iiinova — where your digital self keeps evolving

ttunguz.bsky.social @ttunguz.bsky.social · 04/05/2026

What about heavier workloads? Agentic coding burns 10 to 20x more tokens than passive chat. At that rate, ad-only models can’t keep up. But a hybrid works : $10/month plus 8 ads per day covers 2 million tokens. It won’t fund a tokenmaxxing habit, but it will keep you shipping.

140

ttunguz.bsky.social @ttunguz.bsky.social · 04/05/2026

But there is the question of utilization. All of these figures assume the cluster stays busy. Idle GPUs raise the per-user cost.

130

ttunguz.bsky.social @ttunguz.bsky.social · 04/05/2026

One content ad every 90 seconds still covers the cluster, comparable to what mobile users already tolerate. On the other end of the spectrum, rewarded video clears at $40 to $50 CPM with near-100% fill in gaming. A single round of rewarded video across the cluster nearly covers an hour of compute.

110

ttunguz.bsky.social @ttunguz.bsky.social · 04/05/2026

There are nuances. Assuming ad fill rates (the share of ad requests that return a paid ad) and ad network revenue share, we can model effective CPMs of $1.50. At that floor, the frequency doubles.

100

ttunguz.bsky.social @ttunguz.bsky.social · 04/05/2026

To support these costs, users would see one ad every 3 to 39 minutes. That is well below what users already tolerate : hyper-casual mobile games show six ads per session, roughly one per minute.

100

ttunguz.bsky.social @ttunguz.bsky.social · 04/05/2026

These numbers assume the provider runs 4 B200 Blackwells at 300 users, 50% of the theoretical maximum, leaving headroom for bursts.

100

ttunguz.bsky.social @ttunguz.bsky.social · 04/05/2026

A B200 GPU costs $4.50/hour on spot markets. Google Search ads generate $38.40 CPM (cost per thousand impressions), while Google Display runs $3.12.

100

ttunguz.bsky.social @ttunguz.bsky.social · 04/05/2026

When Anthropic pulled Claude Code from the $20 plan last month, it signaled an industry assumption : frontier intelligence requires frontier pricing. For open models, the economics flip.

100

ttunguz.bsky.social @ttunguz.bsky.social · 04/05/2026

One search ad every 40 minutes pays for a trillion-parameter model. One content ad every 3 minutes does the same. The math on ad-supported AI is better than you think.

100

ttunguz.bsky.social @ttunguz.bsky.social · 01/05/2026

Clearly, there are many risks associated with AI, & it’s important to counterbalance them with some of the tremendous advances that are happening every week. tomtunguz.com/what-ai-did-...

tomtunguz.com

This Week's Sign the Apocalypse Isn't Upon Us

From early cancer detection to teacher training and exoplanet discovery, AI is quietly fixing real problems. Here are the good things it accomplished in just two weeks.

000

ttunguz.bsky.social @ttunguz.bsky.social · 01/05/2026

- Twilio posted Q1 2026 revenue of $1.41 billion (up 20%), its highest revenue and gross profit growth rates in more than three years. The company raised its full year forecast.

100

ttunguz.bsky.social @ttunguz.bsky.social · 01/05/2026

Reversing the SaaSpocalypse - Atlassian reported Q3 FY26 revenue of $1.79 billion (up 32% year over year), with cloud revenue surging 29% to $1.13 billion. The company raised its full year growth outlook to ~24% from 22%, and shares soared more than 25%.

100

ttunguz.bsky.social @ttunguz.bsky.social · 01/05/2026

In disaster response - Hong Kong launched an AI flood forecasting system with 3D simulations, enabling faster preemptive emergency responses.

100

ttunguz.bsky.social @ttunguz.bsky.social · 01/05/2026

In the stars - Princeton researchers used AI to scan 83 million stars from NASA’s TESS telescope data and identified 10,091 new exoplanet candidates (11,554 total), sorting in seconds what would take astronomers years by hand.

100

ttunguz.bsky.social @ttunguz.bsky.social · 01/05/2026

On land - Texas A&M demonstrated AI predicting outbreaks of destructive crop pests with ~88% accuracy in open fields.

100

ttunguz.bsky.social @ttunguz.bsky.social · 01/05/2026

In classrooms - A 2025 Harvard physics study found students using AI tutors learned more than twice as much in less time compared with traditional active learning classrooms. - Thailand upskilled 160,000 teachers in AI, impacting 3.3 million students.

100

ttunguz.bsky.social @ttunguz.bsky.social · 01/05/2026

In medicine - Mayo Clinic validated an AI that detects pancreatic cancer up to 3 years before diagnosis on routine CT scans. - J&J reported AI is halving the time to generate new drug development leads and cut clinical trial report preparation from 700 hours to about 15 minutes.

100

ttunguz.bsky.social @ttunguz.bsky.social · 01/05/2026

With all the dire predictions about AI, it’s important to also spend time recognizing the tremendous pace of innovation & the impact AI is having broadly. Here’s what it fixed, taught, and discovered in the last two weeks.

100

ttunguz.bsky.social @ttunguz.bsky.social · 01/05/2026

I remember growing up reading Sports Illustrated. There was a small column called “This Week’s Sign the Apocalypse Is Upon Us.”

100

ttunguz.bsky.social @ttunguz.bsky.social · 30/04/2026

OpenAI committed to consume approximately 2 gigawatts of Trainium capacity through AWS starting in 2027. Anthropic secured up to 5 gigawatts. But Amazon doesn’t own the model layer. Google does. The hyperscaler that owns the model layer is growing the fastest. tomtunguz.com/2026-04-29-t...

tomtunguz.com

The $112 Billion Quarter

Google Cloud shocked the street with 63% growth in Q1 2026. Amazon, Google & Microsoft reported earnings with combined quarterly capex hitting $112B while cloud growth diverged dramatically.

000

ttunguz.bsky.social @ttunguz.bsky.social · 30/04/2026

But debt isn’t the only way to catch up. Amazon is betting on vertical integration. It landed 2.1 million AI chips over the past twelve months & its chips business has crossed a $20 billion annual revenue run rate, growing triple-digit percentages year-over-year.

100

ttunguz.bsky.social @ttunguz.bsky.social · 30/04/2026

Microsoft, by contrast, is funding its buildout from operating cash flow. Google & Amazon are levering up to close a gap. Microsoft is already ahead.

100

ttunguz.bsky.social @ttunguz.bsky.social · 30/04/2026

Bank of America forecasts hyperscaler debt issuance will hit $175 billion in 2026, more than six times the $28 billion annual average of the prior five years.

100

ttunguz.bsky.social @ttunguz.bsky.social · 30/04/2026

How they’re financing the gap is revealing. Alphabet sold a rare 100-year “century bond,” the first by a tech company since Motorola in 1997, as part of a $32 billion debt offering. Amazon raised roughly $54 billion in March.

100

ttunguz.bsky.social @ttunguz.bsky.social · 30/04/2026

Amazon’s free cash flow collapsed to $1.2 billion as a $59.3 billion year-over-year surge in infrastructure spending consumed nearly all of its $148.5 billion in operating cash flow. Google still generated $64.4 billion in TTM free cash flow. Microsoft produced roughly $15 billion quarterly.

100

ttunguz.bsky.social @ttunguz.bsky.social · 30/04/2026

Google is now outspending Microsoft on capex, despite running a cloud business about 37% the size. That gap will widen. Google raised full-year 2026 capex guidance to $180-190 billion, while Microsoft is tracking toward roughly $120 billion. The smaller player is spending more to catch up.

100

ttunguz.bsky.social @ttunguz.bsky.social · 30/04/2026

All three hyperscalers reported extraordinary capital expenditure in Q1, a combined $112 billion in quarterly infrastructure spending.

100

ttunguz.bsky.social @ttunguz.bsky.social · 30/04/2026

This is consistent with what enterprises like Uber & BlackRock have disclosed : internal AI budgets are eclipsing initial estimates because usage grows exponentially once models are deployed in production.

100

ttunguz.bsky.social @ttunguz.bsky.social · 30/04/2026

Even at the stated minimums, those 330 customers alone represent a floor of roughly $1.6 billion in annual token consumption. And they are growing into their commitments faster than planned : “Customers outpaced their initial commitments by 45%, accelerating over last quarter.”

100

ttunguz.bsky.social @ttunguz.bsky.social · 30/04/2026

The customer scale is staggering : “330 Google Cloud customers each processed over 1 trillion tokens. 35 reached the 10 trillion token milestone.”

100

ttunguz.bsky.social @ttunguz.bsky.social · 30/04/2026

With vertical integration, it is driving down the marginal cost per token : “TPU 8i delivers cost-effective, low-latency inference with 80% better performance per dollar than the prior generation.”

100

ttunguz.bsky.social @ttunguz.bsky.social · 30/04/2026

These are committed contracts that cannot be fulfilled until new capacity comes online in late 2026 & 2027. Gemini is now processing 16 billion tokens per minute via direct API use by customers, up 60% from last quarter. Google is not just scaling volume.

210

ttunguz.bsky.social @ttunguz.bsky.social · 30/04/2026

Pichai disclosed the scale of enterprise deal flow : “We are seeing strong deal momentum, doubling the number of $100 million-$1 billion deals year-on-year & signing multiple $1 billion-plus deals.”

100

ttunguz.bsky.social @ttunguz.bsky.social · 30/04/2026

Google Cloud’s backlog nearly doubled quarter-over-quarter to over $460 billion, more than twice its trailing-twelve-month cloud revenue. (By comparison, Microsoft’s commercial RPO of $627 billion includes Office 365, Dynamics & LinkedIn, not just Azure.)

100

ttunguz.bsky.social @ttunguz.bsky.social · 30/04/2026

Google could not build data centers fast enough to satisfy the AI workloads its customers wanted to run. Pichai confirmed it on the call : “We are compute constrained in the near term. Our cloud revenue would have been higher if we were able to meet the demand.”

100

ttunguz.bsky.social @ttunguz.bsky.social · 30/04/2026

Sundar Pichai gave the clearest explanation on the earnings call : “Our enterprise AI solutions have become our primary growth driver for cloud for the first time in Q1.”

100

ttunguz.bsky.social @ttunguz.bsky.social · 30/04/2026

Whether that explains the full gap is unclear, but the structural advantage is not : Google owns Gemini & TPUs top to bottom, with no licensing fees to OpenAI or Anthropic. Its growth may be more profitable too.

100

ttunguz.bsky.social @ttunguz.bsky.social · 30/04/2026

The divergence is striking. AWS & Azure resell compute. Google bundles compute with its own models.

100

ttunguz.bsky.social @ttunguz.bsky.social · 30/04/2026

Google Cloud grew 63% year-over-year in Q1 2026. Amazon Web Services posted 28%. Microsoft Azure hit 40%. All three are exceptional. Only one hit 63%.

100

ttunguz.bsky.social @ttunguz.bsky.social · 29/04/2026

The database market produced Oracle, MongoDB, Databricks, & Snowflake. A $100B inference market fragmenting the same way creates room for similar winners. tomtunguz.com/inference-ma...

tomtunguz.com

Darwinian Specialization in AI

The $100B AI inference market is fragmenting into specialized workload types, just as databases evolved from one category into dozens. Here's the map.

000

ttunguz.bsky.social @ttunguz.bsky.social · 29/04/2026

Tesla runs vision models on FSD chips drawing 72 watts. Quantized models, specialized chips, & limited memory create different optimization challenges than cloud inference.

100

ttunguz.bsky.social @ttunguz.bsky.social · 29/04/2026

Edge (On-Device & On-Premise) Privacy requirements, connectivity constraints, & latency sensitivity push inference to edge devices. Mobile phones, industrial sensors, medical devices. Apple runs a 3-billion-parameter model on-device for Apple Intelligence...

100

ttunguz.bsky.social @ttunguz.bsky.social · 29/04/2026

A single image requires 50 sequential passes through the model. Different architectures, different constraints, different infrastructure.

100

ttunguz.bsky.social @ttunguz.bsky.social · 29/04/2026

Multimodal (Image, Video, Audio) The bottleneck shifts. For chatbots, the problem is memory. The model holds the entire conversation in its head, & that memory grows with every turn. For image & video generation, the problem is raw compute...

100

ttunguz.bsky.social @ttunguz.bsky.social · 29/04/2026

Batch (seconds to hours) handles document processing & content generation at scale. Cost efficiency matters more than speed, so workloads run during off-peak hours on spot instances.

100

ttunguz.bsky.social @ttunguz.bsky.social · 29/04/2026

Near-real-time (100ms-2s) covers chatbots, code completion, & search augmentation. Most LLM applications today operate here, where batching & queuing optimize throughput without degrading experience.

100

ttunguz.bsky.social @ttunguz.bsky.social · 29/04/2026

Latency Tiers : Real-Time, Near-Real-Time, & Batch Latency defines three distinct segments. Real-time (sub-100ms) serves voice assistants, live translation, & autonomous vehicles. Users won’t wait, so infrastructure must be geographically distributed with dedicated capacity.

100

ttunguz.bsky.social @ttunguz.bsky.social · 29/04/2026

The model ecosystem reflects this. A few dominant LLMs with long half-lives sit alongside 90,000+ image generation models on Hugging Face, with new variants appearing daily. Each model type has different serving requirements, which fragments the infrastructure. Today, we see these segments :

100

ttunguz.bsky.social @ttunguz.bsky.social · 29/04/2026

The inference market is fragmenting for the same reason : workloads are different. Images & video are compute-heavy. Longer context windows demand more memory for KV cache. Edge devices have power constraints. A single architecture can’t optimize for all of them.

100