{"slug": "your-boss-has-token-regret", "title": "Your boss has token regret", "summary": "SpaceX set a proposed IPO price of $135 per share, valuing the company at $1.7 trillion on a fully diluted basis, according to Renaissance Capital. The space launch and AI company plans to make up to 30% of its IPO shares available to retail investors, with a small post-IPO float of around 4% of total shares.", "body_md": "# Your boss has token regret\n\n## And: SpaceX's first IPO valuation! Muse Spark! Ramp's mega-round!\n\n* Welcome to *.\n\n[Cautious Optimism](https://www.cautiousoptimism.news/), a newsletter on tech, business, and power. Modestly upbeat\n\n**Thursday.** Today, we’re looking at **SpaceX’s **IPO valuation, **Ramp’s **massive fundraise, AI regulation, **Broadcom **earnings, and everyone freaking out about AI spending (which is hilarious). To work! — Alex\n\n*I interviewed*yesterday. Chatted with**Impulse Space** CEO Tom Mueller and**Dusty Robotics** CEO Dr. Tessa Laua few days ago. And I got on the horn with**Cortical Labs** founder Dr. Hon Weng Chong and**Pyka** co-founder and CEO Michael Norciaa few days before that. That’s space logistics, construction robotics, neuron-silicon computers, huge drones, drone deliveries, and GPS-free drone navigation! Have fun!**Manna** founder Bobby Healy and**Theseus** co-founder Ian Laffey\n\n## 📈 Trending Up\n\n[Taking public companies public](https://www.bloomberg.com/news/articles/2026-06-04/honeywell-backed-quantinuum-raises-1-68-billion-in-upsized-ipo)…[Taiwan-USA relations](https://www.reuters.com/world/china/foxconn-announces-strategic-collaboration-with-intel-next-gen-ai-infrastructure-2026-06-04/)…[tech layoffs](https://www.bloomberg.com/news/articles/2026-06-04/us-tech-sector-announces-most-job-cuts-in-nearly-two-years)…[100M users for Qwen](https://www.scmp.com/tech/article/3355850/fried-chicken-flight-plans-alibaba-wants-qwen-become-chinas-digital-fixer),[1,000M users for ChatGPT](https://www.reuters.com/technology/chatgpt-app-hits-1-billion-monthly-active-users-record-time-data-shows-2026-06-02/)…[Benchmark aping a16z](https://www.wsj.com/finance/investing/silicon-valley-stalwart-benchmark-breaks-from-past-embraces-mature-startups-d69e2493)? …[Boston startups](https://x.com/RebeccaTorrenc5/status/2062210478840086850)? …[public distrust of data centers](https://www.ft.com/content/ed07dc6c-aabe-4e4d-a508-4d2b4f24a852)…[South Korean exports](https://www.investing.com/news/economic-indicators/us-firstquarter-worker-productivity-labor-costs-revised-lower-4726825)…[celebrations of slavery](https://www.wsj.com/us-news/confederate-statues-christopher-columbus-removed-fb294c49?mod=hp_lead_pos7)…\n\n**SpaceX’s valuation: **The space launch, AI, connectivity, and social media company set a proposed IPO price rather than a range, [slapping a $135-per-share target](https://www.sec.gov/Archives/edgar/data/1181412/000162828026040364/spaceexplorationtechnologib.htm) on its equity ahead of its listing. That’s a first, but tech companies of note have a history of taking unconventional paths when they go public ([ Google’s reverse Dutch auction](https://www.cnbc.com/2014/08/19/es-took-off-but-the-auction-didnt.html),\n\n**Snap’s**\n\n[no-vote public equity](https://corporatefinancelab.org/2017/05/11/not-all-shareholders-are-created-equal-snap-goes-public-with-non-voting-stock/),\n\n**and**\n\n[Palantir](https://www.cnbc.com/2020/09/29/palantir-direct-listing-reference-price-7point25-says-nyse.html)[with their direct listings, and](https://www.cnbc.com/2021/04/14/coinbase-to-debut-on-nasdaq-in-direct-listing.html)\n\n**Coinbase****Roblox’s**\n\n[epic pivot to a direct listing after filing for a traditional IPO](https://www.investopedia.com/roblox-chooses-direct-listing-over-ipo-madness-5101253)!)\n\n**Renaissance Capital** writes that SpaceX’s valuation at the proposed price is [$1.7 trillion on a fully diluted basis](https://www.renaissancecapital.com/IPO-Center/News/119552/SpaceX-sets-terms-for-record-breaking-$75-billion-IPO), not counting “1.3 billion restricted Class B shares granted to Elon Musk that are not vested.”\n\nThe IPO group hazards that SpaceX’s set price may indicate it “has lined up significant commitments from public investors.” SpaceX also intends to make as many as 30% of its IPO shares available to retail investors; the Falcon 9 maker will have a very small float post-IPO ([around 4% of total shares](https://fortune.com/2026/06/03/spacex-ipo-share-price-index-funds-valuation-public/)).\n\n- At SpaceX’s $26.5 billion annualized Q1 run rate, the company would sport a 64x revenue multiple at $1.7 trillion. Tesla, for reference, has a prices/sales ratio (trailing)\n[of 15x today](https://finance.yahoo.com/quote/TSLA/key-statistics/).\n\nIs SpaceX worth $1.7 trillion? Some folks think so, [and some folks do not](https://www.cautiousoptimism.news/microsoft-i-owe-you-an-apology-i-wasnt-familiar-with-your-game/). The market will decide. I would simply hazard that we should not treat the first few days’ post-debut performance as oracular whispers of future trading patterns. In recent years, post-IPO tech companies have at times posted very strong initial results, only to lose steam in later quarters.\n\n- The SpaceX IPO has several purposes, including returning capital to venture investors and their backers, funds often long locked in illiquid equity. What will happen to venture capital firm fundraising after SpaceX lists will prove\n*very*interesting. How much of the capital gets recycled,\n\n**Ramp:** Once a **Brex** also-ran, **Ramp’s **foray into the world of corporate spend management continues to scale. The startup is [now worth $44 billion](https://www.bloomberg.com/news/articles/2026-06-04/ramp-notches-44-billion-valuation-in-new-funding-round) after raising an additional $750 million in a massive new funding round. In a blog post, Ramp CEO Eric Glyman [argued](https://ramp.com/blog/ramp-at-44-billion-the-third-pillar) that the world is moving away from the traditional two-pillar accounting setup (people and vendors) and towards a new *three-*pillar world, one in which AI spending needs to be accounted for and managed intelligently. Given concern about AI spending (see below), the idea has some merit.\n\n**Muse Spark, finally? **It turns out that Meta wanted to drop an API for its Muse Spark model when it announced the new AI tool. The API wasn’t ready. And it *stayed* not ready, [though the Journal reports](https://www.wsj.com/tech/ai/meta-keeps-delaying-the-release-of-its-new-ai-model-to-developers-f8569c8c) that we’ll be able to take Muse Spark for a spin later this month. Here’s hoping this was a one-time delay as Meta moves from internal AI use to providing access to its models to third parties.\n\n- More competition is more good, so I am rooting that Alexandr Wang and gang get Muse Spark out quickly, learn lots, and then come back with an updated edition. Muse Ignition, perhaps. Then we could get Muse Flame. After that, perhaps Muse Fire? Then Muse Conflagration. And, of course, the model lineup will conclude with the inevitable moniker\n*Muse Fireball*.\n\n[📉](https://finance.yahoo.com/news/servicenow-pledges-1-5bn-investment-110000403.html) Trending Down\n\n[📉](https://finance.yahoo.com/news/servicenow-pledges-1-5bn-investment-110000403.html)\n\n[Human-first places on the Internet](https://www.404media.co/companies-are-using-reddit-to-manipulate-chatgpt-and-google-ai-search/)…[private credit funds](https://www.wsj.com/finance/investing/investors-seek-to-pull-10-from-blackstones-giant-private-credit-fund-7cbca527)(again) …[humans online](https://radar.cloudflare.com/traffic#bot-vs-human)…[Nvidia not competing with its customers](https://x.com/Techmeme/status/2062302312920658083)…[France as the new MENA (in AI terms)?](https://www.ft.com/content/8be3cb2e-c65e-4bef-a085-744e99a9dbd1)…[Idiocracy](https://www.theguardian.com/science/2026/jun/04/vaccine-studies-rfk-jr)…\n\n**Avoiding regulation?** In the wake of the [recent Executive Order](https://www.cautiousoptimism.news/microsoft-i-owe-you-an-apology-i-wasnt-familiar-with-your-game/) establishing a voluntary framework for AI companies to share their upcoming models with the government 30 days before release, OpenAI has a different idea. The company thinks that there’s work to be done.\n\nIn a [policy paper](https://cdn.openai.com/pdf/25752ecb-0e5c-47f9-b9e4-c0f4d76f8d3d/a-blueprint-for-a-federal-framework.pdf), OpenAI argued for stiffer AI regulations — though nothing catastrophic or potentially innovation-slowing, based on a first read. Its plan? Invest in the civilian-led Center for AI Standards and Innovation (CAISI), part of the Department of Commerce and the main touchpoint between industry and government on AI. After it’s sufficiently beefy, OpenAI wants the CAISI given the power to evaluate frontier models before release.\n\nEven under OpenAI’s more robust AI safety framework, the government is more consultant than boss (emphasis added):\n\n[The CAISI evaluations] should assessfrontier capabilities, the effectiveness of associated safeguards and mitigations, andthe resulting risk profile of the deployed system.The requirement should be narrowly targetedat the most capable systems and should allow for iterative deployment by establishing clear thresholds for when subsequent model versions require reevaluation.CAISI’s role should be to conduct evaluations and recommend mitigations—not to approve or block deployments. Developers should remain responsible for deployment decisions, publicly disclose evaluation findings and how they responded, and remain accountable through transparency and reporting requirements.\n\nNo matter, OpenAI’s AI regulatory policy papers are moot with the current Congress, I reckon, so this is more pre-midterm signaling than anything else. *Look, AI companies agree that they need to be regulated! Don’t turn them into public monsters!*\n\nWe’re always on the hunt for regulatory capture. Here, OpenAI makes it clear that it doesn’t support AI regulation that affects all models, only those at the cutting edge. Scare-mongering? I think that Anthropic’s Mythos model made it clear that there will be times when dropping a new AI model the moment it’s done cooking would be *dicey* for the global economy. Perhaps Mythos was a one-off, but would you take that bet? ([Speaking of which](https://x.com/AlecStapp/status/2062519741839741054).)\n\n- Stuff\n[like this](https://www.ft.com/content/86b2440a-60ce-4a5b-94ba-a6a4456ae574)doesn’t help tech’s image, for whatever that is worth.\n\n**Tech companies, after earnings: **In brief, to preserve column inches, Broadcom (networking, storage, wireless tech, etc) [reported 48% year-over-year revenue growth](https://www.cnbc.com/2026/06/03/broadcom-avgo-earnings-report-q2-2026.html), forecast $29.4 billion worth of top line in the current quarter after reporting $22.19 billion in its trailing three-month period, and *still *got whacked by investors. Its sin? Not raising full-year guidance. *Damn.*\n\n- Similarly,\n**CrowdStrike**(cybersecurity) swung to net income in its most recent quarter, revenue rose 26% ($1.39 billion), ARR rose 24% ($5.51 billion), and the company raised its guidance (from $5.87-$5.93 billion to $5.91-$5.96 billion). Result?[A 10% pre-market decline](https://finance.yahoo.com/markets/stocks/articles/crowdstrike-q1-2027-earnings-profit-114408133.html).*Damn*. **Veeva** beat expectations and raised guidance. Result?[Down a few points pre-market trading](https://finance.yahoo.com/markets/stocks/articles/veeva-systems-inc-veev-q1-050028156.html).**NetSkope**, a recent IPO,[beat expectations and got annihilated in post-earnings trading](https://www.ktvb.com/article/syndication/associatedpress/netskope-fiscal-q1-earnings-snapshot/616-776f7111-2b30-4dc1-9848-68ffeeb06191).*Damn*.\n\nThe much-vaunted recovery in tech stocks is perhaps more fragile than we first thought.\n\n## It’s hilarious that AI costs are terrifying business leaders\n\n*We’re an AI-first business now. You must default to AI. Here’s a token-burn leaderboard. We’re tracking employee AI usage to ensure adoption. AI usage is part of your* *OKRs*. *You won’t lose your job to AI, but to someone else who uses AI. We’re firing you to invest in AI.* **All that corporate talk has flipped. The new narrative? *** We’re spending more on tokens than people. We blew through our AI budget in a single quarter. Do you need Opus for that task? You are using too many tokens. We can’t afford to let everyone use as much AI as they want.* *Can’t you do that by hand?*\n\nYes, everyone is worried about AI spending. It’s a real problem, one brought on by the sheer insanity of the *pace* of the AI era. To put the present moment into context, large companies are *still* transitioning to the cloud; SaaS took decades to finally strangle on-prem, and single-sale software. AI? Not only is it evolving more quickly than the technologies we just listed, but companies are adopting it at a pace that crushes prior periods of change.\n\nWith *tokenmaxxing* rapidly evolving into *spendminimizzing*, we face a new period of the AI era. Which is to say that, over the next few months, companies will tackle unsustainable AI spending in a number of ways. These will include:\n\n- Doubling down on AI spend while limiting human labor costs.\n- Lurching towards open-weight/open-source models in an attempt to maintain existing AI practices without needing to change more than the underlying API call.\n- Fine-tuning in-house models in a bid to lower costs.\n- Investing in model-routing technologies to lower near-term AI spend by offloading certain queries to cheaper models.\n- Pulling back on AI more generally.\n\nI think all those are likely except for the final bullet. Companies that expect the *fastest* improvement in AI and operate in the *most* competitive spaces are probably those least willing to limit their AI spending and their use of frontier-level AI models. Companies that expect *sustained*, not explosive, AI improvements and operate in *less-*competitive markets may prefer to bring model routers, open-source models, and the like into their AI setup.\n\nCorporate AI customers will be fine. They went all-in on AI before they understood how to use it or how to incentivize its use within their walls. Their spending woes are annoyances, not existential threats.\n\nWill a pullback in enterprise spending (forward budgeting) harm the major AI labs? Some, of course. If your best customers suddenly find religion and go to rehab, they may buy less of your product. My view is that, much like the **DeepSeek **R1 fear cycle, the ‘AI is too expensive’ moment will also fade. And the AI race will keep hammering along.\n\nWhy? Because once you become accustomed to using AI, it’s hard to go back; because employees have become trained to use AI at all times; because the effective price of intelligence is falling even if frontier-level inference is still expensive; because companies will find easy places to trim fat to invest in muscle; and because [we’re seeing progress in making smaller, local models more powerful](https://blog.google/innovation-and-ai/technology/developers-tools/introducing-gemma-4-12b/) which will move some compute demand over to edge hardware. (Sure, the move may limit accidental Opus abuse, but it will also free up compute for other work, slowing the compute crunch.)\n\nAnd finally, because off-peak models from the major AI labs are really good (shoutout Sonnet 4.6!), the Anthropics of the world already have cheaper models hot and ready to go in the same API that companies used to funnel their spare cash flow to Opus 4.8. It’s simpler to swap to a cheaper model with your existing provider than, say, getting your CSO to approve Kimi K2.6 for internal usage.\n\n- Today, as major AI labs are compute-constrained more than not, they are not competing on price in the same way that they will when compute is unconstrained. Today’s issues with serving inference, period, will flip in time to how best to meet inference demand at the most competitive price point. We just need more compute.", "url": "https://wpnews.pro/news/your-boss-has-token-regret", "canonical_source": "https://www.cautiousoptimism.news/your-boss-has-token-regret/", "published_at": "2026-06-04 14:10:06+00:00", "updated_at": "2026-06-15 11:23:01.088042+00:00", "lang": "en", "topics": ["ai-startups", "ai-infrastructure", "autonomous-vehicles"], "entities": ["SpaceX", "Renaissance Capital", "Elon Musk", "Falcon 9"], "alternates": {"html": "https://wpnews.pro/news/your-boss-has-token-regret", "markdown": "https://wpnews.pro/news/your-boss-has-token-regret.md", "text": "https://wpnews.pro/news/your-boss-has-token-regret.txt", "jsonld": "https://wpnews.pro/news/your-boss-has-token-regret.jsonld"}}