Actualités financières en direct

candlefeed The official Python client for CandleFeed — crypto market data that lands straight in a pandas DataFrame. OHLCV, funding rates, open interest, liquidations, long/short ratio, taker volume, basis, and Deribit options — across …

candlefeed The official Python client for CandleFeed — crypto market data that lands straight in a pandas DataFrame. OHLCV, funding rates, open interest, liquidations, long/short ratio, taker volume, basis, and Deribit options — across Binance (perp + spot), Bybit, OKX, dYdX, Hyperliquid, Huobi and Deribit. Auth, cursor pagination, and rate limits are handled for you. One call, one DataFrame. REST only — there is no WebSocket feed and no announced date for one. pip install candlefeed Time to first DataFrame You'll need an API key: get a free one. Free plan, no card, takes about a minute. from candlefeed import CandleFeed cf = CandleFeed ( api_key = "cf_live_..." ) # or set CANDLEFEED_API_KEY df = cf . get_ohlcv ( "BTCUSDT" , interval = "1h" , limit = 5 ) print ( df ) open high low close volume quote_volume time 2026-06-06 18:00:00+00:00 69841.0 70120.5 69770.0 70011.2 1183.42 8.27e+07 2026-06-06 19:00:00+00:00 70011.2 70250.0 69905.1 70180.9 964.18 6.77e+07 ... The frame is indexed by a tz-aware DatetimeIndex and every numeric column is a float — ready for .resample() , .rolling() , or a backtest loop. Authentication Pass your key directly or via the environment: cf = CandleFeed ( api_key = "cf_live_..." ) # or export CANDLEFEED_API_KEY = cf_live_ ... cf = CandleFeed () Get a free key at candlefeed.ai/signup. Tier Requests/day Max rows/request Symbols Datasets History Exchanges Free 100 1,000 BTC, ETH, SOL, XRP, DOGE OHLCV + funding last 30 days Binance Builder 1,000 10,000 all Binance REST datasets; L2 sample days only full Binance Advanced (id pro ) 10,000 50,000 all + Deribit options; L2 sample days only full all venues Pro (id pro_depth ) 10,000 50,000 all Advanced + full published L2 archive full all venues Builder ($29/month) includes Binance REST datasets; Advanced ($99/month) adds other venues and Deribit options. Both include L2 sample days only: the 1st of each month, subject to the 10 GiB monthly allowance. Pro ($149/month) adds the full published L2 archive. Pro includes the full published L2 archive: book from 2026-06-04 and trades from 2026-06-06 for the original 10 symbols; the other 15 symbols start on 2026-09-26. First days are partial. Everything beyond Binance — Bybit, OKX, dYdX, Hyperliquid, Huobi, Deribit — needs Advanced or higher. Endpoints Method Data Key params get_ohlcv / get_candles OHLCV candles symbol, exchange, interval, start, end, limit get_funding_rates Per-exchange funding symbol, exchange, ... get_funding_rates_aggregated OI-weighted funding across venues symbol, interval, exchanges get_open_interest Open interest symbol, exchange, interval get_liquidations Liquidation events (tick or bucketed) symbol, exchange, side, interval get_liquidations_aggregated Pre-aggregated liquidation history symbol, exchange, interval get_long_short_ratio Long/short account ratio symbol, exchange, interval, ratio_type get_taker_volume Taker buy/sell volume symbol, exchange, ... get_basis Futures basis / premium symbol, exchange, interval get_combined Time-aligned multi-dataset frame symbol, interval, fields get_options Deribit options chain + greeks currency, instrument_name, expiry symbols / exchanges / datasets / status Metadata — Intervals — OHLCV: 1m 5m 15m 1h 4h 1d · aggregated funding: 1h 4h 1d · open interest: 5m 15m 1h 4h 1d · bucketed liquidations: 1m 5m 15m 1h 4h 1d · aggregated liquidations: 1h 4h 6h 8h 12h 1d · long/short ratio: 5m 15m 1h 4h 1d · basis: 5m 1h 4h . The same lists are importable as candlefeed.OHLCV_INTERVALS , candlefeed.BASIS_INTERVALS , and friends. How far back the data goes The honest version, because a backtest will find out anyway. Full matrix at candlefeed.ai/coverage. Dataset Depth OHLCV Binance spot 2017 · Binance perp 2019 · OKX/Bybit 2020 · Hyperliquid 1d 2023 / 4h 2024 / 1h 2025 / 1m–15m 2026 · dYdX 2025 Funding rates Binance 2020 · Hyperliquid hourly from each contract's HL listing (BTC/ETH/SOL 2023-05-12) · OKX Dec 2025 · Bybit Jan 2026 · dYdX Mar 2026 Open interest Binance Sep 2020 · other venues Mar 2026 Liquidations — aggregated 1d runs from each contract's listing: Binance 2019-09, Bybit 2020-01, OKX/Huobi 2022-04, Hyperliquid 2026-05. Sub-daily buckets are forward-built (2023→2026 by venue) — use interval="1d" for multi-year work Liquidations — tick Forward-collected from late May 2026: OKX 05-27, Binance/Bybit/Hyperliquid 05-28, Huobi 06-02. No pre-2026 tick history anywhere Long/short ratio, taker volume Binance only — ratio 2020, taker volume from listing (BTC 2019-09) Basis (spot–perp) Binance only, all 52 symbols — 5m from 2026-04-27, 1h from 2025-12-10, 4h from 2025-05-31 Deribit options Rolling ~60-day API window (older snapshots archived); greeks native from Jun 2026, Black-76 before that Symbol universe: 52 on Binance, 50 Bybit, 45 OKX, 19 Hyperliquid, ~5 dYdX. Where an endpoint knows its own depth it tells you — meta.history_from and meta.source come back on cf.last_meta and on each frame as df.attrs["meta"] . Ask for a window and the client transparently follows the API's next_cursor until the range is complete, concatenating into one DataFrame: df = cf . get_funding_rates ( "BTCUSDT" , exchange = "binance" , start = "2026-01-01" , end = "2026-03-01" , ) # many pages → a single tidy frame Cap the result with max_rows=... , control page size with limit=... , or disable auto-paging entirely with paginate=False to fetch a single page and inspect the cursor yourself. agg = cf . get_funding_rates_aggregated ( "BTCUSDT" , interval = "1h" , exchanges = [ "binance" , "bybit" , "okx" ], ) agg [[ "weighted_funding_rate" , "total_oi_usd" , "exchange_count" ]] . tail () panel = cf . get_combined ( "BTCUSDT" , interval = "1h" , fields = [ "ohlcv" , "funding_rate" , "open_interest" ], ) # OHLCV base timeline with funding + OI forward-filled onto each candle Order book (L2) and tick trades: daily files Binance USD-M order book diffs with REST snapshots ( "book" ) and raw tick trades ( "trades" ) come as daily Parquet files, not DataFrames. download_l2 fetches them, checks each file's size and SHA-256, and skips anything already on disk with a matching hash, so a rerun picks up where it stopped. Downloads need CPython on Linux or macOS: files are written with directory-relative, no-follow operations so nothing can be redirected outside dest_dir , and on platforms without them (Windows) download_l2 refuses to run rather than fall back to weaker path checks. cf = CandleFeed ( api_key = "cf_live_..." ) out = cf . download_l2 ( "book" , "BTCUSDT" , "2026-09-01" , "2026-09-30" , "data/" ) # data/book/binance/BTCUSDT/2026-09-01/depth/00.parquet ... depth/23.parquet, snapshot.parquet, manifest.json print ( len ( out [ "downloaded" ]), "files," , out [ "bytes" ] / 1e9 , "GB" ) print ( out [ "missing" ]) # days not available, each with a reason cf.l2_files(dataset, symbol, start, end) returns the listing itself: per day the files (name, size, sha256, a download link valid for 15 minutes) and a QC summary from that day's manifest. Pro and Enterprise get every day. Other plans get the 1st of each month, up to 10 GiB of new files per account per month; downloading the same file again that month doesn't count twice. When an allowance runs out the client raises QuotaExceededError straight away instead of retrying. cf.l2_coverage() and cf.l2_gaps() return the public coverage table and gap log (no plan needed). A few sample days can be downloaded without an account (version 0.3.2 or later), with the same checks: from candlefeed import CandleFeed public = CandleFeed ( public = True ) # never sends a key print ( public . l2_sample ()[ "samples" ]) # the days on offer public . download_l2_sample ( "BTCUSDT" , "2026-10-01" , "data/" ) Rebuilding the book candlefeed.l2book turns a downloaded book day into an order book you can query. It needs pyarrow: pip install "candlefeed[l2]" from candlefeed.l2book import L2Book book = L2Book . load ( "data/" , "BTCUSDT" , "2026-09-01" ) # one day, or pass an end date for a range view = book . book_at ( "2026-09-01T13:05:00Z" , levels = 10 ) # top 10 a side after every event up to then print ( view . best_bid , view . best_ask , view . spread_bps ) print ( view . bids . head ()) # price, qty, best first spreads = book . spread_series ( "1s" ) # bid, ask, mid, spread, spread_bps depth = book . depth_at ( "2026-09-01T13:05:00Z" , bps = [ 10 , 50 ]) # base and quote size within 10 and 50 bps of mid for v in book . iterate ( "2026-09-01T13:00Z" , "2026-09-01T14:00Z" , every = "1min" ): ... It applies the published rule: anchor on a snapshot that isn't in_gap , apply the first event with u >= lastUpdateId whole even when the snapshot's id falls inside it (the straddle case: 848 of the 861 BTCUSDT snapshots on 25 September 2026), and drop the book at every break in the update chain until a snapshot anchors it again. Moments with no trustworthy book raise BookUnavailable in book_at and come back as NaN in spread_series . book.segments() lists the unbroken stretches of the day and when each was anchored. Each query anchors on the latest snapshot at or before it, so results don't depend on where you start reading. A 10-million-row synthetic day loads in under a second and gives a one-second spread series for the full day in about 4 seconds on an M-series Mac; a BTC day has about 100 million rows. Real-day runtime and peak memory have not been validated; synthetic timings are not production benchmarks. Depth outside the anchor snapshot's known price window is partial. Report completeness per band and exclude incomplete samples from full-depth statistics. depth_at and spread_series(depth_bps=...) carry a *_complete flag per band, and book_at returns only levels inside the window ( bids_complete / asks_complete say when that's fewer than asked for; an empty side has no best price). Days are checked against their manifest before use, so an incomplete or mixed download raises IncompleteDay , and a moment after the last loaded event raises BookUnavailable . row_cache_bytes (default 2 GiB) limits the decoded diff rows kept between queries, and an hour file that wouldn't fit in it is refused. It isn't a bound on the rebuild's total memory: the event index (about 80 bytes per diff event, roughly 70 MB for a BTCUSDT day), the decoded snapshots, Arrow's read buffers and any series you build come on top. Hard per-file caps are checked from each footer before decoding: snapshot files at 2 million rows and 512 MiB decoded, hour files at 2 GiB decoded, footers at 64 MB, with column types validated and strings read as dictionaries. The manifest has to be the one for that exchange, symbol, date and generation, every file has to sit under its full published key, and each file's rows have to carry the requested symbol. The SHA-256 checks prove the files are consistent with what CandleFeed published for that day. They aren't a signature: hashes delivered by the same service as the files can't detect that service itself being compromised or malicious. Error handling Every API error maps to a typed exception carrying the API code and message : from candlefeed import ( CandleFeed , AuthenticationError , TierRestrictedError , InvalidParameterError , RateLimitError , ) cf = CandleFeed ( api_key = "cf_live_..." ) try : df = cf . get_options ( currency = "BTC" ) except TierRestrictedError as e : print ( e . message ) # "...requires the Advanced plan or higher... Upgrade at https://candlefeed.ai/pricing" except RateLimitError as e : print ( "retry after" , e . retry_after , "seconds" ) except ( AuthenticationError , InvalidParameterError ) as e : print ( e . code , e . message ) Exception HTTP When AuthenticationError 401 missing / invalid / revoked key TierRestrictedError 403 symbol, dataset, exchange, or history window above your plan InvalidParameterError 400 / 422 bad symbol, interval, or timestamp RateLimitError 429 raised only after the client's bounded backoff retries are exhausted QuotaExceededError 429 L2 monthly sample allowance or daily download limit reached; not retried CandleFeedError — base class; network / unexpected errors On HTTP 429 the client honors Retry-After / X-RateLimit-Reset and retries with exponential backoff before giving up. Remaining-quota headers are exposed on cf.last_rate_limit . Requirements Python ≥ 3.9, requests , and pandas . That's the whole dependency surface. Free API key — candlefeed.ai/signup Site — candlefeed.ai Pricing — candlefeed.ai/#pricing The software is MIT-licensed. CandleFeed data, including samples, is licensed for internal use under Terms §5.3. Published charts, statistics, and research must not include Raw Data or Substantially Raw Derivatives.

The October 5 filing brings BitGo Bank & Trust on board as an additional custodian for the HYPE tokens held by Grayscale's HYPG fund. Grayscale has filed a Form 8-K with the SEC covering …

The October 5 filing brings BitGo Bank & Trust on board as an additional custodian for the HYPE tokens held by Grayscale's HYPG fund. Grayscale has filed a Form 8-K with the SEC covering its Hyperliquid Staking ETF, which trades under the ticker HYPG. The filing, dated October 5, 2026, reports that BitGo Bank & Trust has been added as an additional custodian for the fund’s HYPE holdings, effective September 30, 2026. What the filing changes An 8-K is the SEC form companies use to flag material events between their regular quarterly and annual reports. Grayscale’s filing reports an amendment that lets BitGo hold part of the trust’s HYPE alongside the fund’s existing custody setup. As of October 5, 2026, HYPG held approximately $217.8 million in assets under management, with 6.54 million shares outstanding. HYPG charges a management fee of 0.29%. That was the lowest among US HYPE exchange-traded products when the fund launched. Advertisement Approximately 88.66% of the fund’s assets are staked, generating average gross staking rewards of 2.26%. Once fees come out, the yield investors actually pocket typically lands below that figure. A short but busy history HYPG has not existed for long, yet it has already changed names once. The fund started life as the Grayscale HYPE ETF before being renamed on May 26, 2026, a move intended to tie it more closely to Hyperliquid’s core branding. The news moving money, markets, and the world—before your day starts. Daily. Free. Join 34,000+ readers across crypto, finance, and policy. Email address Subscribe free We respect your privacy. Unsubscribe anytime. The SEC declared the fund effective on June 2, 2026. Trading on Nasdaq began the following day, June 3. Structurally, HYPG is a grantor trust. Investors hold an interest in the underlying assets the trust owns, rather than shares in an operating company that happens to own tokens. The asset at the center of all this is HYPE, the native token of the Hyperliquid network. HYPE has a maximum supply of 1 billion tokens. The token entered the world through a community airdrop in November 2024. The competitive field Grayscale is not alone in this lane. 21Shares and Bitwise both launched their own HYPE products in mid-May 2026, trading as THYP and BHYP, respectively. Both rivals charge 0.30%. Grayscale undercut them by a single basis point. What this means for investors and issuers Spreading a fund’s holdings across more than one custodian can reduce reliance on any single provider, which matters for a product whose entire value sits in one token. Staked tokens are actively used to support a network, so the custodian handling them needs to manage both safekeeping and participation in network operations. HYPG’s staking ratio of roughly 88.66% means nearly nine out of every ten tokens the fund holds are earning rewards. A lower staking ratio would mean more idle tokens, while a higher one could limit how much liquidity the fund keeps on hand for redemptions. HYPE’s price can swing sharply, and a gross reward rate of 2.26% offers little cushion against a meaningful drop in the token’s value. Net returns also depend on network reward rates and how the Hyperliquid ecosystem evolves. Staking rewards are historical figures, not guarantees.

The video-generation spinoff is reportedly targeting an early 2027 listing after raising approximately $2.8 billion at an $18 billion valuation Kuaishou Technology’s AI video unit wants its own ticker. Kling AI plans a Hong …

The video-generation spinoff is reportedly targeting an early 2027 listing after raising approximately $2.8 billion at an $18 billion valuation Kuaishou Technology’s AI video unit wants its own ticker. Kling AI plans a Hong Kong initial public offering that would raise at least $1 billion. Reports from October 6, 2026, indicate the unit has brought on CICC, Goldman Sachs and UBS as underwriters. The target is an early 2027 listing. For a business that only launched in 2024, that is a fast trip from product demo to public markets. The deal on the table The IPO would carve Kling AI out of its parent, Kuaishou Technology, which trades in Hong Kong under the ticker 1024. Kling is also known as Keling AI, or 可灵AI in Chinese. The listing comes on the heels of a large private round. In July 2026, Kling AI raised approximately $2.8 billion, with the round capped at $3 billion. That financing reportedly valued the unit at around $18 billion post-money. Advertisement Tencent, Alibaba and Baidu all participated in the round. Kuaishou’s ownership is expected to shrink as a result. The parent’s stake in Kling is anticipated to be diluted to about 68%. The numbers behind the pitch In the second quarter of 2026, the unit’s revenue topped RMB 850 million. That reflects growth of more than 200% year on year. AI, tech, and the markets they move—in one daily briefing. Daily. Free. Join 34,000+ readers across crypto, finance, and policy. Email address Subscribe free We respect your privacy. Unsubscribe anytime. Kling’s ARR reached roughly $500 million by March 2026. That was up from about $240 million in December 2025. On September 28, 2026, Kling unveiled Kling 4.0. The update supports video clips of up to 30 seconds, adds enhanced controls, and accepts multiple reference inputs. Kling 4.0 is positioned against ByteDance’s Seedance, another Chinese AI video tool. How Kuaishou got here In May 2026, Kuaishou filed a restructuring plan that consolidated Kling’s assets into an entity called Beijing Keling. That restructuring came with investor protections tied to a future listing. If Kling does not go public by October 30, 2031, investors hold repurchase rights. What this means For Kuaishou shareholders, the spinoff is a way to put a visible price on an asset that has been buried inside a larger company. A standalone listing gives the market a direct read on what Kling is worth, since Kuaishou is expected to keep about 68% of the unit. An $18 billion private valuation sits at many multiples of a roughly $500 million revenue run rate. Public market buyers will need to believe the 200%-plus growth pace can hold, at least for a while. Seeking at least $1 billion against an $18 billion valuation means the IPO would float a relatively modest slice of the company. The presence of Tencent, Alibaba and Baidu means Kling’s biggest shareholders include companies with their own AI ambitions, which could complicate strategic decisions down the road.

Crypto exchange OKX launched a standalone app on Tuesday that lets users convert more than 50 local currencies into U.S. dollar-backed stablecoins, as the company tries to expand its offerings beyond just being an …

Crypto exchange OKX launched a standalone app on Tuesday that lets users convert more than 50 local currencies into U.S. dollar-backed stablecoins, as the company tries to expand its offerings beyond just being an exchange. OKX Money, which went live at the c… Crypto exchange OKX launched a standalone app on Tuesday that lets users convert more than 50 local currencies into U.S. dollar-backed stablecoins, as the company tries to expand its offerings beyond…

The prediction market is retiring the Gnosis framework it has relied on since 2019 in favor of a unified, modular, audited system Polymarket has begun rolling out Protocol V2, a rebuild of its prediction …

The prediction market is retiring the Gnosis framework it has relied on since 2019 in favor of a unified, modular, audited system Polymarket has begun rolling out Protocol V2, a rebuild of its prediction market infrastructure designed to replace the collection of smart contracts and adapters that underpin its current system. The new architecture uses a single ERC-1155 positions contract, one collateral asset called pUSD, one exchange supporting multiple market types and a unified router. Polymarket said its existing infrastructure was built around Gnosis’ Conditional Tokens Framework from 2019, forcing the platform to add separate contracts and adapters as it introduced new products. Protocol V2 is intended to consolidate those components and make future upgrades easier. The system initially supports binary markets, Atomic Neg-risk, Incremental Neg-risk and combinatorial markets. Position IDs encode the market type, market and outcome directly, allowing the protocol to identify a position without relying on additional contracts. Advertisement Resolution is also being redesigned through a new OracleAggregator that can connect to multiple oracle providers, including UMA and Chainlink, while leaving room for additional sources in the future. Protocol V2 also introduces native support for eventually bridging positions, collateral and market resolutions across networks. Polymarket said those capabilities are intended to support a future multichain expansion but will not be activated immediately. The news moving money, markets, and the world—before your day starts. Daily. Free. Join 34,000+ readers across crypto, finance, and policy. Email address Subscribe free We respect your privacy. Unsubscribe anytime. Unlike the current contracts, the new system is upgradeable through a governance process, allowing Polymarket to add features without requiring users and integrators to migrate to entirely new contracts each time. The company said the architecture also lays the groundwork for features currently in development, including scalar resolution and directional collateral return. Polymarket plans to run several canary markets using Protocol V2 from Oct. 5 through Oct. 30. The platform is tentatively targeting Nov. 2 to begin routing newly created markets through the new protocol. Alongside the smart contract upgrade, Polymarket is launching Data API V2, a Rust-based service powered by its internal onchain indexer. The new API supports Protocol V2, standardizes response formats and adds cursor-based pagination for trade and activity feeds. Polymarket said Protocol V2 has been audited by Cantina, Certora, Quantstamp, Sigma Prime, Zellic and Pashov, while Certora also formally verified the code. The protocol carries a bug bounty of up to $5 million for critical vulnerabilities.

NYSE’s Parent Just Filed to Tokenize 63 US Stocks on a Crypto-Native Exchange — And the SEC Already Gave It Permission The most conservative institution in US market infrastructure just filed to tokenize 63 …

NYSE’s Parent Just Filed to Tokenize 63 US Stocks on a Crypto-Native Exchange — And the SEC Already Gave It Permission The most conservative institution in US market infrastructure just filed to tokenize 63 of America's most recognizable stocks on a crypto-na… NYSEs Parent Just Filed to Tokenize 63 US Stocks on a Crypto-Native Exchange And the SEC Already Gave It PermissionThe most conservative institution in US market infrastructure just filed to tokenize…

Google will delete all remaining nodes and RPC endpoints on December 15, 2026, and is pointing users to Quicknode as the place to migrate Google Cloud is getting out of the blockchain node business. …

Google will delete all remaining nodes and RPC endpoints on December 15, 2026, and is pointing users to Quicknode as the place to migrate Google Cloud is getting out of the blockchain node business. The company will fully shut down its Blockchain Node Engine and Blockchain RPC services on December 15, 2026. Customers are being pointed to Quicknode, which Google has named its officially recommended RPC infrastructure provider. Teams that miss the date will find their infrastructure deleted. What’s changing and when The wind-down is happening in two stages. The first is already in effect: since June 15, 2026, Google Cloud has disabled new node creation and stopped provisioning new Blockchain RPC endpoints. Between that date and the final cutoff, current nodes and endpoints stay online with limited support. Google says it will keep shipping critical updates, but customers should expect the product to stay where it is rather than gain new features. The second stage is the shutdown itself. On December 15, 2026, every remaining node and endpoint will be deleted. Google’s migration documentation lays out a fairly standard checklist: Advertisement Provision replacement infrastructure Update RPC URLs and application configurations Validate workloads on the new setup Delete legacy Blockchain Node Engine nodes There is one asymmetry worth noting. Google will automatically delete any remaining RPC endpoints at shutdown. Dedicated nodes, however, are on the customer’s list to clean up. Where Quicknode fits in Google is not leaving customers to find a replacement on their own. It has partnered with Quicknode as the recommended destination for teams that need to migrate and provision nodes. AI, tech, and the markets they move—in one daily briefing. Daily. Free. Join 34,000+ readers across crypto, finance, and policy. Email address Subscribe free We respect your privacy. Unsubscribe anytime. There are two ways to make the switch. Customers can provision through the Quicknode listing on Google Cloud Marketplace, which comes with Google Cloud incentives. Or they can sign up directly on Quicknode’s own platform. What Blockchain Node Engine actually did Every blockchain app has to talk to a blockchain. It does that through a node, a computer that keeps a copy of the chain and answers questions about it. Running your own node is tedious. You have to sync the chain, keep up with client upgrades, monitor uptime and pay for storage. Blockchain Node Engine was Google’s managed answer to that headache. The service offered dedicated full nodes and archive nodes. A full node tracks the current state of the chain. An archive node keeps the complete history of past states, which matters for analytics, indexing and some auditing tasks. Support focused mainly on Ethereum Mainnet and its testnets, along with Polygon and Solana. Alongside the dedicated nodes, Google ran a shared Blockchain RPC service, limited to Ethereum Mainnet and the Holesky testnet. What this means for builders and the infrastructure market The immediate impact lands on engineering teams. Anyone running production workloads on Blockchain Node Engine now has a project on the roadmap whether they wanted one or not. Archive node users should pay particular attention. Historical data queries are often the hardest to replicate cleanly, so teams relying on them have extra reason to test early rather than in December. On the competitive side, the arrangement is a clear win for Quicknode. Being named the official recommended provider by a hyperscaler, with a migration path built into Google’s own documentation, is the sort of distribution infrastructure companies usually have to fight for. The date to watch is December 15, 2026. Teams that treat it like a real deadline will be fine. Teams that treat it like a suggestion will learn what deletion means.

The Hangzhou AI lab that spent three years living off its founder's hedge fund is now lining up serious outside money before going public DeepSeek is close to raising at least $12 billion in …

The Hangzhou AI lab that spent three years living off its founder's hedge fund is now lining up serious outside money before going public DeepSeek is close to raising at least $12 billion in fresh funding before it goes public in 2027, according to Bloomberg. The Hangzhou-based AI lab is preparing for a potential listing on Shanghai’s STAR Market. The capital raise would arrive as the company builds out the financial plumbing a public offering requires. The details The money is coming in layers. DeepSeek closed its first external funding round in June 2026, raising approximately $7.4 billion. That round gave the company a post-money valuation of over $50 billion. Founder Liang Wenfeng put in approximately $3 billion of his own assets. DeepSeek is now finalizing talks on another round. It is valued at around 50 billion yuan, roughly $7.4 to $7.5 billion. Advertisement The pricing on that second round is the eye-catcher. It reportedly values DeepSeek at about 500 billion yuan before the new money goes in, or roughly $74 to $75 billion. The revenue story helps explain why. As of September 2026, DeepSeek reported an annualized revenue run rate of $1 billion, more than double its previous figures. AI, tech, and the markets they move—in one daily briefing. Daily. Free. Join 34,000+ readers across crypto, finance, and policy. Email address Subscribe free We respect your privacy. Unsubscribe anytime. Much of that growth came from DeepSeek’s API business, which lets developers plug its models into their own products. Those price increases ranged from 2.3x to 4.5x, and the API unit posted an 82.9% gross margin. From hedge fund side project to IPO candidate DeepSeek was founded in 2023 as a spin-off from High-Flyer, the hedge fund run by Liang. High-Flyer bankrolled the lab on its own for three years, with outside investors only let in during 2026. The company has brought on CITIC Securities as lead underwriter and appointed a new chief financial officer. DeepSeek’s models have gained traction that rivals US labs, despite American export restrictions on semiconductor technology. The company’s response has been to run its models on domestic Huawei hardware. The company has also leaned into open-source releases and research. What this means For investors, DeepSeek is shaping up as a test case for how public markets will price Chinese AI. A $1 billion run rate and an 82.9% API margin give it real numbers to sell. Moving from a post-money valuation above $50 billion to a pre-money figure around $74 to $75 billion is a big step in a short window. Running on Huawei chips insulates DeepSeek from further US export curbs. That same positioning ties DeepSeek’s fortunes to the state of US-China tech relations. The things to watch between now and 2027 are straightforward: does the second round close at the reported valuation, does the $1 billion run rate keep climbing, and does the API business hold its margins once the price increases fully settle in.

is trading around the $85,000 sell wall as leverage cools and new on-chain money arrives, according to Glassnode. That mix could put $96.7K in play. The same report shows profit-taking running well above its …

is trading around the $85,000 sell wall as leverage cools and new on-chain money arrives, according to Glassnode. That mix could put $96.7K in play. The same report shows profit-taking running well above its normal range. Here is what each signal measures and… is trading around the $85,000 sell wall as leverage cools and new on-chain money arrives, according to Glassnode. That mix could put $96.7K in play.The same report shows profit-taking running well ab…

IBIT keeps absorbing institutional demand as US spot Bitcoin ETF flows recover from a rough mid-2026 stretch BlackRock’s Bitcoin ETF clients went shopping again. This time, the receipt came to $69.85 million worth of …

IBIT keeps absorbing institutional demand as US spot Bitcoin ETF flows recover from a rough mid-2026 stretch BlackRock’s Bitcoin ETF clients went shopping again. This time, the receipt came to $69.85 million worth of Bitcoin. How the buying works When investors put money into a spot Bitcoin ETF, the fund doesn’t park that cash. It buys actual Bitcoin to back the new shares. The $69.85 million figure reflects exactly that conversion. Gloria One app, one screen — every morning, Gloria Finance briefs you on what happened and how it affected your portfolio. Read your brief → For scale, this purchase sits below some of IBIT’s bigger days this year. The trust took in $195.6 million on October 1 and logged a $183.41 million inflow in July. The bigger backdrop is a nine-day inflow streak that wrapped up in late September 2026. Over that stretch, US spot Bitcoin ETFs pulled in approximately $3 billion, their strongest run of inflows since October 2025. Advertisement IBIT grabbed a hefty slice of that pie. In one strong week in September, the fund captured approximately $1.2 billion, while rivals including Fidelity’s FBTC and ARK 21Shares’ ARKB also drew money. From a $5.8 billion hole to positive territory During the mid-2026 downturn, cumulative flows were heading toward a potential loss of approximately $5.8 billion. By late September, institutional buying had flipped that tally to modestly positive. The news moving money, markets, and the world—before your day starts. Daily. Free. Join 34,000+ readers across crypto, finance, and policy. Email address Subscribe free We respect your privacy. Unsubscribe anytime. On certain days, IBIT’s inflows actually exceeded the net inflows for the entire category, meaning money was leaving other funds while still arriving at BlackRock’s door. All of this played out while Bitcoin hovered around the mid-$80,000s during trading periods. The fund that ate the category BlackRock launched IBIT in January 2024 after securing SEC approval, among the first US spot Bitcoin ETFs to reach the market. The fund quickly climbed to the top in both assets under management and inflows, offering regulated Bitcoin exposure in a familiar brokerage product, with no private keys to misplace. Throughout 2026, IBIT has frequently accounted for the majority of net category flows, even as Bitcoin’s price turned choppy. What this means for Bitcoin and investors Concentration is one notable feature of this market structure. With so much institutional money funneling through one product, IBIT’s performance and reputation carry outsized influence. On days when IBIT’s inflows exceeded total category net inflows, competitors like Fidelity and ARK 21Shares were experiencing net outflows even as BlackRock continued to attract capital.