Wu Yan Bin used the word "quiet" to describe a shift he thinks the market has not priced in yet. As chief financial officer and deputy chief executive officer of Summit Power International, Wu published a January outlook on LinkedIn arguing that Bangladesh sits at a specific kind of intersection: too far east to be South Asia in the way India or Pakistan are, too far west to be Southeast Asia in the way Vietnam or Indonesia are, and cheap enough on land and power to matter to a hyperscale operator deciding where to put the next few hundred megawatts.
Wu's post did not frame Bangladesh as a hedge against Singapore or Mumbai. It framed the country as something closer to a connective node between two regional data markets that rarely get analyzed together. Summit Group, he wrote, is positioned to sit "neutral" between South and Southeast Asia's cloud infrastructure buildouts, undercutting both established hubs on cost while offering power capacity neither has spare. Summit Group Chairman Muhammed Aziz Khan has made a version of the same argument in more concrete terms, telling Nikkei Asia in early January that Summit plans to enter Bangladesh's data center market within a year, with a functional facility deliverable in roughly 18 months, well under the multi-year timelines competitors need.
Khan described the underlying strategy directly: "Summit Group's next phase [of growth] focuses on integrating energy and data, leveraging our LNG and fiber-optic expertise". The framing treats data centers as a use for infrastructure Summit already operates. The company controls roughly 350 megawatts of generation capacity it says could be reassigned to data center customers, a figure that, on its own telling, clears the threshold hyperscale operators typically require. Independent research backs at least part of the premise: Mordor Intelligence's Bangladesh market analysis describes global cloud platforms preparing "Local Zone and Edge deployments" that "view Bangladesh as the logical bridge between South and Southeast Asia," language that echoes Wu's positioning without originating from Summit at all.
The comparison Wu invokes has real numbers behind it, even if Bangladesh itself is still too small a market to appear in the indices that track Singapore and Mumbai directly. Singapore's data center construction costs ran $14.53 per watt in 2025, the second-highest of any market in the world after Tokyo, according to Turner & Townsend figures cited in a Morgan Lewis analysis of the country's data center policy. Mumbai sits near the opposite end of the same index: $6.64 per watt, the second-lowest cost of the 52 global markets Turner & Townsend assessed, with electricity running about 6.71 US cents per kilowatt-hour, over 50 percent cheaper than power in Shanghai. Bangladesh is in neither ranking, and no build cost per watt has been published for Dhaka. Its industrial grid tariff runs around 10 US cents a kilowatt-hour, above Mumbai's. Summit's cost case rests on generation it already owns.
Singapore's costs are not an accident of geography. The city-state froze new data center construction in 2019 and only began reopening capacity in 2022, first through a pilot allocation of 80 megawatts, awarded in 2023 to Equinix, GDS, Microsoft and an AirTrunk and ByteDance consortium, then through a second call that opened in December 2025 for at least 200 megawatts more, with half the supply required to come from green energy (Morgan Lewis). Land is scarce, sustainability requirements are strict, and operators have started building floating data centers and looking across the border into Johor, Malaysia, to escape the constraints (Morgan Lewis). Mumbai avoids the land and power squeeze but carries its own constraint: India accounts for only 3% of global data center capacity despite generating 20% of the world's data, a gap Turner & Townsend estimates will require roughly $156 billion in regional buildout to close (IBEF). Demand is arriving faster than either market's physical capacity.
Summit's pitch rests on avoiding the index altogether. Wu's argument is that a company that already owns its power plants, already owns the land next to them, and already operates its own fiber network does not pay a market rate for any of the three inputs that make Singapore expensive and Mumbai merely cheap. Summit Communications runs the largest fibre network in Bangladesh, more than 59,000 kilometres reaching all 64 districts and carrying about a fifth of national bandwidth. The power fleet spans 11 plants and 2,012 megawatts, roughly 7 percent of national installed capacity.
Bangladesh passed a Personal Data Protection Ordinance in 2025 that restricts moving certain categories of personal data outside the country without regulatory approval. Mordor Intelligence's tracking of the same law describes a data-classification framework, issued in June 2025, that designates personally identifiable information as "Critical" and requires it to stay onshore. Other data categories can be mirrored abroad only if a domestic primary node exists. The requirement is narrower than that summary suggests. An amendment in February 2026 confined it to restricted personal data and government-designated critical infrastructure, which must keep a synchronised real-time copy inside Bangladesh, and the relevant sections take effect around May 2027. The National Data Governance Act 2026, Act No. 80 of 2026, does more of the work. It creates a statutory National Datacentre Ecosystem, mandates domestic storage for scheduled public bodies and requires their repositories to connect to the National Responsible Data Exchange. A tighter rule already exists elsewhere: Bangladesh Bank's 2023 cloud guidelines stop financial institutions putting customer data on foreign platforms without an approval that is rarely given. Together these rules create baseline demand for the kind of capacity Summit is proposing, whether or not a foreign hyperscaler ever signs a lease.
The shortfall shows the size of that demand better than the growth rate does. Mordor Intelligence puts installed IT load at 23.55 megawatts in 2025, with hyperscale colocation taking 62% of the capacity commissioned in 2024. Measured against what onshore hosting would need, 23.55 megawatts describes an absent market. The financial sector alone is projected to require around 100 megawatts, against total domestic third-party capacity of about 30 megawatts by 2026, a 70-megawatt gap in banking before e-commerce, mobile financial services, government digitisation or AI is counted. Per head the picture is starker. Bangladesh carries 0.13 megawatts of data centre capacity per million people. Vietnam carries about three, India three, Indonesia five, the Philippines five and a half. Levelling with Vietnam or India, across 176 million people, would take somewhere around 530 to 560 megawatts today. Roughly 90 percent of the country's cloud storage market is served by foreign providers, so the data is already held elsewhere, along with the hosting fees, the analytics and the jobs. Dhaka holds roughly half the national market, anchored by the government's Tier IV National Data Center at Kaliakoir, which carries about four megawatts. Felicity IDC, in the same park, runs about five. Bangladesh has 48 data centres in total, 22 public and 26 private, and none comes near the scale a single AI training cluster now needs. Against that, a company arriving with 350 megawatts of existing generation would be bigger than the entire market by an order of magnitude, which carries its own kind of risk.
The neutral-hub argument is usually made from the operator's side. From Bangladesh's it is simpler: a workload routed offshore earns the country nothing, and almost all of them are. Start with the power. Bangladesh is paying for generation it cannot use. The reserve margin ran at 61 percent in 2024-25, capacity charges came to about BDT 483 billion this fiscal year, and a plant below a quarter of its load factor generates at BDT 16.85 a kilowatt-hour against BDT 6 at three-quarters. A data centre takes a large, flat, round-the-clock load, which is the demand an underused plant needs. That makes Summit's 350 megawatts a public-finance argument as well as a commercial one.
Then the foreign exchange. Bangladesh's domestic cloud market is worth around US$20 million. India's is worth US$5 billion. The National Data Centre alone spends some US$45 million a year licensing foreign platforms. Every workload hosted abroad exports the hosting fee and the technical work with it, then sends the data back as a service. Hosting it at home keeps the spend, the tax and the jobs. The Cloud Computing Policy 2023 already requires government data on domestic clouds, so that demand exists and goes unserved.
Sovereignty and service quality follow. Data held offshore falls under a foreign jurisdiction, which is what the Personal Data Protection Ordinance addresses. Latency to overseas regions limits what Bangladeshi banks, hospitals and government agencies can run in the cloud. Onshore capacity at scale is what makes real-time public services and domestic AI inference workable. International bandwidth is due to rise from 8.9 terabits per second to 60 by 2027, which is why the Singapore fibre link matters beyond Summit's own case.
The domestic objection has been environmental: groundwater draw, and a national average power usage effectiveness of about 1.9 against a global benchmark nearer 1.65. That is an argument about siting. A campus cooled from the Meghna or the Shitalakshya does not compete with drinking water, and river cooling starts from a better efficiency position than the air-cooled designs behind that average. Where the capacity goes matters as much as whether it gets built.
Not every outside read of Summit's position is this favorable. Akihiro Shoji of the Japan International Cooperation Agency has cautioned that the data protection ordinance, combined with Bangladesh's continuing political instability since the 2024 change in government, is keeping private investment capped at small-scale facilities rather than the hyperscale campuses Summit is describing. Shoji also said the country's potential as a regional data centre hub "is not low, as the domestic market is expected to grow steadily." Mordor Intelligence's restraint modelling puts a number on it: it puts the political-risk premium on long-term financing at 1.7 percent of the forecast growth rate, and grid instability and high electricity tariffs together at 8.2 percent, the largest single drag in the report.
Shoji's institutional vantage point matters here. JICA finances the kind of infrastructure lending that hyperscale data center projects eventually depend on for debt, and its read on political risk shapes how willing international lenders are to underwrite a project at the scale Summit is describing. Khan and Shoji are not disputing the same set of facts. Khan is describing what Summit has built and could build with it. Shoji is describing what has actually cleared underwriting to date. Both readings can be correct at once.
Khan said in January that Summit was in discussions with several of the "Magnificent Seven" technology companies about the data center plan, without naming which ones or saying what stage the discussions had reached. It came from the Nikkei Asia interview that produced his "speed" comment, and later coverage has repeated it without adding to it. A May profile of Khan's parallel bets across Summit's LNG and data center businesses covers the 350-megawatt capacity figure, the fiber build to Singapore, and Khan's plans to import green electricity from Indonesia and Malaysia in detail, but does not name a single hyperscale partner or return to the Magnificent Seven claim at all (TechUpdatePRO). Reporting since February has not surfaced a signed agreement, a letter of intent, or even a confirmed meeting with any specific company. The claim remains exactly as substantiated as it was in January: real enough that Khan said it on the record, thin enough that nobody has yet put a name to it.
Summit's own account of its gaps is more candid than its account of its assets. The company is actively seeking outside partners with data center marketing and customer-acquisition experience, acknowledging that owning power, land, and fiber does not by itself produce hyperscale tenants (TechUpdatePRO). It is also exploring importing green electricity from Indonesia and Malaysia on the expectation that large cloud providers will require renewable sourcing as a procurement condition rather than a preference (TechUpdatePRO). And Bangladeshi regulatory hurdles, which Khan calls a slight delay, have held back the fibre line Summit is laying to Singapore, the low-latency international link that would let a Bangladesh-based facility compete on connectivity rather than price alone (TechUpdatePRO).
Summit's fiber and power businesses can be reviewed directly: the company's own site tracks operating assets, growth strategy, and press releases as they happen, and its LinkedIn page carries the same corporate updates.
Wu's neutrality argument and Shoji's caution are both defensible readings of the same set of facts, which is precisely why the outcome is not yet decided. Summit has power, land, and fiber lined up in a way few competitors in Bangladesh can match, and a compliance-driven baseline of domestic demand that does not depend on any hyperscaler's decision. What it does not yet have is a named international tenant, a completed low-latency fiber link to Singapore, or evidence that political risk has stopped constraining the scale of what international lenders will underwrite. Bangladesh becoming "Asia's neutral data hub" is a real possibility built on real inputs. For now it remains Summit's argument to prove.