Chunghwa Telecom 中華電信: Taiwan's State-Anchored Telecom Giant Bets on AI

I. Cold Open & Roadmap

On the afternoon of August 5, 2026, a group of institutional investors will dial into a one-hour teleconference from Taipei to hear how 中華電信 Chunghwa Telecom performed in the second quarter.²¹ It is an unglamorous ritual — no stage, no product launch, no keynote. A century-old phone company reads out its numbers, takes a handful of questions from sell-side analysts, and hangs up.

And yet the numbers that will be read out belong to one of the stranger stories in Asian infrastructure right now. This is a company in which the Republic of China government, through its Ministry of Transportation and Communications, still owned roughly 35.29% of the outstanding shares as of the end of 2024 — a stake large enough that every non-independent board member is a government appointee.² It is a company whose tariffs, spectrum, and universal-service obligations are supervised by a regulator, the 國家通訊傳播委員會 National Communications Commission, that was carved directly out of its own former parent bureau.² By every structural marker, this should be a sleepy rate-regulated utility.

Instead, it has spent the last two years repositioning itself as what its own executives call an "AI-Native Telco" — building liquid-cooled data halls rated for racks drawing 100 to 200 kilowatts apiece, laying trans-Pacific submarine cable, buying multi-orbit satellite capacity, and pushing capital expenditure up 14.6% in a single year to NT$31.91 billion for 2026.¹³⁴

The scale, first. Full-year 2025 consolidated revenue reached NT$236.11 billion, up 2.7% — an all-time high for the company. Net income was NT$38.69 billion and earnings per share NT$4.99, both eight-year highs.¹ In Taiwan's mobile market, Chunghwa held about 39% of subscribers and over 40% of revenue, a lead it has now sustained for years.⁹ The shares trade in Taipei under 2412 and the American depositary shares trade in New York under CHT, a dual listing dating back to October 2000 and July 2003 respectively.²

Here is the framing worth carrying through the rest of this story: Chunghwa Telecom is really two businesses stapled together, and they behave nothing alike.

The first is a national utility. Mobile, fixed broadband, fixed voice, television. It grows at low single digits, throws off enormous cash, employs a large unionised workforce, and answers to a regulator and a ministry. Its job is to be boring and to fund the dividend.

The second is an infrastructure venture: internet data centres retrofitted for AI workloads, sovereign and hybrid cloud, cybersecurity, submarine cable capacity, satellite, edge computing. Group ICT revenue grew 25% year over year in the first quarter of 2026.¹² This part of the company is growing four to ten times faster than the first, is far more capital-hungry, and is competing — at least notionally — in the most crowded global capital-allocation race of the decade.

The interesting question is not whether the utility works. It plainly does. The question is whether the second business is a real option worth underwriting, or a growth narrative layered on top of a cash cow because the cash cow alone no longer excites anyone. That question has a specific test attached to it, and we will come back to it repeatedly: has management disclosed anything about the unit economics of its AI data centres — utilisation, contracted capacity, return on the incremental capital — or only about the capital going in?

The roadmap from here. We start with how a government bureau became a listed company without ever quite becoming independent. We anatomise the core telecom business and the 2023 consolidation that reshaped Taiwan's market from five carriers to three. We treat the 2020 5G spectrum auction as the company's defining capital-allocation event and grade it like an acquisition. We meet the management team installed in September 2024. We interrogate the AI strategy on its own evidence. We map the geopolitical overlay that makes this company's cables a national-security asset. Then financials, the bull and bear cases through Porter and Helmer, the risk radar, and what to watch.

Start where the company started: not as a company at all.

↑ Back to Top

II. From State Monopoly to Public Company

For most of the twentieth century, telephony in Taiwan was not an industry. It was a department.

The entity that placed your calls, ran the lines, and set the prices was the Directorate General of Telecommunications — a bureau inside the Ministry of Transportation and Communications, staffed by civil servants, budgeted through the state, and accountable to the legislature rather than to shareholders. There was no competitor because competition was not a concept that applied. There was no share price because there were no shares. Network build-out followed development policy, not demand curves.

Then, on July 1, 1996, the bureau's operating arm was cut loose and incorporated as Chunghwa Telecom Co., Ltd., operating under its own dedicated statute.² Note the mechanism, because it matters: Chunghwa was not privatised by legislation that dissolved state control. It was corporatised — given a balance sheet, a board, and a legal identity — while the state remained the owner. Privatisation, in the sense of the government actually giving up majority ownership, would take another nine years.

What followed was an unusually deliberate, unusually patient sequence. Common shares listed on the Taiwan Stock Exchange under the code 2412 in October 2000. Domestic public offerings ran in tranches from August 2000 through July 2003. An international offering in the form of American depositary shares was completed on July 17, 2003 and listed on the New York Stock Exchange under CHT.² Then, on August 9, 2005, the ministry auctioned shares domestically and closed a second international offering the following day. When those transfers settled on August 12, 2005, government ownership fell below 50% and the privatisation plan was formally complete.²

Nine years, four separate capital-markets events, two exchanges, and a legal statute to make it all work. Compare that to the more common Asian pattern of a single dramatic IPO followed by continued state majority control, and Chunghwa's path looks genuinely careful. It also had a hard consequence for employees: at privatisation the company settled all existing defined-benefit pension obligations in full, and every continuing employee was deemed to have started employment on August 12, 2005 for seniority purposes under the post-privatisation plans. The ministry took over annuity payments to everyone who had already retired.² A clean break, and an expensive one — but it meant the listed company was not carrying a century of unfunded bureau liabilities into the public market.

Here is the part that makes the whole story work differently from a normal privatisation narrative. Twenty-one years after the government crossed below 50%, it has not kept going. As of December 31, 2024, the ministry held approximately 35.29% of outstanding common shares, and the broader ROC government position — ministry plus government-linked entities — was around 40.57% as of February 28, 2025.² The company's own risk disclosure is admirably blunt about what that means. Because all non-independent board members are ministry appointees, the government "may continue to have the ability to control" matters including the composition of senior management, the timing and distribution of dividends, the approval of the annual budget, the election of a majority of directors, and the company's business direction.²

That is not a residual legacy holding. That is de facto control at roughly a third of the equity, exercised through board appointment rather than voting majority. It is the cheapest form of control there is.

Layered on top is a second structural constraint that most utility investors never encounter. Under the Telecommunications Management Act, effective July 1, 2020, foreigners may directly hold no more than 49% of Chunghwa's shares, and direct plus indirect foreign holdings may not exceed 60%.² The penalty for breach is not a fine — or not only a fine. The network establishment approval granted to the company may be abolished. And the company concedes it cannot police this: it cannot control who buys its shares, cannot block transfers, and cannot force holders to sell. It could be sanctioned "even if there is no fault of our own."² In practice the caps are far from binding — foreign direct holdings sat at roughly 15.67% as of April 2, 2025² — but their existence tells you what the state thinks this asset is. It is not a utility that happens to be strategic. It is a strategic asset that happens to be listed.

Regulatory authority moved too. Before March 1, 2006, Chunghwa answered to the ministry and to its own former parent bureau. On that date the National Communications Commission was created under its own organisation act, with seven full-time commissioners nominated by the Premier and confirmed by the legislature for four-year terms.² In August 2022, a Ministry of Digital Affairs was launched to absorb part of what the NCC, the ministry, and the Executive Yuan had been doing.² The regulatory architecture around Chunghwa has therefore been rebuilt twice in twenty years, always by political process.

For an investor, the implication is uncomfortable but clarifying. Every major decision this company makes — how much to spend, what to spend it on, how much to pay out, who runs it — passes through a filter that includes Taipei's policy preferences. Sometimes that filter is a tailwind: national-champion status, strategic capex support, guaranteed relevance. Sometimes it is a tax: tariff oversight, hiring rigidity, political-cycle leadership turnover. It is never absent. Anyone modelling this business as a clean private-sector compounder is modelling the wrong company.

Which brings us to the thing the state actually built: the best network in Taiwan.

↑ Back to Top

III. The Core Business: Taiwan's Telecom Oligopoly

In December 2023, Taiwan's mobile industry stopped being a five-player market and became a three-player one. Two mergers completed in the same month: 台灣大哥大 Taiwan Mobile absorbed 台灣之星 Taiwan Star, and 遠傳電信 Far EasTone absorbed 亞太電信 Asia Pacific Telecom.² For an island of roughly 23 million people that had been supporting five mobile network operators, this was less a strategic land grab than an admission that five was always two too many.

Chunghwa did not participate. It did not need to. It was already number one, and consolidation handed it the benefit without the integration risk — the classic position of the incumbent in a consolidating oligopoly.

Start with the anatomy of what it actually sells, because the segment structure is the analytical key to this company.

The Consumer Business Group is the largest by far: mobile, fixed broadband, and MOD television sold to households. It generated NT$143.37 billion of revenue in 2025, up 2.4%.¹ This is the utility.

The Enterprise Business Group sells to companies and government: ICT projects, systems integration, data centre and cloud services, cybersecurity, IoT. It generated NT$77.24 billion in 2025, up 2.5% — but that annual figure conceals violent quarterly swings, which we will come back to.¹

The International Business Group is the smallest at NT$9.52 billion in 2025, and it actually shrank 4.1% for the year.¹ It handles international connectivity, roaming, and overseas business. Small, but strategically outsized, because this is where the submarine cables live.

Now the competitive position. In mobile, Chunghwa held roughly 11.26 million users and a 39.09% share of the mobile telecom market in early 2025, having sustained over 40% revenue share for seven consecutive quarters.⁹ Total mobile subscribers across all categories reached 13.34 million by the first quarter of 2026, up 1.7%.³ Post-paid mobile ARPU excluding IoT was NT$573, up 3.6% year over year.³

That ARPU number deserves a moment, because it carries the single most important piece of evidence about the post-consolidation market. Between roughly 2015 and 2020, Taiwan's mobile industry was one of the most value-destructive in developed Asia — five carriers, an unlimited-data price war, and ARPU grinding downward year after year. Today ARPU is rising at three to four percent annually across the industry. That is not a Chunghwa achievement. That is what happens when the two subscale players are removed from the board.

The evidence that this is industry-wide rather than company-specific is worth stating plainly, because it cuts against the easy bull argument. After its merger, Far EasTone deliberately shed 881,073 accounts in a cleanup that included fraud-related disconnections, and its ARPU rose to NT$513.1 — edging past Chunghwa's NT$512.7 on the comparable measure. Taiwan Mobile, meanwhile, took the opposite route, running a lower-ARPU, higher-volume position at NT$448.1 across 9.38 million subscribers, and delivered a seven-year high in telecom operating profit with roughly 30% year-over-year improvement and 23% net income growth.⁹

Read that carefully. The two challengers grew profits faster than the incumbent in the post-merger period, from a lower base, with self-help levers Chunghwa does not have. A merger gives you cost synergies, network decommissioning savings, and a one-time subscriber-quality reset. Chunghwa has none of those available. Its earnings growth has to come from price, mix, and new products. That is a harder game, and it is the honest counterweight to "market leader with 40% share."

Where Chunghwa genuinely does lead is in the physical plant. It emerged from the 2020 auction holding 90MHz in the 3.5GHz band and 600MHz at 28GHz, on top of 40MHz at 900MHz, 60MHz at 1800MHz, 40MHz at 2100MHz, and 60MHz at 2600MHz — a spectrum portfolio deeper than either rival's, and the widest 5G channel in the market.² Spectrum depth is not a marketing claim; it is arithmetic. A wider contiguous channel at 3.5GHz produces higher peak and median speeds for the same radio hardware and the same cell density. That is why Chunghwa consistently posts the fastest median 5G download speeds in Taiwan. It is a cornered resource in the literal sense — the frequencies exist in finite quantity, they were allocated once, and no amount of capital can conjure more.

The quieter half of the business is the fixed network, and it is arguably the better one. Broadband subscribers reached 4.45 million with ARPU of NT$818 in the first quarter of 2026, and fixed broadband revenue grew 3.0% to NT$11.81 billion.³ Note what is driving that: not subscriber growth, which was a rounding-error 0.5%, but ARPU up 2.5% as households migrate to higher-speed fibre tiers.³ This is the purest form of pricing power available in a telecom business — selling a faster version of the same connection over the same glass, at a higher price, with essentially no incremental capital. Fixed-line voice subscribers stood at 8.63 million at the end of 2025, a legacy base that shrinks slowly and profitably.¹

Chunghwa also distributes differently from its rivals. It ran 17 operations offices, 442 service centres, 14 customer-service call centres, and 240 exclusive SENAO retail stores as of the end of 2024.² For an urbanised market, that is an enormous physical footprint — and it is a legacy of the bureau era, when the mandate was to serve every township regardless of economics. It now functions as a distribution moat and a brand-trust asset, particularly among older and rural customers, and particularly for government-adjacent services. It is also a fixed-cost anchor that the challengers do not carry.

Then there is the regulator. Under the Telecommunications Management Act, obligations are tiered. General obligations bind everyone: disclosure of service conditions and network quality, separate accounting for telecom and non-telecom services, confidentiality of communications, complaint handling, record retention, and — critically — sharing the cost of universal service if annual telecom revenue exceeds a threshold. Special obligations bind those granted specific resources: free emergency calling, number portability, equal access, and mandatory information-security maintenance plans. Designated obligations can be imposed for disaster prevention, communication security and surveillance, accessibility for people with disabilities, and universal service.²

And there is a further lever. The NCC may designate an operator as having "significance in the specific telecommunications service market," which unlocks a set of asymmetric remedies: compelled disclosure of interconnection terms and pricing, prohibitions on discriminatory refusal to provide interconnection or infrastructure access, and mandated publication of template interconnection agreements. Chunghwa's own filing acknowledges it "may be regarded as the enterprise with significance."²

Translate that from regulatory language. The company that owns the last-mile copper and fibre into most of Taiwan's homes can be forced to sell access to that plant to its competitors, on terms the regulator can influence. That is the structural ceiling on incumbency here. Chunghwa's fixed-network advantage is real, but it is a leased advantage — held at the regulator's discretion, and revocable in economic effect if not in title.

So the core business is: a durable number-one position with genuine physical advantages, in a market that just got structurally healthier, run under a regulator that can cap the value of exactly the assets that make it number one. Low growth, high certainty, limited upside from here.

Which is why what happened in early 2020 mattered so much — the one time this company had to write an enormous cheque to defend the position we have just described.

↑ Back to Top

IV. Capital Deployment I: The 5G Spectrum Auction

Spectrum auctions are supposed to be boring. A regulator sets a reserve price, carriers bid in ascending rounds, and the whole thing resolves in a few days at a modest premium to the floor. Taiwan's 5G auction, which opened in December 2019 and did not finally conclude until February 21, 2020, was not boring. It was a slow-motion demolition of everyone's budget.

By the time the first phase closed, bidding had reached NT$138.08 billion. A second, positioning phase — in which carriers bid not for how much spectrum they got but for where in the band they got it — added NT$4.11 billion more, bringing the total to NT$142.19 billion.⁷⁸

The final tally by carrier: Chunghwa Telecom NT$48.373 billion. Far EasTone NT$43.042 billion. Taiwan Mobile NT$30.606 billion. Taiwan Star NT$19.708 billion. Asia Pacific Telecom NT$412 million.⁸

Five carriers, all of whom would within four years become three, collectively paying nearly NT$142 billion for the right to build networks in a market of 23 million people. Chunghwa's share alone — NT$48.37 billion — was roughly equal to its entire consolidated capital expenditure for the following two years combined.

What did it buy? Ninety megahertz in the 3.5GHz mid-band and 600 megahertz at 28GHz.² And in that final positioning phase, for an incremental NT$2.08 billion, it secured 3.42–3.51 GHz — the premium slot within the band, the one with the deepest global equipment ecosystem behind it.⁷ Far EasTone took 3.34–3.42 GHz. Taiwan Mobile landed at 3.51–3.57 GHz, a position noted at the time for thinner handset and equipment support. Taiwan Star drew 3.3–3.34 GHz, exposed to potential satellite interference.⁷

Run the arithmetic on the mid-band alone and the number is arresting: roughly NT$5.3 billion for every 10MHz of 3.5GHz spectrum. Contemporaneous industry commentary treated Taiwan's clearing price as among the most expensive 5G spectrum bought anywhere in the world at the time. Whether or not it held the literal global record, the direction is not in dispute — this was a market where everybody paid too much.

Chunghwa's chairman at the time responded to the outcome with the sort of line that sounds like confidence and reads, in hindsight, like a man boxing himself in: "Despite a higher-than-expected cost, our plan for 5G deployment remains unchanged. We aim to launch 5G service in July."⁷ The company in fact launched commercial 5G in June 2020, a month ahead of that commitment.²

Now grade it as an acquisition, which is the right lens.

The case against. Nothing about the auction created value. Every carrier got spectrum; the relative competitive positions barely moved. NT$142 billion of shareholder capital transferred from the industry to the treasury, and the industry's competitive structure was unchanged the day after. Chunghwa wrote the largest cheque of anyone and gained no share against Taiwan Mobile or Far EasTone that it did not already hold. In the narrowest sense, this was a defensive tax on incumbency — the price of not losing, paid in cash, with no revenue attached to it.

The case for. Consider the counterfactual seriously. Had Chunghwa bid conservatively and ended up with 60MHz instead of 90MHz, or with the 3.51–3.57 GHz block instead of 3.42–3.51, its entire competitive claim would have collapsed. The whole basis of premium pricing at Chunghwa is network superiority. Lose the speed leadership and you are selling an undifferentiated commodity — at which point the ARPU premium goes, and with it the earnings that fund the dividend that supports the equity story. On that reading, NT$48 billion was not a bid for growth. It was insurance on the existing franchise, and the correct question is not "what return did it earn" but "what would it have cost not to buy it."

Where the evidence has landed since. Six years on, the honest verdict is: mostly vindicated, but not because of the auction. Chunghwa still leads on network quality. Post-paid ARPU is rising, not falling.³ Mobile capital expenditure has now declined for five consecutive years and fell another 24.4% year over year in the first quarter of 2026 as the 5G build passed its peak.¹¹¹² The heavy spending phase is genuinely over, and the asset is now in harvest.

But note what did the real work. It was not the spectrum. It was the December 2023 consolidation, which removed two price-cutting competitors and let the whole industry raise prices. Chunghwa paid NT$48 billion to defend a position that was then rescued, two years later, by a structural change it neither paid for nor engineered.

That is the uncomfortable investing lesson buried in this episode. Defensive capital deployment by an incumbent is extremely hard to evaluate at the time, because success looks identical to having overpaid — you keep what you already had. The only genuine test is the counterfactual, and the counterfactual is unobservable. What an investor can do is watch what management does next with large, discretionary, hard-to-reverse capital. Which is precisely the question hanging over the AI data centre programme.

And that question landed on a new set of desks in September 2024.

↑ Back to Top

V. Current Management: Continuity Inside a State-Controlled Board

On September 30, 2024, Chunghwa Telecom changed its chairman and its president on the same day.¹⁰

In most listed companies, a simultaneous CEO-and-president transition signals crisis: a blown quarter, an accounting problem, an activist. Here it signalled nothing of the kind. It was a board-driven succession inside a state-controlled governance structure, and the people who came in had spent essentially their entire careers inside the building.

Chih-Cheng Chien became chairman. His résumé is a tour of Chunghwa's own org chart: president of the International Business Group, president of the company's Telecommunications Training Institute, president of the Mobile Business Group, and president of the Network Technology Group. He holds a doctorate in engineering technology from Taiwan's institute of industrial technology.¹⁰ He succeeded Shui-Yi Kuo.

Rong-Shy Lin became president, succeeding Chau-Young Lin. His path is similarly internal and similarly technical: vice president of the Enterprise Business Group, vice president of the information department, vice president and then president of Telecommunication Laboratories, president of the Data Communications Business Group, chief technology officer, and chief information security officer, with a doctorate in computer science from National Chiao Tung University.¹⁰

Two engineers, two lifers. The chief financial officer role sits with Wen-Hsin (Audrey) Hsu, who fields the capital-allocation questions on the earnings calls.¹¹

The self-description they launched with was "New Team, New Vision," and Chien's opening framework was three words: simplification, pragmatism, intelligence — cutting the product catalogue and simplifying processes, applying an 80/20 rule to focus effort, and pushing intelligent systems through the operation. He summarised it as "Do the right things Right."¹⁰

Take that seriously for a moment, because it is more revealing than it first appears. The first two pillars are not growth strategies. They are the language of an operator who has looked at a formerly-state-owned organisation with 442 service centres and a sprawling product catalogue and concluded that a meaningful share of the earnings improvement available to him comes from doing less, not more. That is a credible read on where the value is in this kind of business — and it is a considerably more grounded starting position than a growth story.

Then consider the incentive structure, which is where a normal governance analysis breaks down.

Chunghwa's executives hold negligible direct equity. This is not an oversight; it is a structural feature of ministry-appointed leadership. Meaningful equity incentives at the group level do not exist for the parent's senior team, though listed and semi-listed subsidiaries including 是方電訊 Chief Telecom and other group entities have run their own employee stock option plans.² The chairman does not get rich if the shares double. He does not get poor if they halve.

Strip out the equity, and what is left to align management with shareholders? Two things, and only two: hitting the guidance, and paying the dividend. Both are public, both are annual, and both are unambiguous. That is a thinner alignment mechanism than an equity-heavy Western telecom board — but it is not a worthless one, because it is unusually falsifiable. There is nowhere to hide.

So test it.

Guidance discipline. The company set 2025 targets, then beat them. Chairman Chien told the 2026 annual general meeting that Chunghwa had exceeded its financial guidance in 2025, with revenue and EPS both reaching records.⁵ That is verifiable rather than rhetorical: revenue NT$236.11 billion and EPS NT$4.99, up 4.0%.¹ Then the first quarter of 2026 beat again — revenue NT$59.99 billion, up 7.5%, the highest first quarter since 2012, with EPS of NT$1.30, a decade high.³ Chien's framing on that call was specific rather than atmospheric: "All financial metrics exceeded quarterly forecasts."³ Two consecutive beats against public targets, from a team barely eighteen months into the job. On the narrow test available, this is a pass.

Dividend consistency. The 2026 annual general meeting approved a cash dividend of NT$5.2 per common share against 2025 earnings — a payout ratio of 104.2%.⁵ That is up from NT$5.00 the prior year. Chunghwa is therefore paying out slightly more than it earns while simultaneously raising capital expenditure 14.6%.⁴ That combination is only possible because depreciation runs far ahead of maintenance capital needs on a mature network, and because the balance sheet carries very little debt. It is not reckless. But it is worth naming precisely: a payout above 100% of earnings, during a capex ramp, leaves no retained-earnings buffer. If the AI programme requires materially more capital than currently budgeted, something has to give — the payout, the balance sheet, or the programme.

Willingness to explain a miss. This is the sharpest available test, and it produced the most interesting exchange of the last two years. In the fourth quarter of 2025, Enterprise Business Group revenue fell 7.9% year over year to NT$22.02 billion, even as Consumer grew 5.9%.¹ Group ICT revenue declined 6% in the quarter.¹¹ For a company selling an ICT-led growth story, a sharply negative quarter in exactly that segment is the moment to watch.

President Lin's explanation was concrete rather than atmospheric: major projects had been recognised in earlier quarters, creating a high comparison base, while recurring ICT revenue — the part that repeats — grew 15% year over year.¹¹ That is the right answer if it is true, and it is a checkable claim. The distinction between lumpy project recognition and recurring revenue is exactly the disclosure an investor needs in a systems-integration business, and management volunteered the recurring number rather than hiding behind the headline.

The follow-through supports it. By the first quarter of 2026, group ICT revenue was up 25% year over year with recurring ICT up 11%.¹² Enterprise Business Group revenue rose 8.5% to NT$18.81 billion.³ The Q4 wobble was, on the evidence so far, a base effect rather than a demand problem.

Where they were less forthcoming. On the same fourth-quarter call, J.P. Morgan's analyst pushed on two things: why non-mobile capital expenditure was rising nearly 24%, and why operating costs were guided up 3.5–4% against roughly 2% revenue growth. CFO Hsu answered with categories — fixed-line maintenance, satellite, cables, internet data centres on the capex side; talent acquisition especially in AI roles, electricity cost uncertainty, and depreciation from prior-year spending on the cost side.¹¹ Reasonable answers. But the analyst came back asking for a more granular breakdown of the non-mobile capex components in future disclosure — which is the polite analyst way of saying the answer was not specific enough.¹¹

That request has not, so far, been met. On the first-quarter 2026 call, when analysts probed the durability of ICT growth and the underlying capex trend, management pointed to concrete projects — an internet data centre construction pipeline, an APCN cable expansion to 18 terabits per second — but declined to give forward capex percentages, multi-year guidance, or any quantified revenue contribution expected from AI initiatives.¹²

That is the honest picture of this management team as of August 2026: credible on delivery, disciplined on guidance, generous on dividends, and notably reticent on the forward economics of the single largest new thing they are doing.

There is one more standing item that has nothing to do with performance. Because the chairman's seat is effectively a government appointment, leadership continuity here is a function of Taiwan's political cycle as much as of Chien's track record. The 2024 transition was orderly. There is no structural guarantee the next one will be timed by the business rather than by an election. Treat management-change risk at Chunghwa as a permanent, low-probability, always-on condition rather than an event risk to be modelled once.

Now to the thing they are actually spending the money on.

↑ Back to Top

VI. The New Growth Story: "AI Everywhere" and the AI Data Center Bet

The most useful way to understand what Chunghwa Telecom is attempting is to look not at the strategy deck but at the floor.

A conventional telecom data hall is built for racks that draw two to five kilowatts each, sitting on a floor rated to about 500 kilograms per square metre. That is the standard for servers doing what servers have done for twenty years — running databases, hosting websites, terminating network traffic.

An AI rack is a different physical object. A single rack packed with accelerators can exceed 1,250 kilograms and draw 100 to 200 kilowatts — with next-generation deployments pushing toward 140 kilowatts per rack as a working target. To host that, Chunghwa has been rebuilding its facilities to floor loadings of 2,000 to 2,500 kilograms per square metre and converting from air cooling to liquid cooling, customised per customer.¹³¹⁴

Stated plainly: a rack that used to weigh as much as a motorcycle and draw as much power as a few households now weighs as much as a small car and draws as much power as a city block. The building has to be structurally reinforced, the power feed rebuilt, and the heat removed with circulating fluid rather than moving air. This is not a software upgrade. It is closer to converting a warehouse into a foundry.

That physical reality is the entire investment case, and the entire risk. It means AI data centre capacity cannot be conjured quickly — which protects incumbents with land, power connections, and cooling water. It also means the capital is sunk, specific, and non-recoverable if demand does not materialise. You cannot rent a liquid-cooled 200kW hall to a customer who wants to run a website.

The strategy wrapping this is branded "AI Everywhere" under a corporate vision Chunghwa calls "Igniting the Future with AI," structured around three pillars — digital resilience, smart empowerment, and a sustainable future — and three execution guidelines: advancing technology (cybersecurity, pre-6G), expanding business (internet data centres, cloud, sovereign AI), and extending footprint (subsidiary investments, asset optimisation).⁴ The architectural choice worth noting is distributed AI data centres rather than a few hyperscale campuses — using the carrier's existing national network footprint to place compute in multiple locations for resilience and scalability.¹³

On the product side, President Lin has described a proprietary "CHT AI factory" platform bundling full-stack solutions, compute capacity, and a portfolio of AI modules, models, and agents, with the stated ambition to "convert our AI capabilities into our service offering" for enterprise customers integrating AI into operations and compliance.¹²¹¹ Adjacent to this sits genuine technical credibility: Chunghwa deployed what has been described as the world's first IOWN all-photonics network in 2024 and sits on the IOWN Global Forum board, having invested in a US$500 million IOWN-related AI fund alongside NTT, SK Telecom and others.¹³

Now the evidence.

What is proven. The revenue is real and it is accelerating. Group ICT revenue grew 25% year over year in the first quarter of 2026, with recurring ICT up 11% — and management identified the specific drivers as cybersecurity, internet data centres, and international public cloud.¹² The International Business Group, the runt of the segment structure, grew 10.7% to NT$2.70 billion on the back of an 89% surge in United States revenue driven by a single large-scale AI supply chain project.³ President Lin's characterisation on that call was direct: "We continue to monetize our AI infrastructure, delivering solid revenue growth, particularly driven by AI data center."¹² Meanwhile President Lin has told investors that pre-6G-adjacent opportunities spanning AIoT, satellite, and big data should together clear NT$10 billion of revenue in 2026.¹

Those are not vapour numbers. They are reported segment results with named drivers.

What is not proven. Chunghwa has not disclosed AI data centre utilisation rates. It has not disclosed contracted versus available capacity, megawatts deployed or under construction, customer concentration, contract duration, or the return profile on AI-specific capital. When analysts asked on the first-quarter call about the durability of ICT growth and AI's impact on the IT services business, the response stayed at the level of "digital transformation demand" and agentic AI's ability to upgrade services and provide value — without quantifying an expected revenue contribution from AI initiatives.¹² The most granular forward-looking disclosure offered was a cable capacity figure and the existence of an IDC construction pipeline.¹²

That is a real gap, and it should be stated without euphemism: as of August 2026, the AI data centre programme is capex-ahead-of-disclosed-economics. The company is reporting growth in a bucket labelled ICT and attributing part of it to AI infrastructure. It is not yet showing investors the unit economics of the AI infrastructure itself.

A second data point that complicates the picture. Chunghwa guided to approximately NT$32.4 billion of capital expenditure for 2025, allocated across 5G, IDC construction, FTTx, submarine cables, network transformation, service platforms, cloud, IoT and ESG.² Actual 2025 capex came in at NT$27.84 billion.¹ That is roughly a 14% underspend against plan.

Two readings are available, and both are worth holding. The benign one: construction, permitting, and equipment lead times slipped, and the money moves into 2026 — which is consistent with 2026 capex being guided up 14.6% to NT$31.91 billion.⁴ The less benign one: a company that talks about a 40%-growth AI infrastructure market and then underspends its own capital plan by NT$4.5 billion is either supply-constrained or demand-constrained, and it has not told investors which. For a business whose bull case rests on capturing AI infrastructure demand, an unexplained capex underspend is not a footnote. It is a question that deserves a direct answer on a call.

On the sustainability and cost side, there is a genuine constraint that most AI infrastructure discussions in Taiwan skate past: power. Roughly 40% of Chunghwa's data centre electricity came from renewable sources as of 2025, against a target of 100% by 2030 for internet data centres and 2040 for the company as a whole.¹³² Meanwhile, the CFO has flagged electricity cost uncertainty as a driver of rising operating costs.¹¹ On an island with constrained generating capacity and a contested energy policy, power price and power availability are not ESG talking points for this business. They are input costs and capacity limits.

The bolt-ons, briefly, because they are small and should be treated as such. Chief Telecom, founded in January 1991 and majority-acquired by Chunghwa in September 2006, listed on the Taipei Exchange on June 5, 2018 under 6561 and runs carrier-neutral data centres including a newer AI-oriented facility.⁶ It is a useful asset in the AI story but a modest one relative to group scale. In March 2025 Chunghwa acquired Pingnan Cable TV, a small regional cable operator — a consolidating move in a declining category, economically immaterial. And in a genuinely trivial example of the company's investment activity, a July 9, 2026 disclosure recorded a US$78,000 long-term strategic commitment to a fund partnership, described by the company itself as a very small portion of total assets.²⁰

There is one legacy diversification worth flagging as a second-layer item. Chunghwa invested in Next Commercial Bank in 2020; the bank launched services on March 29, 2022 and, as of the most recent annual filing, had yet to generate profits.² Small in the scheme of a NT$236 billion revenue company, but a live reminder that this management culture is comfortable making adjacent bets that take a long time to pay — which is exactly the disposition that makes a large AI capex programme worth watching rather than assuming.

So what does an investor do with all this? The AI layer is the only part of Chunghwa capable of moving the growth rate. Its early revenue is real, its drivers are named, and its physical build is genuinely hard to replicate quickly. But the case currently rests on segment-level growth rates and management framing, not on disclosed returns. Until Chunghwa publishes something resembling AIDC utilisation, contracted capacity, or incremental return on invested capital, an investor is underwriting a capability, not an economic result. That distinction is the single most important one in this story.

And a good part of why the state is comfortable funding it has nothing to do with returns at all.

↑ Back to Top

VII. Strategic Infrastructure and the Geopolitical Overlay

On January 3, 2025, a cargo ship called the Shunxin-39 — Chinese-owned, crewed entirely by Chinese nationals, registered in Cameroon and Tanzania — was operating in waters off Yehliu, a cape on the northern coast of New Taipei City. When it left, four cores of the Trans-Pacific Express cable were damaged.¹⁷

Seven weeks later, on February 25, 2025, Taiwan's coast guard detained a Togo-flagged vessel, the Hong Tai, crewed by eight Chinese nationals, after it had been lingering near a cable route off the southwest coast since the previous Saturday. The cable it is suspected of damaging — TPKM-3 — connects Taiwan to the Penghu Islands.¹⁷

These were not the first. In February 2023, two cables serving the Matsu Islands were severed, leaving residents without internet service for weeks.¹⁷ Two further domestic cables linking Taiwan and Lienchiang County were damaged in January 2025.¹⁶ Taiwan's coast guard has blacklisted 96 suspicious vessels, predominantly Chinese-owned but sailing under flags of convenience from Mongolia, Cameroon, and Sierra Leone.¹⁷

Here is the context that makes those incidents matter economically rather than merely dramatically. Taiwan's connection to the rest of the world runs through roughly 24 submarine cables — 14 international and 10 domestic.¹⁷ That is thin redundancy for a technology-exporting island economy. Repairs take six to eight weeks or longer, cost millions of dollars, and require specialised cable ships. More than 100 cable incidents occur globally each year for entirely mundane reasons — anchors, fishing gear, dredgers — and Chunghwa's own filings acknowledge that sand-pumping dredgers routinely break its cables, suspending fixed-line, television, broadband and mobile services.²¹⁷ The pattern around Taiwan is distinguishable from that background rate mainly by who is nearby when it happens.

Chunghwa sits at the centre of this. It is the Taiwanese carrier that invests in, lands, and operates the systems the island depends on — from legacy trans-Pacific routes to the newest builds — and it has been methodically deepening that position.

The design principle it has adopted is dual landing. Rather than concentrating international cable arrivals at a single point, Chunghwa lands systems at both Tamsui District in New Taipei City in the north and Fangshan Township in Pingtung County in the south, with additional landing infrastructure at Toucheng in Yilan.¹⁵ Geographic separation of hundreds of kilometres means a single incident — deliberate, accidental, or seismic — cannot sever the country's connectivity at one stroke. It is the network equivalent of not keeping the spare key under the mat.

The build programme moved fast in 2025 and 2026. The Southeast Asia–Japan 2 system, SJC2, was activated in July 2025 after a construction delay of nearly three and a half years against the original plan — a delay Chunghwa attributed to permitting complexity.¹⁵² The system runs roughly 10,500 kilometres connecting Taiwan, Singapore, Thailand, Vietnam, Hong Kong, China, South Korea and Japan across eleven stations, delivers up to 126 terabits per second, and cost Chunghwa NT$1.36 billion, about US$46.2 million.¹⁵¹⁶ The Apricot system — some 12,000 kilometres linking Japan, Taiwan, Guam, the Philippines, Indonesia and Singapore — is scheduled for completion in 2027.¹⁶¹⁴ In the same week SJC2 went live, Chunghwa signed into the Asia United Gateway East consortium, committing NT$2.4 billion for a system due in 2029.¹⁵¹⁹ Domestically, a fourth Taiwan–Penghu–Kinmen–Matsu cable is targeted for 2026.¹⁴ Total undersea cable spending was lifted to roughly NT$2 billion in 2025.¹⁶

Chairman Chien's framing of the SJC2 activation was explicitly national rather than commercial: Chunghwa "will continue to take concrete actions to enhance the digital resilience of Taiwan's communications network."¹⁵

That sentence is the tell. A normal telecom operator justifies a cable investment with a bandwidth demand forecast and an internal rate of return. Chunghwa justifies it with resilience. Both things can be true — international bandwidth sales, IDC connectivity, and terrestrial link revenue all depend on cable capacity, and the company's own filings note that disruption to new cable projects would endanger exactly those revenue lines.² But the ordering of the justification tells you which shareholder is being addressed.

Which brings us to the concrete example of political logic overriding commercial logic.

In November 2023, the NCC publicly pressed Chunghwa to reconsider a network arrangement with China Unicom Hong Kong. The arrangement was an international roaming agreement that let users access Taiwanese telecom services using Hong Kong phone numbers on prepaid cards. The regulator's objections were specific: the cards created a fraud-prevention loophole because online purchasers bypassed Taiwan's two-forms-of-ID requirement, and the arrangement was non-reciprocal — it worked in Taiwan but not in Hong Kong. The NCC had already suspended 250,000 such numbers in Taiwan over fraud involvement. Vice Chairman Wong Po-tsung's public warning was that if numbers were involved in fraud, "telecom services would be suspended," and he urged the company to "exercise self-discipline."¹⁸

Chunghwa pushed back on the compliance allegation, noting it had blocked 25,500 SIM cards from roaming since August, and then said it planned to terminate the roaming contract with China Unicom Hong Kong entirely.¹⁸

Read the sequence. A profitable, legal, contractual commercial relationship was publicly questioned by the regulator on security and fraud grounds, and the company terminated it. No legislation was passed. No licence condition was cited. The regulator applied pressure and the company folded — in a direction that happened to align with national security policy.

That is not a scandal. It may well have been the right call. But it is a documented instance of how commercial decisions get made at this company when Taipei has a view, and an investor should generalise from it rather than treating it as a one-off.

The net position is genuinely two-sided, and both sides need to be held simultaneously. Being the national digital-resilience champion is a durable business advantage: it means state support for capex, priority in national infrastructure programmes, a near-unassailable position in government and defence-adjacent contracts, and a brand that means something during a crisis. It also means being a named target. In any serious cross-strait escalation, Chunghwa's cables, landing stations and core network are not incidental infrastructure — they are among the first things an adversary would want to degrade. And the company's own disclosure notes it does not carry insurance covering damage from natural disasters, with losses on property, plant and equipment from earthquakes and typhoons running at roughly NT$17.44 million in 2024.² That figure is small. The disclosure of the practice is the point.

You cannot own this business and hedge that risk away. It is embedded in the asset.

Time to look at what the whole machine produced.

↑ Back to Top

VIII. Financial Snapshot as Narrative

The temptation with a company like this is to read out the numbers. Resist it, because the numbers only mean something in a specific shape.

The shape is this: a large, slow, high-certainty base growing at roughly two to three percent, with a smaller, fast, less predictable layer growing at double digits on top. The blended result — and the blended result is the only thing that matters to an equity holder — is a company that grew revenue 2.7% in 2025 and then 7.5% in the first quarter of 2026.¹³

That acceleration is the headline, and it deserves scrutiny rather than applause.

For the full year 2025, revenue reached NT$236.11 billion, operating income NT$48.55 billion (up 3.6%), EBITDA NT$88.77 billion (up 2.6%), and net income NT$38.69 billion with EPS of NT$4.99 — the strongest earnings in eight years.¹ Fourth-quarter revenue was NT$65.65 billion, up only 0.5%, with quarterly EPS of NT$1.20.¹ Note the internal composition of that quarter: Consumer up 5.9%, Enterprise down 7.9%, operating income down 2.2%.¹ The utility carried the quarter while the growth engine went backwards on a base effect.

Then the reversal. First-quarter 2026 revenue of NT$59.99 billion, up 7.5% — the highest first quarter since 2012. Operating income NT$13.10 billion, up 4.6%. EBITDA NT$23.30 billion, up 3.4%. Net income attributable to shareholders NT$10.11 billion, up 3.2%, with EPS of NT$1.30, the best first quarter in a decade.³ All three segments grew: Consumer 6.2%, Enterprise 8.5%, International 10.7%.³

Here is the detail that a numbers dump would bury, and it is the most analytically important line in the quarter. Revenue grew 7.5%. Operating income grew 4.6%. EBITDA grew 3.4%. Operating margin fell from 22.44% to 21.75%, and EBITDA margin fell from 40.37% to 38.85%.³

Margins compressed on record revenue. That is not a failure — it is arithmetic, and it is exactly what you would expect. The fast-growing revenue is ICT: systems integration, equipment resale, project delivery, data centre services. That revenue carries a materially lower gross margin than selling a mobile plan over a network you built a decade ago. As the mix shifts toward ICT, the blended margin falls even when every individual business is performing well.

The implication is one investors in this name should internalise now rather than discover later: Chunghwa's revenue growth and its earnings growth are going to diverge, structurally, for as long as the AI and ICT layer is the growth driver. A shift from a 40% EBITDA-margin business toward a lower-margin one is dilutive to margin by construction. That does not make it value-destructive — it can create considerable value if the incremental return on capital is adequate. But anyone anchoring on "record revenue" as evidence of a successful transition is reading the wrong line.

Management's own 2026 guidance concedes the point with unusual candour. Revenue is guided to NT$241.99–243.68 billion, up 2.5–3.2%. But net income is guided to NT$37.39–38.94 billion — a range spanning negative 3.4% to positive 0.6%. EPS is guided at NT$4.82–5.02, against NT$4.99 delivered in 2025. EBITDA is guided up 1.7–3.4% with a margin of 37.3–37.7%, down from 2025 levels.⁴

In plain terms: management is guiding to revenue growth with flat-to-lower earnings. That is a mix-shift and cost story — higher depreciation from the capex ramp, hiring in AI-related roles, and electricity cost uncertainty, per the CFO's own explanation.¹¹ It is also, notably, conservative guidance from a team that has now beaten twice. Whether that is prudence or sandbagging will be visible over the next several quarters.

The trajectory through mid-2026 has run well ahead of the guided pace. Cumulative January–June revenue reached NT$121.35 billion, up 7.83% year over year, with June alone delivering NT$21.34 billion of revenue and NT$3.55 billion of net income.²⁰ Half-year revenue growth of nearly 8% against full-year guidance of 2.5–3.2% implies either a substantial upgrade ahead or a sharply decelerating second half — and the second-quarter call on August 5, 2026 is where that gets addressed.²¹

Which leaves the dividend, and the reason it functions as the load-bearing element of the equity story here.

Cash dividends have moved from NT$4.76 to NT$5.00 to the NT$5.2 per share approved at the 2026 annual general meeting — a payout ratio of 104.2% against 2025 earnings.⁵ Sustaining and raising a payout above 100% of net income, while simultaneously lifting capital expenditure 14.6% and carrying a conservatively geared balance sheet, is the clearest behavioural evidence available that management is not planning to fund an AI land grab by cutting shareholder returns.

It is also a constraint they have chosen to impose on themselves. A company paying out everything it earns cannot quietly triple its capital programme. If the AI opportunity turns out to be larger than budgeted, Chunghwa will have to either issue debt, cut the payout, or pass — and each of those choices would be a highly visible signal. For a business with weak equity-based alignment, that visibility is worth something.

So what does the market actually own here? A low-growth, high-certainty cash machine, deliberately capped in its capital ambitions by an above-100% payout, running an option on AI infrastructure with capital it can spare rather than capital it must find. Whether that option is worth much is the argument.

↑ Back to Top

IX. Powers, Position, and the Bull/Bear Case

Time to war-game it properly, using two frameworks and one adversary.

Porter's five forces, applied honestly.

Barriers to entry: very high, and structurally so. You cannot enter Taiwanese telecom. Spectrum is finite and was allocated at auction in 2020 at a clearing price of NT$142 billion across five bidders.⁸ Cable landing rights require permits that, in Chunghwa's own documented experience with SJC2, can delay a project by three and a half years.² Last-mile fixed plant into millions of homes cannot be economically duplicated. Universal-service and information-security obligations impose costs a new entrant would have to absorb from day one.² And the market just consolidated downward from five players to three — the opposite of an environment attracting entrants.² This force is about as favourable as it gets.

Rivalry: moderate, and improved — but not permanently. Three rational players with differentiated positioning: Chunghwa at the premium quality end, Far EasTone running a quality-over-quantity ARPU strategy, Taiwan Mobile at higher volume and lower ARPU.⁹ Industry ARPU is rising. That is textbook post-consolidation discipline. The caveat is that discipline is a choice, not a structure. Three players is enough for a price war if any one of them decides share matters more than margin.

Supplier power: low. Network equipment, handsets, and increasingly AI hardware are competitively supplied. The genuine exception is electricity, where Taiwan's constrained generation and contested energy policy give the input real leverage — which is why the CFO named electricity cost uncertainty as a driver of 2026 cost guidance.¹¹

Buyer power: split. Consumers have almost none — three carriers, switching costs that are low but not zero, and a market where everyone's prices are drifting up together. Enterprise and ICT buyers have considerably more: they can multi-source, run competitive tenders, and in the AI and cloud domain can go straight to global hyperscalers. This is precisely why the fast-growing ICT revenue carries thinner margins than the mobile business.

Substitutes: negligible for the last mile, real at the layer above. Nothing replaces the fibre and the radio access network. But the services delivered over them are substitutable — messaging killed SMS revenue years ago, and international public cloud can substitute for domestically-hosted IDC unless there is a sovereignty or latency reason not to. Chunghwa's answer to that is sovereign and hybrid cloud, which is a genuine differentiator in a jurisdiction with acute data-sovereignty concerns.

The 7 Powers lens: which power is this, really?

The temptation is to call this a network-effects business. It isn't. Mobile customers derive no benefit from other customers being on the same network; interconnection regulation ensures a call to a rival's subscriber works identically.

It is also not a switching-cost story of any strength. Number portability is mandated.² Contracts run twelve to twenty-four months. Churn between three carriers is a live and continuing phenomenon.

What Chunghwa actually has is two of Helmer's powers, and they are the right two for an infrastructure business.

Cornered resource is the strongest. Ninety megahertz of contiguous 3.5GHz spectrum, plus holdings across 900MHz, 1800MHz, 2100MHz, 2600MHz and 28GHz.² Cable landing rights and dual landing stations at Tamsui and Fangshan.¹⁵ Rights of way and duct for a national fixed-line plant assembled over a century of state monopoly. None of these can be bought at any price by a competitor, because they are not for sale.

Scale economies run second. Fixed network costs are amortised across the largest subscriber base in the market, which is why Chunghwa can carry 442 service centres and still post a 37–38% EBITDA margin.⁴²

There is a third, weaker power worth naming: something adjacent to branding, though it is closer to institutional trust. In a crisis — an earthquake, a typhoon, a cable cut, a cross-strait incident — Chunghwa is the network the state, the military, and older households assume will still be working. That is not a premium you can charge for on a rate card. It is a customer-acquisition and retention advantage that shows up in low churn and in government contract flow.

What is conspicuously absent: process power, counter-positioning, and cornered scarcity in anything AI-related. In the AI data centre business specifically, Chunghwa's powers are considerably weaker. It has land, power connections, network adjacency and sovereignty positioning — real advantages inside Taiwan. It has no proprietary technology, no scale advantage against global hyperscalers, and no evident cost advantage. The moat that protects the telecom business does not extend to the growth business. That asymmetry is the single most under-discussed feature of this story.

The bull case, stated at its strongest.

A quasi-utility with genuinely defensible economics generates a large, predictable cash flow. That cash flow funds a dividend that has risen for consecutive years to NT$5.2 per share.⁵ It also funds — from surplus, not from leverage — a call option on AI infrastructure in a jurisdiction that is simultaneously the world's most important semiconductor manufacturing base and one of the most acutely security-conscious digital economies. Chunghwa is the natural sovereign-cloud and sovereign-AI provider for a country that has every reason to want its compute domiciled at home and defended. The ICT growth is already visible in reported segment results, and the physical build — liquid cooling, reinforced floors, hundreds of kilowatts per rack — is genuinely hard to replicate at speed.¹²¹³ Meanwhile the December 2023 consolidation permanently improved industry pricing behaviour, ARPU is rising for the first time in a decade, and the 5G capital cycle is finished with mobile capex down for five straight years.¹¹ State backing lowers political and funding risk domestically. You are paid to wait, and the option is free.

The bear case, stated at its strongest.

Government control is not a background condition — it is the operating constraint. Tariffs are regulated, universal-service obligations are mandated, the regulator can designate the company as market-significant and force it to open its plant to rivals, and every non-independent director is a ministry appointee.² The November 2023 Hong Kong roaming episode demonstrated that a commercially rational arrangement can be abandoned under regulatory pressure.¹⁸ Capital allocation therefore has a permanent political input.

On growth: the core is structurally low-single-digit and cannot be otherwise. The ICT layer that provides the growth is lower-margin, which is why management's own 2026 guidance pairs revenue growth with flat-to-declining earnings.⁴ The AI programme has no disclosed unit economics — no utilisation, no contracted capacity, no incremental returns — and the company underspent its own 2025 capital plan by roughly NT$4.5 billion without a clear public explanation.²¹ Global AI data centre capacity is being built at extraordinary scale by far better-capitalised players; a Taiwanese incumbent telecom is not obviously advantaged outside its own borders, and its international segment is small enough that one large project moved it 89% in a quarter.³

On the shareholder base: foreign direct ownership is capped at 49%, total foreign ownership at 60%, and breach can theoretically cost the company its network establishment approval — a structural deterrent to concentrated foreign accumulation.² And the entire asset sits inside the most-watched geopolitical flashpoint in Asia, uninsured against natural disaster damage and physically targeted in the grey zone already.²¹⁷

The activist stress test.

A state-controlled telecom is not a conventional activist target — you cannot win a proxy fight against a ministry that appoints the board. But run the questions anyway, because they are the right questions regardless of who can force an answer.

On capital allocation: the dividend record demonstrates discipline in returning cash. It does not demonstrate discipline in deploying it. Those are different skills, and the company is currently being asked to demonstrate the second one at a scale it has not attempted since the 5G auction. The most pointed version of the question: if the AI data centre business is generating attractive returns, why not disclose them and let the market re-rate the stock? If it is too early to disclose returns, is it too early to be raising capex 14.6%?

On disclosure: a J.P. Morgan analyst asked directly, on the fourth-quarter 2025 call, for a more granular breakdown of the non-mobile capex components in future disclosure.¹¹ Two quarters later, that breakdown had not appeared; the first-quarter 2026 answers offered project examples but no forward percentages or multi-year guidance.¹² That is a legitimate, specific, unmet ask from a covering analyst — the kind of thing an activist would escalate.

On portfolio complexity: a listed subsidiary in Chief Telecom, a small cable TV acquisition, a bank investment that had not turned profitable as of the latest annual filing, retail distribution through SENAO with related-party arrangements, and a fund commitment small enough to be a rounding error.²⁶²⁰ None of these is individually alarming. Collectively they describe an organisation with a mild disposition toward adjacency, which is worth monitoring precisely because it is about to deploy large discretionary capital.

On accountability: with negligible executive equity ownership, there is no financial mechanism by which a poor AI capital allocation decision costs management personally. Guidance and dividend delivery are the only accountability levers, and both are annual and backward-looking.

On the counter-argument: the strongest defence is that management has actually delivered — beating 2025 guidance and the first quarter of 2026, explaining the Q4 ICT decline with a specific, checkable recurring-revenue distinction rather than boilerplate, and raising the dividend through a capex ramp.⁵³¹¹ That is a better behavioural record than most telecom management teams can show. It earns them time. It does not settle the AI question.

Why this company wins from here, and what breaks the case.

It wins if the sovereign-AI thesis is real — if Taiwanese enterprises, government, and defence-adjacent buyers pay a premium for compute that is domestically hosted, network-adjacent, low-latency, and operated by the carrier the state trusts. In that world the AI data centre business is not competing with global hyperscalers at all; it is competing for a protected domestic niche where Chunghwa's cornered resources still apply. The ICT growth rates, and the specific attribution to cybersecurity, IDC and sovereign cloud, are early evidence consistent with this.¹²

It breaks if the AI layer turns out to be low-margin infrastructure resale — buying accelerators, hosting them, and passing through most of the economics — while the depreciation from the buildout lands squarely on earnings. In that world revenue grows, margins compress, EPS stagnates, the payout ratio above 100% becomes untenable, and the equity is left as a lower-quality version of the utility it started as. The 2026 guidance of revenue growth with flat-to-lower net income is, uncomfortably, exactly what the early stage of that scenario would look like.⁴

Distinguishing between the two requires data the company has not yet published. That is the state of the argument as of August 2026.

↑ Back to Top

X. Risk Radar

Five risks matter here, and each has a specific mechanism rather than a generic label.

Geopolitical and cable-sabotage risk. The mechanism is concentration. Taiwan's international connectivity depends on roughly 14 international submarine cables, with Chunghwa the principal Taiwanese participant across the systems that matter.¹⁷ The threat is documented and recent rather than hypothetical: the Shunxin-39 incident damaging TPE cores in January 2025, the Hong Tai detention in February 2025 over TPKM-3, the Matsu severances in February 2023, and a coast guard blacklist of 96 suspicious vessels.¹⁷ Repairs run six to eight weeks. The company's mitigation — dual landing at Tamsui and Fangshan, plus new capacity through SJC2, Apricot, E2A and a fourth Taiwan–Penghu–Kinmen–Matsu domestic cable — is genuine and expensive, and it addresses redundancy rather than deterrence.¹⁵¹⁶¹⁴¹⁹ In a serious escalation scenario, no amount of route diversity protects a company whose network is the target.

Regulatory and political risk. The mechanism runs through two channels. First, tariff and obligation: the NCC sets the framework for pricing, universal service, interconnection, and information security, and can impose asymmetric remedies on an operator designated as significant in a specific market — a designation Chunghwa concedes may apply to it.² Second, ownership: the ministry appoints every non-independent director and thereby influences senior management composition, the annual budget, and dividend timing.² The 2023 Hong Kong roaming episode is the concrete demonstration of how regulatory pressure translates into commercial decisions without any formal order being issued.¹⁸ Add the leadership-continuity dimension — the chairman's seat is effectively a political appointment — and this is a standing condition, not an event.

AI capex execution risk. The mechanism is asset specificity. Liquid-cooled halls with floors rated to 2,000–2,500 kg/m² and power feeds for 100–200kW racks cannot be repurposed for general-purpose hosting.¹³¹⁴ Capital committed is capital sunk. With 2026 capex guided to NT$31.91 billion and rising non-mobile spending across IDC, satellite and cables, the exposure grows each year.⁴¹¹ The specific concern is not that the spending is large — it is that no utilisation, contracted-capacity, or return disclosure exists to test whether it is productive, and the 2025 underspend against a NT$32.4 billion plan suggests the pipeline may be less linear than the strategy language implies.²¹ Depreciation from this programme is already cited as a driver of the flat-to-lower 2026 earnings guidance.¹¹⁴

Competitive-intensity risk. The mechanism is behavioural, not structural. The post-2023 three-player market is delivering rising ARPU across all participants.⁹ But Far EasTone has demonstrated it can reset its subscriber base aggressively, and Taiwan Mobile has demonstrated it will trade ARPU for volume.⁹ Both extracted merger synergies unavailable to Chunghwa, giving them room to price aggressively while still growing profits. If either concludes that share matters more than margin — most plausibly in the value-plan segment where Chunghwa's network-quality premium is least persuasive — the industry's pricing discipline could unwind faster than it took to establish.

Currency, ADR and ownership-structure risk. The mechanism is mechanical rather than fundamental. Substantially all revenue is in New Taiwan dollars while the ADR is priced in US dollars, so a non-Taiwan holder's return embeds a currency position they may not have chosen. Layered on that are the foreign ownership caps of 49% direct and 60% total, breach of which could in principle cost the company its network establishment approval even absent any fault of its own — a structural deterrent to concentrated foreign accumulation and, in practice, a constraint on which investor bases can build meaningful positions.² Sell-side coverage of the ADR is thin, which tends to widen the gap between the Taipei and New York listings' price discovery.

One second-layer item worth logging rather than dramatising: the company does not carry insurance against earthquake, typhoon or other natural disaster damage, or the resulting business interruption, and recorded roughly NT$17.44 million of such losses in 2024.² The historical loss experience is trivially small relative to a NT$236 billion revenue base. The disclosure matters because it establishes that a self-insurance posture on physical catastrophe is deliberate policy — on an island subject to major seismic events.

↑ Back to Top

XI. Playbook: What This Company Teaches

Four transferable lessons, and none of them are specific to telecom.

Privatisation is not a transition — it can be a permanent structure. The instinct when reading "privatised in 2005" is to treat state ownership as a legacy that decays over time. Twenty-one years later, the ministry holds roughly 35% and appoints every non-independent director.² That is not a company on a path to independence; it is a company in a stable equilibrium where the state has found the minimum ownership required to retain control. When you encounter a partially-privatised state asset anywhere in the world, the right default assumption is that the current ownership level is the destination, not a waypoint — and every capital-allocation, dividend, and management-succession decision should be modelled accordingly.

A record price in a strategic auction is not automatically a bad deal — but the burden of proof is on strategic necessity, not on growth ambition. Chunghwa's NT$48.37 billion in the 2020 spectrum auction was the largest cheque any bidder wrote, at a clearing price that no participant had planned for.⁸ Judged as a growth investment it was a failure — it bought no share. Judged as insurance on an existing franchise, the logic is defensible: lose spectrum depth, lose network leadership, lose the ARPU premium, lose the dividend. The general rule that falls out: distinguish defensive capital deployment from expansionary capital deployment before grading either. They have different success criteria, and management teams have a strong incentive to describe the first as the second.

Legacy cash flow is a legitimate funding source for new-economy bets — but only with a visible discipline mechanism attached. What makes Chunghwa's AI programme underwritable rather than alarming is not the strategy. It is the payout ratio of 104.2%, which structurally prevents management from quietly redirecting a decade of retained earnings into an unproven business.⁵ Any material escalation of the AI programme would require a visible decision — debt, a dividend cut, or a pass. Investors evaluating any incumbent funding a transformation should look for the equivalent constraint. If there isn't one, the transformation is unbounded, and unbounded transformations are where incumbent capital goes to die.

Infrastructure that matters to national security is valued and scrutinised on a different axis. A submarine cable justified as "digital resilience" rather than as an internal-rate-of-return calculation is being financed partly by a non-financial buyer.¹⁵ That cuts both ways, and both sides are real. The upside is a lower effective cost of capital for strategically-aligned projects, priority in national programmes, and a business that governments will not let fail. The downside is a permanent political input into commercial decisions, demonstrated concretely when a legal and profitable roaming arrangement was terminated after public regulatory pressure.¹⁸ You do not get one without the other. Price the pair, not just the half you like.

↑ Back to Top

XII. Epilogue & What to Watch

Narrow it to the three things that actually determine how this story resolves.

KPI one: ICT revenue growth, and specifically the recurring share of it. This is the single most important number Chunghwa reports. The distinction between recurring ICT and project ICT is the difference between a durable services business and a lumpy integration contractor. Management drew that distinction itself when explaining the fourth-quarter 2025 decline, citing recurring ICT growth of 15% against a headline group ICT decline of 6% — and then delivered group ICT growth of 25% with recurring at 11% in the following quarter.¹¹¹² Watch both figures together. A widening gap between headline and recurring growth means the reported acceleration is being carried by one-off projects. A rising recurring line means the AI and cloud services are actually annuitising.

KPI two: mobile post-paid ARPU. This is the health check on the entire post-consolidation industry structure, and it is the number that funds everything else. It stood at NT$573 in the first quarter of 2026, up 3.6%.³ As long as it rises, the three-player market is behaving rationally and the utility is doing its job. If it flattens or turns down, the price discipline that made the 2023 consolidation valuable has broken, and the cash flows underwriting both the dividend and the AI programme come under pressure simultaneously.

KPI three: the dividend payout, watched against capital expenditure. At NT$5.2 per share and a 104.2% payout ratio, alongside 2026 capex of NT$31.91 billion, the two are in visible tension.⁵⁴ The direction in which that tension resolves will tell investors more about management's true priorities than any strategy document. A maintained payout with capex delivered on plan is the confirmation case. A capex step-up funded by a payout cut is a signal that the AI opportunity has been judged large enough to override the company's most public commitment — informative either way, but a materially different investment.

Near-term catalysts. The second-quarter 2026 results call on August 5, 2026 is the immediate one, and it carries a specific question: half-year revenue growth of 7.83% is running far ahead of full-year guidance of 2.5–3.2%, which means either a guidance raise or an explanation of expected second-half deceleration.²⁰⁴ Beyond that: whether the granular non-mobile capex breakdown a covering analyst requested in February 2026 ever appears;¹¹ whether any AIDC capacity, utilisation, or named-customer disclosure emerges; the Apricot cable completion in 2027 and the fourth Taiwan–Penghu–Kinmen–Matsu domestic cable in 2026;¹⁴ progress toward the NT$10 billion combined revenue target management set for AIoT, satellite and big data in 2026;¹ and chairman Chien's accumulating multi-year record, now approaching two years from the September 2024 transition.¹⁰

The closing question. Chunghwa Telecom has spent thirty years proving it can be a very good utility — corporatised in 1996, listed in 2000 and 2003, privatised in 2005, and still, in 2026, delivering record revenue, record earnings, and a rising dividend under a regulator and a controlling ministry.²¹⁵ That much is settled.

What is not settled is whether an organisation built to be reliable can be good at being venturesome. The AI data centre business rewards speed, scale, customer acquisition, and tolerance for stranded capital — a different set of muscles entirely from the ones that made this company excellent. Chunghwa has the land, the power, the network, the sovereignty positioning, and the trust of a government that badly wants domestic AI infrastructure to exist. What it has not yet shown anyone is a single number demonstrating that the capital it is putting into the ground earns an adequate return.

Until it does, the honest description is this: a well-run utility with a genuine option attached, priced by a market that has to decide how much an undisclosed option is worth. That is a legitimate thing to own. It is just not the same thing as an AI infrastructure company, and the distinction is worth holding onto every time the phrase "AI-Native Telco" appears in a press release.

Chunghwa Telecom Co., Ltd.

Table of Contents

Chunghwa Telecom 中華電信: Taiwan's State-Anchored Telecom Giant Bets on AI

I. Cold Open & Roadmap

II. From State Monopoly to Public Company

III. The Core Business: Taiwan's Telecom Oligopoly

IV. Capital Deployment I: The 5G Spectrum Auction

V. Current Management: Continuity Inside a State-Controlled Board

VI. The New Growth Story: "AI Everywhere" and the AI Data Center Bet

VII. Strategic Infrastructure and the Geopolitical Overlay

VIII. Financial Snapshot as Narrative

IX. Powers, Position, and the Bull/Bear Case

X. Risk Radar

XI. Playbook: What This Company Teaches

XII. Epilogue & What to Watch

References