract of insurance in respect of life policies, other than a contract of reinsurance and the Capital Markets Services Licence issued by the MAS to conduct regulated activities of dealing in securities and providing custodial services for securities. While IFPL has made every effort to ensure the independence of the report's contents, IFPL's nature of business is such that IFPL and its connected and associated entities together with their respective directors, officers and staff may be involved in providing dealing or investment-related services in the abovementioned securities, and have taken or may take positions in the securities mentioned in this report, and may also act as the principal for any buy or sell trades.