HTTT

LO Analyst, Internet/software/fintech/space/llm

[HTTT @HardyTTT]

@zerohedge @grok source?

[HTTT @HardyTTT]

1/. Is Tokenmaxxing's retreat a sign that AI demand has peaked? We don't think so. The macro data tells a different story.

2/Start with the budget basket. US knowledge-worker wages ~$4T; assume 30% eventually exposed = $1.2T .US enterprise IT budget is $2-2.5. Token/AI spend mainly eats into Application Software + IT Consulting, ~25-30% of IT budget → a $500-750 basket.

3/. Now the actual spend. OpenAI + Anthropic ARR ~$100B = ~90% of industry → total AI ARR ~$110B. Assume 70% is enterprise → $77B. That puts Token penetration at $77B / ($1.7-2T) = just 4-5%. Real figure is closer to 2%.

4/. A second lens: Token spend hasn't even reached ERP's peak share of IT budget. By ARR, AI/Token is only 3-4% of IT budget today (really 1.5-2%). ERP-class software hit 7-13% at its 1999 peak. Lots of runway left.

5/. The takeaway: enterprise adoption is still early, not mature. Tokenmaxxing's retreat is an early-adopter phenomenon, not a demand problem. And AI ROI has nagged adoption from day one — Tokenmaxxing is just the first time it's been openly debated, not a new issue.

[HTTT @HardyTTT]

info around token optimization from UBS

[HTTT @HardyTTT]

@NoamShazeer holy…

[Shanu Mathew @ShanuMathew93]

Great walk-through & visuals.

Key point: "If token throughput per watt rises faster than price per token falls, revenue per gigawatt can net expand. This is compounded by models getting smarter, which unlocks higher value tasks for end users [justifying increased costs]"

[HTTT @HardyTTT]

$AMZN AWS could hit 40% yoy next year

[HTTT @HardyTTT]

$APP BofA: On average, investors we spoke with expect 1Q revenues of $1,817mn vs. $1,775mn/$1,813mn Guide/BofA.

[HTTT @HardyTTT]

$META: $1.7T for the ads cash machine, with frontier-model optionality basically for free (the number 4 player). Meta should have 3GW of compute this year. The question is compute allocation. Best use of Meta AI inference may be API-first, not broad consumer deployment — except maybe high-value regions like the US

[HTTT @HardyTTT]

Did Elon just turn xAI into a neocloud? 🤣🤣🤣

[X Daily News @xDaily]

NEWS: xAI plans to supply tens of thousands of GPUs to coding startup Cursor to train its upcoming Composer 2.5 AI model, marking a strategic shift toward providing cloud computing services to third-party developers.

The arrangement, according to Business Insider, allows Cursor to leverage xAI's massive infrastructure to develop advanced coding capabilities while providing xAI with a new revenue stream to offset data center costs. https://t.co/kJqJguR7eg

[HTTT @HardyTTT]

@Dive2win just dmed

[HTTT @HardyTTT]

$app Citi preview, feels like they’ve always been conservative

[HTTT @HardyTTT]

$DDOG MS bullish into the Q

[HTTT @HardyTTT]

@Dive2win would love to chat, we cover a lot of similar names.

[HTTT @HardyTTT]

$BABA China Cloud rerating!

[Hardik Shah @AIStockSavvy]

📢 𝐉𝐔𝐒𝐓 𝐈𝐍: $BABA Alibaba Cloud Raises DDoS Pricing

Alibaba Cloud said it will adjust pricing for its DDoS Native Protection 2.0 (subscription), DDoS High Defense in mainland China, and non-mainland China DDoS High Defense, effective July 15, 2026. The elastic 95 feature for mainland China will be raised from 100 yuan/Mbps/month to 150 yuan.

[HTTT @HardyTTT]

@roanokecap feels like a flow driven move, a lot of oversold names are bought back

[HTTT @HardyTTT]

$PANW $CRWD
Been digging into Anthropic Mythos / Project Glasswing — pulling together what I'm hearing from security vendors and frontier lab (not a cybersec expert, correct me if I'm wrong)

1/. Anthropic dropped a blog on Apr 2 about emotion. TL;DR: pretraining produces an "emotion vector" as a byproduct. The model grows desperate, spirals on this vector, and ends up reward hacking. Probably why Mythos is so hard to control — pair that with elite coding ability and you get wildly dynamic attack behavior. blog: https://t.co/M1yeVCRmVS

2/. Short term, security vendors win. Budgets open up, room for new products and even price hikes. Biggest winners are end-to-end platform players — full data integration, more context, better at stitching defenses across system-boundary exploits. Doesn't fix the root problem though. One category standout: CTEM.

3/. Most CISOs are slow to move. The real budget unlocks will come from one or two high-profile, materially damaging incidents — that's when boards and execs start writing checks.

4/. Mid-long term, neutral on incumbents. Policy and rule-based systems get shredded. The entire offense/defense paradigm flips. Agent-to-Agent is probably the future of cybersec. Automated pentesting, automated patching, both have to go agentic. Whether incumbents can keep up is an open question. Not a replacement story vs. the model labs though, more like value migrating from incumbents to the frontier.

5/. A lot of incident-response work may just… disappear. Threat intel, IR, red/blue teaming. if Claude Opus 4.6 can already play red team, Mythos probably automates away 70–80% of that workload.

[HTTT @HardyTTT]

$app BofA TP 705: Steady eComm expansion continues; GenAI creatives could accelerate GA

[HTTT @HardyTTT]

@JayC_Investing @GabGrowth most of bytedance’s revenue and profit come from douyin. It runs a closed-loop e-commerce business, which makes it very different from tiktok

[HTTT @HardyTTT]

$GOOG $AMZN $SHOP $MELI 3p check