Tomasz Tunguz

Student of Startups
Backer of 9 unicorns
Author of http://tomtunguz.com 
Subscribe https://ttunguz.kit.com/ef0a779b9c

Founder

I burned 84 million tokens on February 28th. Researching companies, drafting memos, running agents.

That’s running Kimi K2.5, a serverless model via API. At Claude or OpenAI rates — roughly $9 per million tokens blended — equivalent usage would cost $756 for a single day’s work. My peak days hit 80 million tokens. My average days run 20 million. Cloud inference at frontier-model pricing adds up fast.

This week, Alibaba released Qwen3.5-9B, an open-source model that matches Claude Opus 4.1 from December 2025. It runs locally on 12GB of RAM. Three months ago, this capability required a data center. Now it requires a power outlet.

A $5,000 laptop — a MacBook Pro with enough memory to run Qwen locally — pays for itself after 556 million tokens. At my usage rate, that’s about a month. At 20 million tokens per day, it’s four weeks.

After payback, the marginal cost drops to electricity.

It isn’t an intelligence compromise. Reasoning, coding, agentic workflows, document processing, instruction following : the 9B model matches December’s frontier across the board.

What changes when frontier intelligence runs locally? Everything I send to cloud APIs today — drafting emails, researching companies, writing code, analyzing documents — stays on my machine. No API logs. No third-party retention. No outages. No rate limits.

The tradeoff is parallelization. Cloud APIs handle thousands of concurrent requests. A laptop runs one inference at a time. For simple tasks — summarization, drafting, Q&A — that’s fine.

Queue them up. Let them run overnight. For complex agentic workflows that spawn dozens of parallel threads, local inference may not be worth the wait. The economics favor depth over breadth : fewer tasks, run longer, run cheaper.

Three months from data center to laptop. The buy-vs-rent math just changed.

Last quarter, my AI inference costs hit $100,000 annualized.

I started small. Six months earlier, I was spending $200 a month on Claude. Then I added three agent subscriptions : Codex, Gemini, & Claude Code. I was paying $600 a month.

Next I started using AI to transform my todo list into my done list, increasing tasks to 31 per day. $92 daily inference invoices started arriving. Then $400 per month on browser agents.

Within two quarters, my inference spend grew from $7,200 to $43,000 to over $100,000 run rate.

So I migrated to an open source model. It took a weekend. The key was building the right testing loops : I had six months of historical task data, so I could replay requests through the new model & hill-climb to parity with AI agents working through the night. By Sunday evening, they performed identically. At 12% of the cost.

I’m not the only one paying attention to this cost.

Technology companies are adding a fourth component to engineering compensation : salary, bonus, options, & inference costs. Levels.fyi pegs the 75th percentile software engineer salary at $375k. Add $100k in inference & the fully loaded cost is $475k. That’s 21% in tokens.

The question CFOs will pose : what am I getting for all this inference spend? Can I do it cheaper?

If the metric for a new cloud is gross profit per GPU hour, the employee equivalent is : productive work per dollar of inference.

For me, the answer is 31 tasks a day at $12k annually. The engineer still burning $100k? They’d better be 8x more productive!

Will you be paid in tokens? In 2026, you likely will start to be.

Databricks started later. It built a more complex architecture. It focused on unstructured data; images, documents, logs, audio. Though vast within the enterprise, this data had historically produced little insight. Too hard to process. Too messy to query. Too expensive to store in formats that mattered.

Snowflake took the opposite bet. Structured data. Clean tables. SQL queries that ran fast & returned answers executives could read. The market agreed. Snowflake went public at a $70 billion valuation. Databricks raised private rounds at half that.

Then AI arrived. Suddenly the data that was too messy to query became the data that models needed to train. Unstructured data wasn’t a liability. It was the asset.

Databricks started later. It built a more complex architecture. It focused on unstructured data; images, documents, logs, audio. Though vast within th

Open article: Databricks Overtakes Snowflake by Tomasz Tunguz • 2 min read

Last quarter, my AI inference costs hit $100,000 annualized.

I started small. Six months earlier, I was spending $200 a month on Claude. Then I added three agent subscriptions : Codex, Gemini, & Claude Code. I was paying $600 a month.

Next I started using AI to transform my todo list into my done list, increasing tasks to 31 per day. $92 daily inference invoices started arriving. Then $400 per month on browser agents.

Within two quarters, my inference spend grew from $7,200 to $43,000 to over $100,000 run rate.

Last quarter, my AI inference costs hit $100,000 annualized. I started small. Six months earlier, I was spending $200 a month on Claude. Then I added 

Open article: Will I Be Paid in Tokens? by Tomasz Tunguz • 2 min read

I burned 84 million tokens on February 28th. Researching companies, drafting memos, running agents.

That’s running Kimi K2.5, a serverless model via API. At Claude or OpenAI rates — roughly $9 per million tokens blended — equivalent usage would cost $756 for a single day’s work. My peak days hit 80 million tokens. My average days run 20 million. Cloud inference at frontier-model pricing adds up fast.

I burned 84 million tokens on February 28th. Researching companies, drafting memos, running agents. That’s running Kimi K2.5, a serverless model via A

Open article: Data Center Intelligence at the Price of a Laptop by Tomasz Tunguz • 2 min read

Data Center Intelligence at the Price of a Laptop

For the first time in venture history, three distinct channels share the liquidity burden roughly equally.

A decade ago, secondaries barely registered. They accounted for roughly 3% of exit value in 2015. Today they claim 31% : nearly $95b in the trailing twelve months.

The shift accelerated after 2021’s IPO bonanza. When public markets closed their doors in 2022, investors found alternative routes. Secondaries absorbed demand that would have flowed to traditional exits. When Goldman Sachs acquired Industry Ventures, the transaction signaled secondaries have arrived. Morgan Stanley followed with EquityZen, then Charles Schwab announced its acquisition of Forge Global. Wall Street recognized the structural change before most of venture did.

This matters for founders & investors. When IPOs dominated exits, fund models assumed a small number of public offerings would generate the bulk of returns.

Now liquidity arrives through multiple doors. A founder might sell secondary shares to patient capital while the company remains private. A GP might move positions through continuation vehicles. An LP might trade fund stakes on an increasingly liquid secondary market.

The 830 unicorns holding $3.9t in aggregate post-money valuation cannot all exit through IPOs. The math doesn’t work. At 2025’s pace of 48 VC-backed IPOs, clearing the unicorn backlog would take seventeen years. Secondaries provide a release valve that traditional exits cannot.

Companies like OpenAI have embraced this reality, running employee tender offers while voiding unauthorized secondary transfers. The largest private companies now manage their own liquidity programs rather than waiting for public markets.

Today, secondary liquidity concentrates in the top 20 names. SpaceX, Stripe, OpenAI. For the founder of company #50, the secondary market remains largely theoretical. For secondaries to succeed as a broad asset class, buyers must underwrite positions in companies without household recognition. As the market grows, this coverage gap becomes opportunity.

For LPs starved of distributions since 2022, the expansion of secondary channels offers hope. The $169b in cumulative negative net cash flows needs somewhere to go. More exit paths mean more opportunities to return capital.

When a Series B employee asks about liquidity today, the answer isn’t “wait for the IPO.” It’s “we’re planning a tender offer next year.”

A decade ago, secondaries were a footnote. Now they’re infrastructure. Liquidity flows where it can, not where tradition suggests it should.

[Security vendors please dont ping me. If you're working in Sales, GTM, Marketing - to save your time, please don't ping me with your security/dev-tooling/any product pitches - I'm not decision maker or looking for any tools to buy]

Senior Software engineer with 7+ Years experience (specialization: product security, application security & offensive security focused) well versed in the development, security, and deployment of distributed software applications. expertise in leading & building scalable systems, infra, tooling around LLM & Security.

Open Source contributions: github.com/shivasurya
Secureflow AI Code Reviewer: https://marketplace.visualstudio.com/items?itemName=codepathfinder.secureflow
Building Code-pathfinder, open-source AI native static code analysis (https://codepathfinder.dev) for securing vibe coding, beating traditional static analysis by 10x and reducing false positives. (Not an yet another AI wrapper)

Blog: https://shivasurya.me
HackerOne: https://hackerone.com/libcontainer
Google VRP: https://bughunters.google.com/profile/41d08dfc-2316-41fa-8bb0-885f5401b011

taplio

The Battle

Tomasz Tunguz

Shiva S.

Followers

Posts

Engagements

Likes

Performance ScoreTomasz Tunguz's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Tomasz Tunguz

Shiva S.

Followers

Posts

Engagements

Likes

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Tomasz Tunguz's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreTomasz Tunguz's LinkedIn Performance