tetsugan sakamoto

MD by background.
Japanese | Head of investment at @cryptotimes_mag

[tetsugan sakamoto @CryptoTetsugan]

今年もよろしくお願いします！

[CRYPTO TIMES@仮想通貨メディア @CryptoTimes_mag]

【仮想通貨の今後はどうなる？業界有識者40名が予測する2026年の注目分野 - 後編】
-
記事はこちら👇️
https://t.co/XsWxIME9ih
=====
毎年恒例の特集記事の「後編」が公開されました。後編でも業界で活躍するVC、リサーチャー、KOL等の分野から20名以上の注目コメントを掲載しています。

[tetsugan sakamoto @CryptoTetsugan]

@sjdedic Congrats!!

[Wu Blockchain @WuBlockchain]

Citrini: Hyperliquid Accounts for Nearly Half of Crypto Token Buybacks in 2025

Citrini Research said Hyperliquid is a “compelling” investment, noting that more than 90% of platform fees are directed to its Assistance Fund and used to repurchase HYPE on the open market. According to the report, Hyperliquid's buybacks have accounted for nearly half of all crypto token buyback activity in 2025 by some measures. Citrini added that the protocol still has significant room to gain market share.

https://t.co/nhts7s5zeD

[John Scott-Railton @jsrailton]

NEW: malware developers added nuclear & biological weapons text to to their spyware.

Goal? To trigger LLM safety refusals... so that their spyware wouldn't be analyzed by an AI security scanner.

Cleanest practical example I can think of for why over-indexing on first order safety alignment is risky.

When closed (and open) models ship with aggressive refusals, they will be sprinkled with second-order blindspots that attackers will discover...and exploit.

We are only in the earliest days of attackers leveraging these features, and it wouldn't surprise me if users systems that need to handle complex cybersecurity issues demand that models be less safety-blunted.

In the weeds: @SocketSecurity's post also shows why intention matters in how you design a malware analysis pipeline to avoid prompt manipulation.

H/T to colleagues that shared this with me https://t.co/f3Aj9TYxU4

[Jukan @jukan05]

Nvidia's AI ramp deepens memory squeeze as cloud providers lock up supply through 2028

Memory shortages tied to Nvidia's next wave of AI hardware are expected to intensify through 2027 and into 2028, as major cloud service providers continue to secure long-term supply for data-center buildouts, according to supply-chain sources. The pressure is already rippling through DRAM and NAND markets, with OEMs and module makers warning of tighter availability and weaker room for additional orders.

Cloud providers and OEMs race to secure capacity
As Nvidia's next-generation AI accelerator, Vera Rubin, is set to ship in the second half of 2026 and HBM4 moves into mass production, cloud service providers, or CSPs, are continuing to invest heavily in AI data-center expansion. Supply-chain sources said CSP leaders have already booked all available 2027 long-term agreement, or LTA, capacity and are now moving to secure 2028 supply as well.

The AI boom has driven a wave of memory shortages, and several memory module makers have recently received notices from original equipment manufacturers stating they will not be able to provide any additional supply beyond previously promised volumes. Upstream capacity has been heavily diverted toward Nvidia-related AI servers and toward CSP demand, leaving almost all of 2027's capacity already committed to CSP customers.

Supply-chain sources said shortages in DRAM and NAND are expected to widen from the second half of 2026. OEMs have also disclosed that Apple has continued to secure third-quarter memory capacity for new product launches, while other brands are being forced to move ahead with production plans despite the strain.

That means memory makers must support Vera Rubin mass shipments while also meeting rising demand from other end devices and edge AI. This combination is expected to tighten global supply further. Industry consensus is increasingly that the 2027 shortage will be worse than the 2026 one.

2028 supply talks begin early
After locking up 2027 capacity, CSP leaders are also moving ahead of schedule to secure 2028 LTA supply. Upstream suppliers expect memory prices to remain strong. While they were unwilling to release 2028 capacity before May 2026, they have recently become open to discussing first-quarter 2028 orders, with some HBM and server capacity already allocated.

Industry sources said LTA structures vary by supplier. Only a few require advance deposits or prepayments; most rely on customers first committing to expected volumes, after which suppliers adjust expansion timelines accordingly. Final selling prices are still confirmed only before actual shipments, suggesting memory makers remain confident in contract price gains over the next two to three years.

Although CSP companies were once rumored to be willing to fund dedicated production lines at memory makers and subsidize purchases of expensive semiconductor manufacturing equipment, industry insiders said upstream suppliers are already highly profitable and have already planned new fabs and equipment budgets. They do not need to tie themselves to a single customer to build production lines, but CSPs are still rushing to secure long-term memory contracts to avoid a repeat of the supply crunch in the second half of 2025 to 2026.

Tight DRAM supply drives capacity hoarding and AI server cost pressures
Sources said memory module makers are also using prepayment models for certain product lines, paying suppliers in advance at previously agreed lower prices and then covering the gap when the original maker is ready to ship at market rates. While this increases short-term funding pressure, it helps them lock in critical capacity early.

Industry analysts said server memory and standard memory share similar design architectures, unlike embedded memory, so capacity allocation at suppliers is mainly crowding out standard and PC memory, which together account for roughly 60% to 70% of the entire DRAM market. That shift has pushed PC memory's share of total DRAM down from 11% to 12% in the past to just 9%.

Nvidia CEO Jensen Huang recently confirmed that SK Hynix, Samsung Electronics, and Micron have all received HBM4 certification, have begun production, and are fully supporting Vera Rubin. Addressing rumors that memory usage would be cut sharply, Huang said future systems will still use a great deal of high-speed memory. Still, the shortage must be managed rationally across all systems, and supply must be expanded further.

Industry sources also said AI servers are seeing downgrades in high-capacity memory configurations. While 128GB was originally the mainstream specification, some customers are switching to 64GB or 96GB to cut costs amid high prices and tight supply. Although that does not significantly reduce overall memory usage, it highlights how elevated memory costs are slowing the AI server upgrade cycle.

[TechCrunch @TechCrunch]

xAI fired an engineer who raised alarms about Grok safety, new lawsuit claims https://t.co/gRoLgg3mFp

[金のニワトリ @gosrum]

DiffusionGemma、GGUFが出てるということはllama.cppで動かせるということかな？

精度と推論速度がどう変化するのかが気になる

[Cookie3 🍪 @cookie3]

The CookiePro waitlist just passed 3,000.

Thank you to everyone who signed up, and everyone still waiting.

You're the reason we keep building.

Launch is close. If you read X for a living, there's still time to get in early: https://t.co/IS2XnIi06k

[tetsugan sakamoto @CryptoTetsugan]

Yup this is the way. Only limited people will touch frontier and majority will use outdated model or OSS models from China.

[Ryan Connor | RockawayX 🇺🇸 @ryanconnor]

"we hence see growing signs of a bifurcation in frontier vs 'everyday' AI usage"

Guys, we've been model routing to reduce costs since GPT-4...

GPT-5 is was an improved model router. Your Google search routes based on complexity. Claude answers may use frontiers but suggested follow up questions are a mini-model...

This has always been the standard.

[Milk Road @MilkRoad]

Mike Novogratz: No one cares about anything outside the AI food chain right now.

"Crypto is in that... we tried to make the crypto agentic link. It's actually a real link. It's not a theoretical link..."

"But no one cares about crypto right now."

"Markets need energy."

"And then you've got Saylor selling a little bit. I don't know why he did that, it was a strange move..."

"It was a tiny amount - but if the guy with laser eyes is selling $BTC, that's not good either."

FT @Scaramucci @Novogratz @AllThingsMkts @SkyBridge @SALTConference.

[Wes Roth @WesRoth]

Figure’s BotQ humanoid robot factory hit a record production month in May.

[Polymarket @Polymarket]

JUST IN: GPT-5.6 projected to be released next week, as OpenAI looks to answer Anthropic’s Claude Fable launch.

68% chance. https://t.co/mLQVodbBlk

[Milk Road @MilkRoad]

6th Man Ventures: "There's a new age of assets that have value, that are generating revenue, and then there's the rest of crypto."

$HYPE is up 300-400% vs Bitcoin. Equities are ripping like memecoins. AI is pulling liquidity away.

"I don't see a near-term catalyst for a reversal." Bitcoin has quantum as an overhang. Ethereum is struggling. It's asset-picking season now.

FT @CarlKVogel @LGDoucet @6thManVentures.

[tetsugan sakamoto @CryptoTetsugan]

Free your mind

[Andrej Karpathy @karpathy]

This is a super exciting release - Claude Fable 5 is the same underlying model as Mythos but with added safeguards. The benchmarks are great and it's SOTA on everything by a margin but I'll add that *qualitatively* also, this is a major-version-bump-deserving step change forward (imo of the same order as Claude 4.5 was in November), peaking especially for long problem-solving sessions on very difficult problems. You can give it a lot more ambitious tasks than what you're used to, the model "gets it" and it will just go, and it's never felt this tempting to stop looking at the code at all (but don't do this in prod!). The model still has quirks that people will run into and the safeguards are configured to be a little too trigger happy for launch, which can hopefully be tuned over time.

I feel a lot of things changing as working software increasingly comes out on a tap. The Jevon's paradox kicks in and I feel my own demand for software growing substantially. You can ask for anything - explainers, visualizers, dashboards, bespoke single-use apps (e.g. a full wandb that is hyper-specific just for your project), you can 10X your test suite, auto-optimize code, run giant research projects with custom HTML for the results, anything! "Free your mind" (Matrix ref). Really looking forward to all the things people build!

[ありゃりゃ @aryarya]

ブラックストーン、個人投資家の資金流入が6月に改善＝社長
https://t.co/qyXMy76CLd

[Pruv Finance @PruvFinance]

Tokenization is gaining momentum worldwide.

This week:

🇺🇸 America's largest banks, including JPMorgan, Citi, Bank of America, and Wells Fargo, are reportedly planning a tokenized deposit network targeted for launch by H1 2027.
https://t.co/0vZnajCh2e

📈 Citi's latest Tokenization 2030 report projects tokenized assets growing from $17B today to $5.5T by 2030, with stablecoins reaching $1.9T.
https://t.co/jmToixidNA

🇭🇺 Hungary's new Science and Technology Minister signaled a more crypto-friendly approach, criticizing previous restrictions on digital asset services and indicating potential regulatory changes ahead.
https://t.co/ybEhcyp0mu

🇻🇳 Vietnam continues advancing its digital asset framework, with regulators recognizing crypto assets and tokenized assets as an increasingly important part of the digital economy.
https://t.co/3yrx7Vog7N

📊 Ondo is bringing more utility to tokenized assets. Tokenized stocks will soon be usable as collateral on Ondo Perps.
https://t.co/HgZwb9f0Io

A clear pattern is emerging:
• Governments are building regulatory frameworks.
• Institutions are building infrastructure.
• Onchain Markets are building utility.

[Pruv Finance @PruvFinance]

In one month, 3 institutional-grade premium funds with over $3.3M combined value were tokenized through Pruv.

• Yield8 Private Credit Fund
• Garuda Sports Fund
• SateYield Money Market Fund

Tokenization is inevitable.

[kokko | 古賀興司 @kokko_coco]

今日の米国株の異常な上下動の背後にある要因として無視できないのは、やはりレバETF。AUMが膨らんだレバETFのリバランスフローと、そのフロントランが、指数や主力株の上下動を強烈に増幅している構図。今日も前半の下落で、推計リバランスフローは一時$-30Bn前後まで拡大。金曜の二の舞かと思われた。しかし、この手のフローの怖さは、下方向だけではない。少し巻き戻しが入ると、その巻き戻し自体も増幅される。NY時間の後半になればなるほど、リバランスを執行できる時間は少なくなる。つまり、巻き戻しが遅い時間に始まるほど、フローは一気に逆回転しやすい。昨年11月のNVDA決算翌日もそうだった。決算を受けて高く寄り付き、NY午前中にリバランスフローのフロントランが十分入った後、オプション市場主導で巻き戻しが始まった。そこからレバETFのフローが逆回転し、巨大な陰線を作ったあの木曜日。今日はその逆。下落が少し巻き戻されると、今度はリバランスフローが上方向に追いかける。SPXは一時-2%超まで売られていたにもかかわらず、結局は小幅安圏まで戻す。個人投資家としての対処はシンプル。触らぬ神に祟りなし。こういう構造的乱高下に付き合う必要はない。今は本物のガンマもネガティブに落ちているため、なおタチが悪い。金融市場ウォッチャーとしてどう思うか。レバETFは、そのAUM拡大に伴い、もはや単なる投資商品ではなくなっている。市場全体にダメージを与えかねないVaRショック供給源になりつつある。禁止を含めた規制強化を、特に個別株レバETFについては、真剣に議論すべき段階。

[ClaudeDevs @ClaudeDevs]

We're hosting Claude Fable 5 Build Day in San Francisco on June 13.

Point Fable 5 at a problem worth solving and build a solution with Claude Code.

The Anthropic team will be in the room, with a chance to win from a prize pool of $150K in Claude credits across 3 finalists.

[ぬこぬこ / NUKO 🇯🇵 @nukonuko]

Anthropic の中の人による Fable 5 を用いた自律ループの作り方の記事

Fable 5 の性能を最大限発揮するには、単にプロンプトを入力するのではなく、ゴールやアウトカムなどの環境フィードバックや自動的なメモリ管理を活用した自律ループを設計すると良い

[Vitto Rivabella @VittoStack]

https://t.co/KjnTyZK5Cn