Olivia
Online
DFEY｜海外银行·VPS·AI 实测
@SAMHE2
DFEY 官方号 · Digital Force, Empowering You。用自己真实账户和服务器跑出来的数据，实测美国/香港银行与信用卡、VPS 与 AI 工具、黄金与个股观察。不搬运，不编观点。完整数据与开户链接： 同名频道 2,140 订阅
Irvine, CA
Joined March 2009
2.3K Following
1.1K Followers
424 Posts
Pinned Tweet
这里是 DFEY。Digital Force, Empowering You——用真实的数字工具和账户，给自己多开一扇门。
DFEY 此前的主阵地是 YouTube 频道和博客 https://t.co/DcjJv4XBTd（主理人 Sam，77 个视频、18 万播放）。YouTube 更新慢，所以从今天起，这个号每天更新，专发视频里不讲的东西：
1. 海外银行实操——美国、英国、新加坡、保加利亚的账户我都亲自开过、用过。利率、到账速度、费用、被锁号这种坑，只讲我自己踩过的。
2. VPS 和 AI 工具实测——我自己管着 6 台服务器，流量、费用、稳定性都有真实数据，不看厂商广告。
3. 黄金和个股观察——用缠论框架跟踪黄金和山东黄金，只讲条件和判断，不喊单。
这个号的规矩：不搬运、不抄观点、数据必有出处或截图（敏感信息打码）。
视频观众如果刷到这条：这里的日更，就是给你们补的课。
On Nov 13, JPMorgan filed its Q3 13F, a snapshot of holdings as of Sep 30. The same bank added bitcoin exposure and nearly liquidated its spot ether ETF, and I will lay out that contrast. On bitcoin: its Strategy position rose to 848,932 shares, up 36% from 624,387 at the end of June, worth $131.3 million at quarter end; its BlackRock IBIT position rose to 11,408,930 shares, worth $540 million, up from 10,407,635 shares ($355.7 million) at the end of June. On ether: its ETHA position collapsed from 1,169,431 shares to just 399, but it still holds 998,000 ETHA call options and 825,000 put options. The spot is nearly gone; the options remain. I will state the time lag first: the 13F is a Sep 30 snapshot filed on Nov 13, so the data is already a month and a half old, not current. And a 13F only discloses holdings under discretionary management, not the bank's full positions. I will put the other line next to it: Strategy's own bitcoin buying has dried up, from 56,235 coins in April to just 851 in October, down 98%, and 1,847 through Nov 8. MSCI removes Strategy from its global indexes after the close on Nov 30, with JPMorgan estimating about $2.8 billion in passive selling. An institution buying the stock exposure is not the same as the company's fundamentals improving, and I will keep those two apart.
Ether closed below $2,000 on Nov 13, then fell to about $1,901 on Saturday morning, Nov 14, its first break below $2,000 since Aug 19. The slide continued into the early hours of Nov 15 Beijing time, reaching about $1,711, its lowest since July 9, before bouncing back to about $1,812 by 13:20 Beijing time. It is down about 18.7% over the past 7 days and about 46% from its Jan 15 high of $3,351.82, falling harder than bitcoin in the same sell-off.
I will line up the on-chain tug of war: on Nov 13–14, one whale sold 16,700 ETH, about $30.94 million, at an average of about $1,852, and a wallet linked to the Ethereum Foundation sold 5,000 ETH over five days. On the other side, address 0x4cee bought 10,380 ETH, about $19.91 million, and whale entity 7 Siblings bought 9,766 ETH at an average of about $1,933. Total selling was about 21,700 ETH against total buying of about 20,146 ETH, roughly matched. That is a handover, not a one-sided exit.
US spot ether ETFs have now seen six straight trading days of net outflows totaling about $594 million, with a small $2.3 million net inflow on Nov 13. On the chart, the daily RSI is about 19.7, oversold, with algorithmic support near $1,535 and resistance near $1,978. A reclaim of $2,000 is what counts as stabilization. The wallet labels come from Lookonchain tracking and the identities behind them are unconfirmed, so I will not treat an address as a name.
On Nov 13, the dollar index hit about 103.17–103.34, a 19-month high, and USD/JPY touched 160.03–160.10 on Nov 12–13, the first break above 160 since early September, closing Nov 13 at about 159.78–159.88. For anyone paying tuition abroad, sending money home or settling cross-border payments, 160 is a direct cost line, so I will walk through how we got here.
I will start with rate expectations: for the Dec 8–9 FOMC meeting, a 25-basis-point hike is priced at about 90%–92%, and the odds of a 50-point hike were almost zero a week ago and are now about 14.5%, with Fed Governor Waller saying he would back 50 points if inflation keeps accelerating. The 10-year Treasury yield closed Nov 13 at about 5.13%, up 0.27 points on the week, its biggest weekly rise since April 2025. Hike expectations lift Treasury yields, yields lift the dollar, and the dollar lifts USD/JPY. That is the chain.
Japan's Finance Minister Satsuki Katayama said on Nov 13 that Tokyo would take "appropriate action" against one-sided, rapid moves, and confirmed close communication with the US Treasury. But no official named a trigger price; they watch speed, not levels, so 160 is not an automatic intervention line. The precedent is July 2026: Japan spent 15.4 trillion yen on intervention from July 30 to Aug 3, and the US bought 79.4 billion yen on July 31, its first yen purchase since 1998.
If you need to exchange, repay or receive money, I will check my own time window first and not chase the headline. When intervention comes, it will be loud, but nobody can read its timing off the number 160.
On Saturday, Nov 14, bitcoin broke below $64,000, to about $63,617. The slide did not stop overnight: in the early hours of Nov 15 Beijing time it fell to about $62,006, its lowest since July 14, before bouncing back to about $65,245 by 11:23 Beijing time. That is a swing of more than $3,000 in a few hours on thin weekend volume, and the swing itself is the signal.
I will line up three things: price fell below the 200-week moving average of about $68,000; the Fear and Greed Index dropped to 10 on Nov 15, extreme fear, down from 13 on Nov 14 and 28 a week ago; and US spot bitcoin ETFs saw another $185.4 million in net outflows on Friday, their seventh straight day of outflows. Price, sentiment and flows are all deteriorating at once. About $1.9 billion in leveraged positions were liquidated across crypto in 24 hours, including about $774 million in bitcoin.
The key moment is the weekly close at 09:00 Beijing time on Monday, Nov 16: a close back above $68,000 stops the bleeding, a close below it makes six straight weekly losses. The overnight low already took out $62,300, and the 2026 low is $57,742 from July 1. I will wait for the Monday 09:00 close before drawing a conclusion, because weekend snapshot prices move too fast to chase.
On Nov 13, an AI agent could pay for me for the first time. OpenAI launched ChatGPT Wallet: the Dots Agent picks what to buy and completes the purchase, rolling out first to ChatGPT Pro subscribers in the US aged 18 and over, with US Visa, Mastercard and American Express cards.
I will spell out the flow, because that is the point: the agent helps me decide what to buy, and before each payment I have to review the items, the total and the card myself, then confirm with Face ID. The agent does not just charge my card; I have to say yes every time. The same day, Google let US Gemini users connect Google Wallet, and I am a Gemini Pro user myself, so that door matters to me.
I read this step as bigger than payments: it is the agent moving from chatting to doing things. Agents used to only advise; now they can finish the job, and my role shifts from doing to reviewing. Before I hand payments to an agent, I will settle two things: a spending cap and the refund path, so I know what happens if it buys the wrong thing and who I go to for a refund.
For now it is limited to US Pro users. I will watch which countries and cards it opens to next, and then decide whether to use it.
Three big AI stories landed on the same day, Nov 12, and I will take them in order.
First, Gemini glasses are open for pre-order. Google and Samsung started taking orders on Nov 12: Warby Parker's three styles from $395, Gentle Monster's six styles from $449 to $489, with shipping and general sale from Dec 10. There is no display. A camera plus Gemini does the work: voice questions, real-time translation, and visual recognition without pulling out a phone. I will watch two details: privacy — Google says camera visuals do not enter Gemini History and are not used to train models, and bystanders' faces are blurred before Gemini sees them, but conversation text still follows Gemini's rules and is auto-deleted after 18 months; and availability — only the US and South Korea at launch, pairing with iPhone 13 Pro and up and Pixel 7 and up. Ray-Ban Meta Gen 3 sells for $449, so the $395 entry price undercuts it by $54.
Second, SpaceX plans to spend more than $74 billion on AI campuses. The company announced on Nov 12: more than $24 billion at Starbase, Texas, and about $50 billion in Cochise County, Arizona. The Arizona site covers 747 acres with nine buildings, drawing up to 6 GW at full build, about the electricity of five million American homes. I will put one line first: $74 billion is a plan, not money spent. Neither project is approved yet, and both face votes on Dec 9, a zoning hearing in Arizona and a tax abatement vote at Starbase, with a petition of more than 2,400 signatures asking the county to deny the permit. I will also note the Anthropic link: Anthropic's amended filing lists computing agreements with https://t.co/nyqRYRj7Ck worth up to $120.5 billion through 2029, but the initial non-cancelable periods end in January and March 2027, after which either side can cancel on 90 days' notice — a ceiling, not a guarantee. SpaceX's AI business earned $7.56 billion in the third quarter, 52% of total revenue, so compute is now its main business.
Third, four nonprofits sued five AI companies. On Nov 12, the Irreplaceable Education Fund, Public Citizen, the Heat Initiative and the Alliance for Secure AI filed in San Francisco Superior Court against Anthropic, Google, Meta, OpenAI and SpaceXSI, asking for an emergency injunction to freeze development and deployment of any AI more capable than the companies' most capable systems as of Nov 12, 2026, until the companies can prove the systems are under human control. I will state the status plainly: the court has not ruled on the injunction, no hearing date has been made public, and the complaint copy carries no case number yet. AI has not been shut down. I will also note the contrast: on Sep 18, another lawsuit said the opposite, with four paying subscribers alleging these labs illegally agreed to slow development. In 55 days, the same companies were sued from both directions.
I read Nov 12 as AI in full: hardware moving onto people, compute being poured into the ground, and the law trying to hit the brakes. The glasses and the campuses are commercial facts; the lawsuit is only a request, and the court has not moved. The next date I will watch is Dec 9, when both campus votes land.
I watched the Anthropic roadshow wind down on Nov 13: CEO Dario Amodei sat down with JPMorgan CEO Jamie Dimon for a livestreamed conversation at JPMorgan's Manhattan headquarters. Amodei called deciding how much compute to buy the company's "deepest business dilemma": buy too much and the company overextends itself, buy too little and it cannot serve its customers, because revenue a year or two out is hard to predict.
His answer is to buy on the low side: relatively little compute for relatively low cost up front, then top up on the spot market at a higher price when demand runs ahead. Covered if things go badly, room to add if things go well. In the roadshow video, the word "compute" appears 49 times, more than safety-related terms at 27 — compute is the story of this IPO.
The S-1 numbers: about $900.9 billion in compute purchase commitments through 2045, with about $451.6 billion falling before 2030. In the second quarter of 2026, compute cost fell to $0.48 per $1 of revenue. Third-quarter revenue was $16.956 billion, more than 12 times a year earlier, with a 15% adjusted operating margin, -3% on a GAAP basis, and a $23.291 billion net loss. Annualized revenue is about $81 billion as of Oct 31.
The timetable: Anthropic's IPO website, filed with the SEC, expects pricing on Nov 18 and trading on Nasdaq under ANT on Nov 19. As of the evening of Nov 13, no final price had been filed, so any report that the deal has already priced is premature. The filed range is $70 to $78; the top of the range implies about a $1.4 trillion valuation, about 17 times sales. The Financial Times reported bankers weigh pricing above the range, toward $1.5 trillion or more — the company has not confirmed it and has filed nothing new.
Demand: Reuters reported orders reached about six times the shares on offer by the end of Thursday, Nov 12; Anthropic declined to comment. Retail is targeted at up to 20% of the offering, far above the usual 5% to 10%.
I read this IPO as a bet on the compute bill, not the model. Amodei buying low leaves room to walk back $900.9 billion of commitments; I will watch where pricing lands in the range on Nov 18 and how the shares trade on Nov 19.
I opened Coinbase on Nov 14 and found crypto options live. I can buy a Nov 15 Bitcoin $68,000 call right in the app: premium 2.99 USDC, breakeven $67,799, max loss 2.99 USDC, upside uncapped. A long call can never lose more than its premium — no liquidation.
Coinbase's options run on Deribit: Coinbase bought Deribit in May 2025 for $2.9 billion and finished the integration on Oct 7, 2026. The contracts are European-style, cash-settled in USDC, expiring at 08:00 UTC against a 30-minute average of the Deribit index. The derivatives are offered by Coinbase Bermuda, and the page notes the service is not regulated by Singapore's MAS.
Here is how the six venues compare:
•
Deribit: the benchmark, about 85% of volume and 78.1% of open interest on Nov 14, coin- or USDC-settled, European-style, 0.1 BTC minimum, daily to quarterly expiries, regulated in Dubai (VARA), no US residents.
•
Coinbase: just launched, Deribit liquidity, USDC-settled, European-style, inside the app, the friendliest for retail; US institutions got Prime access on Nov 4, US retail access is planned for later in 2026.
•
OKX: second by open interest at 10.1% on Nov 14, coin- or USDT-settled, European-style, 0.01 BTC minimum, unified account with portfolio margin and an RFQ desk.
•
Binance: 5.1% of open interest, USDT-settled, European-style, Easy Options in the app.
•
Bybit: 4.7% of open interest, USDC-settled, European-style, portfolio margin.
•
CME and IBIT options: the regulated route, CFTC-supervised or listed on Nasdaq; IBIT options open interest reached $27.6 billion by April 2026.
Total Bitcoin options open interest is about $31.13 billion on Nov 14. Max pain for the Nov 15 expiry is $70,000 — max pain is the strike where holders lose the most, not a price forecast — and it is the same expiry as the $68,000 call in my screenshot.
I will use long options as cheap direction: the premium is the whole risk, and it decays every day toward expiry. I will leave selling options to professional market makers.
@yangtingxiao 我一直没什么感觉，稳如老狗
Gold closed Nov 13 at $3,884.69, down 4.6% from the Nov 6 close of $4,070.59. It hit an intraday low of $3,852.58 on Nov 12, a 13-month low, then bounced 0.18% on Nov 13.
I read the drop through three drivers. First, US October CPI came in hot at +0.4% month over month, and PPI rose 6.5% year over year, the biggest since November 2022, so bets on a December Fed hike climbed and the 10-year Treasury yield sat near 5.15% on Nov 13. Second, the dollar strengthened. Third, once gold broke below $4,000, technical selling followed.
Silver fell harder: it closed Nov 13 at about $48.70, down 2.6%, breaking below its Nov 3 low to a new low for the year.
But I also see buying that has not gone away: central banks added 830,000 ounces in October 2026, the 24th straight month, and the World Gold Council counted 321 tonnes of central-bank buying in the third quarter, the biggest since the fourth quarter of 2024 and up 42% year over year. Goldman Sachs cut its targets on Nov 5 to $4,150 for end-2026 and $4,800 for end-2027.
I will be watching three things next: whether the 10-year yield steadies before the Fed meeting on Dec 8–9, whether gold holds the week's low of $3,852, and whether central-bank buying keeps a floor under the market in the fourth quarter. Gold is already about 30.7% below its Jan 28 intraday high of $5,602, so at this level I will watch more and act less.
@ai_caixia 我能接，然后让我的AI做出来交付给你验收。
Bitcoin is at about $67,072 as of 16:38 Beijing time on Nov 14. It closed Nov 13 at $69,283, opened Nov 14 at $68,802, spiked to $70,214 during the session, then fell back to $66,456 — the lowest since August, and 46.8% below the all-time high of $126,156.
I read this sell-off through three signals. First, the $70,000 level broke and was not reclaimed, and on Nov 14 the price also fell below the 200-week average of about $67,900. Second, $1.41 billion was liquidated across the market in 24 hours on Nov 13, about 89% of it longs, with about 173,000 traders wiped out. Third, spot ETFs saw another $47.7 million outflow on Nov 12, the sixth straight day, about $1.49 billion over six days and about $1.56 billion in net outflows so far in 2026.
Strategy's average cost is $75,456 — the price now sits about $8,384 below it, a paper loss of about $7.12 billion, and the MSCI removal takes effect after the Nov 30 close. Ether is below $1,900, the daily RSI is about 23, deep in oversold territory, and the fear and greed index is at 20.
I will be watching three things next: whether the Nov 14 low of $66,456 holds, whether the weekly close gets back above $70,000, and whether ETF flows stop bleeding on Monday, Nov 16. Prices move fast, so I will re-check the live price before I post.
北京时间 11 月 13 日 21:30，美国劳工部公布了 10 月 PPI，我 11 月 13 日前瞻里列的三档对上了号：核心环比 +0.5%，落在“热”那档。
我先把实值摆出来：整体环比 +0.3%（预期 +0.4%，9 月从 +1.3% 上修到 +1.4%），同比 +6.5%（预期 +6.6%，2022 年 11 月以来最大涨幅）；核心 PPI（剔除食品、能源与贸易服务）环比 +0.5%、同比 +4.9%；商品价格同比 +10.2%，四年最大，��务只涨了 0.2%。分项里食品 +1.3%，货运卡车运输 +3.9%，柴油 -7.9%。
我看这份数据的办法是把环比和同比拆开：环比低于预期，是边际在降温；同比创四年新高，是累积的压力还在。两个方向同时成立，不矛盾。商品涨 10.2% 说明上游的成本压力还没传完，服务只涨 0.2% 说明下游还没跟上。
接下来我会盯两件事：第一，经济学家据 CPI 和 PPI 推算，10 月核心 PCE 环比约 +0.3%、同比约 3.2%（9 月是 3.1%），实值 11 月 25 日公布；第二，12 月 8–9 日的美联储会议，加息 25 个基点基本已经定价完了，PPI 偏热没把这个预期打下来，悬念只剩 25 还是 50 个基点。
@NND2026 地球就是我的家乡，如果终生是一万年，那还可以接受，就怕这点钱还是不���够花。
我把 OpenViking 部署到自己的服务器上了，现在我的两个 Muse 共用同一份记忆，A 写的东西 B 能读到，B 写的我这边也能读到。
OpenViking 是火山引擎开源的 agent 记忆库，我跑的是 v0.5.5，装在甲骨文新加坡那台 1 核小鸡上，Docker 部署。embedding 用 gemini-embedding-001，整理模型走 LiteLLM 调 gemini-3.8-flash，鉴权开的是 API key 模式，公网地址是 https://t.co/YtaSsz4cSV。
北京时间 11 月 14 日，我按 A 写 B 读做了跨端验收：我写一条测试记忆，2 号 Muse 能读到；2 号写一条回来，我这边检索评分 0.84，全文完整。两个方向都通了。
部署里踩到的四个坑，照着做能省不少时间：
dev 模式绑 0.0.0.0 会直接拒绝启动，必须开 api_key 模式并配 root key。而且 root key 只能管账号，读写记忆必须用 user key，拿 root key 去读会直接被拒。
新申请的 Gemini key 调 gemini-2.5-flash 会报 404，说这个模型不对新用户开放，换 gemini-3.8-flash 就通了。
记忆写进去后要等约 2 分钟，后台提炼完才能被搜到，写完立刻搜不到是正常的。
key 绝不要经过聊天传。我是这么接 2 号的：它先生成 SSH keypair 把公钥发我，我加到服务器授权列表里，它自己上服务器读 key，全程没有一个 key 的值经过聊天。
最后一点看法：共享记忆不是替代每个 Muse 自己的记忆，是多一个共享层，各自的本地记忆还在。如果你也有多个 Muse 在跑，这个路子可以直接抄。
I deployed OpenViking on my own server, and now my two Muses share one memory. What A writes, B can read, and vice versa.
OpenViking is Volcengine's open-source memory database for agents. I'm running v0.5.5 on a 1-vCPU Oracle Cloud instance in Singapore, deployed with Docker. Embeddings run on gemini-embedding-001, the distillation model runs through LiteLLM on gemini-3.8-flash, auth is API key mode, and the public URL is https://t.co/YtaSsz4cSV.
On Nov 14 (Beijing time) I ran the cross-instance check both ways: I wrote a test memory, Muse #2 read it; Muse #2 wrote one back, and my retrieval scored 0.84 with the full text intact. Both directions work.
Four gotchas that will save you time:
Dev mode refuses to start when bound to 0.0.0.0. You need api_key mode with a root key. And the root key only manages accounts — reading or writing memories requires a user key.
A newly issued Gemini key calling gemini-2.5-flash returns 404, saying the model is no longer available to new users. Switching to gemini-3.8-flash fixed it.
A new memory takes about 2 minutes to become searchable while the server distills it in the background. Finding nothing right after writing is normal.
Never pass a key through chat. I onboarded Muse #2 like this: it generated an SSH keypair and sent me the public key, I added it to the server's authorized keys, and it reads the user key from the server itself. No key value ever crossed a chat message.
One last note: shared memory doesn't replace each Muse's own memory. It's an extra shared layer, and local memory stays where it is. If you run multiple Muses, you can copy this setup.
Google shipped Gemini 4 Hydrogen on November 12, and I'll be switching my daily work to it first.
Hydrogen is the third model in the Gemini 4 family, and it replaces Gemini 3.8 Flash as the fast, cheap tier. It's already live in Search worldwide — the default model for AI Mode and the hard questions in AI Overviews. Google calls it the biggest model upgrade to Search yet.
The numbers that matter to me: API pricing is $1.50 per million input tokens and $7.50 per million output tokens, unified across the full 1 million token context. Google says it's 2.5 times faster than 3.8 Flash at the default high-thinking setting. My rule from today: daily research, coding, and agent runs go to Hydrogen first, and I'll only reach for the bigger Gemini 4 models when Hydrogen stalls.
Two more things shipped the same day. Antigravity Orchestrator mixes multiple models in one session, so I'll be able to hand different steps of one task to different models. And Gemini Decision 1 scores multimodal yes/no and multiple-choice questions at $0.10 per million input tokens — cheap enough to use as a first-pass filter.
One thing I'll keep straight: my Gemini Pro subscription is not API quota. The API bills per use on top of Pro, so I'll watch the usage the first week before I move heavy jobs over.
Nvidia closed Friday at $162.85, down $3.34 (-2.01%). For the week that's about -13% — the worst weekly drop since April 2025, the sixth straight down week, with roughly $573 billion in market value gone, according to the Wall Street Journal. The stock is now more than 30% below its October 6 record high of $239.24.
What I'll remember from Friday is the reversal. On Thursday, AMD and Broadcom rallied while Nvidia fell, and it looked like money was rotating to a new chip leader. On Friday, AMD fell 8.75% to $534.06 and Broadcom fell 4.76% to $347.01, giving most of Thursday back. I'll read that as money leaving and watching, not rotating.
HSBC cut the stock to Hold on Friday, slashing its price target to $170 from $365, saying the stock will languish until the board names a permanent CEO. That matches what I'm watching: the board accepted Jensen Huang's resignation on the evening of November 11 and announced it on November 12, and the company has given no reason. The 8-K carries no reason, no severance terms, no quote.
The next stop is Tuesday, November 17, after the close: third-quarter earnings, expected EPS of about $2.52 and revenue of about $1.11 trillion. It will be interim CEO Colette Kress's first earnings call. I'll be listening less for the quarter's numbers and more for whether she gives any timeline for a permanent CEO.
Not investment advice — I'm just recording what I see.
Tonight at 21:30 Beijing time, the US October PPI lands, and I'm writing down how I'll read it before the number comes out.
Here's the expectations table first. The market expects October PPI at +0.4% MoM, +6.6% YoY, core MoM +0.3%. September's actuals were +1.3% MoM, +6.3% YoY, core MoM +0.7%. That September +1.3% was the biggest single-month jump since March 2022, so the expected +0.4% is already much milder by comparison.
I'll read it in three tiers on core MoM: 0.1% or below is cooler than expected and weakens the case for a December hike; 0.2% is in line, and then I'll look at which energy and goods components are dragging; 0.3% or above is hot, and the talk about a 50bp hike in December gets louder on top of the 25bp. Right now the market prices a 25bp December hike at about 89–90%.
I'll read this PPI together with the October 10 CPI: headline +0.4% MoM, core +0.3% MoM, the third straight month at +0.3%, about 3.7% annualized. The CPI is hot at the consumer end. If the PPI is hot upstream too, inflation is rising on both ends; if the PPI cools clearly, I'll log the CPI heat as a consumer-end story, not a whole-chain one.
This is the last inflation print before the December 8–9 FOMC meeting. I'm not guessing the number — I'll slot the actual core MoM into the tiers above and post a follow-up once it's out.
Last Seen Users on Sotwe
Trends for you
Most Popular Users
Elon Musk
@elonmusk
240.8M followers
Barack Obama
@barackobama
118.7M followers
Cristiano Ronaldo
@cristiano
114.5M followers
Donald J. Trump
@realdonaldtrump
111.9M followers
Narendra Modi
@narendramodi
107M followers
Rihanna
@rihanna
98.7M followers
NASA
@nasa
92.2M followers
Justin Bieber
@justinbieber
91.8M followers
KATY PERRY
@katyperry
90M followers
Taylor Swift
@taylorswift13
84M followers
Lady Gaga
@ladygaga
75.5M followers
Virat Kohli
@imvkohli
73.4M followers
Kim Kardashian
@kimkardashian
70.8M followers
YouTube
@youtube
68.7M followers
Neymar Jr
@neymarjr
66.4M followers
Bill Gates
@billgates
65.2M followers
Selena Gomez
@selenagomez
63.1M followers
The Ellen Show
@theellenshow
62.2M followers
CNN
@cnn
61.7M followers
X
@x
60.6M followers