Summary

Blocking exits does not discipline Big Tech; it starves startups, reduces capital available to them, and ultimately strengthens incumbents. Balaji traces the squeeze from Sarbanes-Oxley—public companies and IPOs declined, forcing startups to stay private—through four years of blocked M&A, citing DOJ interference with JetBlue’s acquisition of Spirit, after which Spirit went bust, and Roomba’s difficulties. His investable mechanism: acquisitions are incumbent “surrenders” whose proceeds fund challengers, so “the actual way of regulating big companies is with a thousand startup piranhas.”
Corporate M&A is a power-law venture portfolio, not a series of retrospectively obvious monopoly grabs. Steven says acquisitions are provably net destroyers of value and that big companies make the wrong strategic bet roughly 90% of the time; yet successful outliers get retconned as inevitable. Instagram had no revenue, had just raised at a $500 million valuation, cost Facebook $1 billion—about 25% of its cash—weeks before its IPO, without prior board consultation; “everybody wants a piece of the reward” while accepting none of the original risk.
Antitrust pressure has created an “acqui-fire”: selected talent moves and money remains in the left-behind entity, but the company itself is not acquired. Balaji groups Scale, Character, Inflection, Adept, Covariant and Windsurf into variations on this structure, which can execute faster than conventional M&A. In his account, Google took roughly 40 Windsurf people while leaving about 200 employees and more than $100 million in the company; the missing consideration was status, prompting the remainder to seek a second transaction with Cognition.
AI is a platform shift in which strategically irrational spending on tools and exceptional people can still be economically rational. Steven compares today’s coding-tool proliferation with DOS, BASIC and the overinvestment that consolidated PC operating systems: tooling may never be the largest standalone business, but every aspiring platform “has to have it.” Balaji adds that AI amplifies the best researchers and engineers, making selective talent deals more attractive even when integrating whole companies would not be.
The speakers see a fundamental mismatch between dynamic software markets and regulation built for railroads, coal and geographically constrained distribution. HHI-style market definitions break when an iPhone is simultaneously a phone, camera and programmable internet device, or when Apple, Google and Facebook compete across overlapping product sets. Balaji’s “network versus state” framing adds the political mechanism: Uber, YouTube and other platforms became practical regulators, while governments remain monopoly platforms whose users lack comparable exit.
US leadership in AI could be squeezed simultaneously by copyright litigation, energy constraints and restrictions on Chinese models. Balaji rejects “this is the worst it’ll ever be,” pointing to Napster and Google Books as products degraded by legal attack; Kimi, Qwen and DeepSeek, he says, are already good open-weight models even if not fully open source. Erik frames China’s strategy, in Christensen’s terms, as commoditizing American strength. Steven compares it to Google releasing Google Docs for free, while Balaji pushes back that copyright and intellectual property also helped create the technology industry.
Their policy prescriptions converge on markets but differ in scale: Steven favors predictable rules and letting transactions fail, while Balaji wants organized jurisdictional competition. Steven calls predictive merger blocking statistically indefensible and compares it to “rent control on investing.” Balaji proposes model legislation for all 50 states and 190 sovereign countries, backed by 10 CEOs, founders or investors—or a broader group representing stated revenue or AUM—promising capital where it passes: build at “the speed of physics, not permits.”

Deep dive

1. Closing exit markets entrenches the incumbents regulators target

Balaji’s causal chain begins with Sarbanes-Oxley: legislation intended to prevent another Enron instead reduced public companies and IPOs, pushing technology businesses to remain private longer and depend on private-equity-style financing. Aggressive FTC and DOJ merger enforcement then constricted the second exit route. “For four years, it was just a desert.”
His sharpest examples are JetBlue–Spirit—after DOJ intervention, “Spirit went bust”—and Roomba, which “had issues,” alongside “a bunch of companies” that silently died. The state first made IPOs harder, then M&A harder, while also pursuing crypto, AI-compute and debanking actions Balaji characterizes as an “all-out anti-tech assault.”
Figma’s eventual IPO does not vindicate the blocked Adobe sale in Balaji’s telling: Figma succeeded “absolutely no thanks to” the intervention. Lina Khan taking credit was, he says, “like the assassin congratulating themselves for helping to elect Trump”—a specimen of Washington claiming visible successes while ignoring the longer, less visible trail of failures.

2. Software markets defeat the static categories antitrust depends on

Steven’s historical puzzle is that computing “swallowed the economy” largely without licensing, hearings or prior approval. Outside IBM’s long antitrust case and FCC approval for computers containing radios, neither software engineers nor software products required licenses—an extraordinary exception in a society that licenses haircuts, makeup and massage.
Antitrust doctrine, by contrast, arose from tangible bottlenecks such as railroads, railcars and coal. Modern analysis still assumes a definable market, measurable shares and a fixed roster of competitors before turning those assumptions into an HHI calculation. But what is the word-processing market: dedicated applications, everywhere people type, or only software that prints?
Erik’s disruptive-innovation example is the early iPhone camera: inferior to a DSLR on image quality, yet ubiquitous, internet-connected, programmable and effectively zero incremental cost. It was not initially recognized as a camera competitor, illustrating how a product can enter along an unexpected axis before becoming a substitute.
Competition is also a fuzzy set: Google and Apple contest operating systems, Google and Facebook contest advertising, while Facebook and Apple contest headsets. Formulas can therefore become “a substitute for judgment” or a mask for animus, especially when regulators treat “millionaires and billionaires” as adjacent categories despite the 1,000-fold difference.

3. Networks grew from applications into rival systems of governance

Balaji’s framework is “network versus state.” The internet is invisible but upstream of phones, AI, drones, social media and elections—“perhaps the most popular thing in the world”—yet is rarely treated as an independent actor because its infrastructure sits directly in front of everyone without appearing tangible.
As networks reached billions of users, they displaced practical regulatory functions. Uber and Lyft regulate drivers through real-time tracking and bilateral ratings; YouTube and Facebook regulate speech through platform rules. What technology builders viewed as peaceful expansion looked to state actors like lost market share in authority, so “the empire struck back.”
Running a platform complicated Balaji’s own libertarian instincts. A system administrator cannot negotiate millions of defaults and algorithmic parameters with every user; someone must “flip a switch.” That helped him understand Elizabeth Warren’s governing mindset, but he retains one distinction: Apple, Microsoft and Google face competition and user exit, whereas Washington offers no comparably practical switching option.
Steven sees that collision in Europe’s Digital Markets Act. Apple deliberately constrained iPhone APIs, drivers and system access after learning how PCs accumulated privacy and security failures; Europe then demanded PC-like openness while also demanding security. Apple’s answer, in Steven’s paraphrase: “We did that once already.”

4. M&A is a portfolio of speculative bets with power-law returns

Steven’s starting premise is unfriendly to corporate dealmakers: the business literature repeatedly finds M&A a net destroyer of value. He encountered that difficulty firsthand at Microsoft, where acquisitions disrupted Bill Gates’s demands for synchronization across products. Regulators nevertheless focus on blocking deals rather than on the statistically more common risk that buyers destroy them.
His visual exhibit was the New York Times front page after Google bought YouTube: “Dot-com boom is echoed in the deal for YouTube,” followed by concerns about copyright, kitten videos, a tiny team and overpayment. Today’s retrospective certainty removes precisely what Google was buying—the uncertain potential to solve those problems.
Instagram was an even cleaner venture bet. It had no revenue, had raised at a $500 million valuation the previous day, and Zuckerberg offered $1 billion—roughly 25% of Facebook’s $4 billion cash balance—weeks before Facebook’s IPO, without consulting the board first. Contemporary reactions called it a “huge waste of cash,” not an obvious monopoly foreclosure.
Steven estimates big companies choose the wrong potential roughly 90% of the time, often believing their salesforce or platform can rescue a weak asset; HP–Autonomy is his specimen. Retrospective enforcement treats the rare transformed winner as predetermined while erasing the initial possibility that it could have become another failed acquisition.

5. Acquisition math favors huge buyers and category leaders

Balaji’s rule of thumb is that an acquirer should be about 100 times the target’s size. At only 10 times, the purchase consumes roughly 10% of the buyer’s equity—possibly more than several years of spending—and demands company-wide conviction. His exceptional case was Illumina’s acquisition of Solexa, whose sequencing technology became foundational.
That asymmetry explains why startup-to-startup combinations usually disappoint: neither side has the money to make the combination work. A shutdown followed by the team joining the survivor can work, but “M doesn’t work, but A works”—mergers usually fail where genuine acquisitions sometimes succeed.
Smart buyers value a target as product multiplied by the acquirer’s distribution. Weak buyers instead assume “magical distribution beans” can turn the cheaper number-two or number-three asset into the leader. Steven’s failure list includes Microsoft’s $5 billion aQuantive purchase, Google–Motorola, Sprint–Nextel, Microsoft–Nokia and chip companies buying second-tier modem suppliers.

6. Acquisition proceeds finance the next wave of challengers

The well-meaning case for blocking deals—force startups to become independent competitors—ignores businesses that prove a technology but cannot finance a standalone path to IPO. Balaji offers Oculus as the specimen: heavy cash burn made a deep-pocketed acquirer more plausible than continued independent operation.
A large acquisition is also a public surrender by the incumbent: Google already had Google Video but still paid roughly $1.6 billion for YouTube. That payout tells founders the category is valuable and “throws fertilizer” on the field; Instagram’s billion-dollar exit helped inspire Snapchat, TikTok and many other attacks on Facebook.
Steven adds the internal reality behind apparent “tying.” When an adjacent feature emerges, half the company argues it can be built and half argues it must be bought, all with different levels of hysteria. Regulatory discovery can then elevate the most alarmist sales memo or engineering email into evidence of an exclusionary strategy.

7. The best acquisitions can import technical DNA, not revenue

Steven’s transformative Microsoft example is FrontPage, acquired around 1996. Office first fought Internet Explorer internally over ownership of the deal, then competed with Netscape for the Boston company—an illustration of how much disagreement hides behind a transaction later described as a single corporate decision.
FrontPage let users edit a website on a PC, press a button and publish it to an internet server. The product never became Microsoft’s defining franchise, but its people taught the company that web editing was not simply Word: it required HTML, scripting, programming and browser-native publishing.
That imported expertise helped Microsoft understand dynamic websites and in-browser editing. Regulators focused on a static pie miss this reinvention function, Steven argues, because technology companies assume their current products have a finite “sell-by date”; a Windows company cannot safely resolve to make Windows forever.

8. Antitrust pressure invented the “acqui-fire”

Balaji distinguishes three structures. A conventional acquisition buys the company and distributes consideration through its capitalization table and liquidation waterfall. An acquihire usually leaves the acquired company with little or no money, though some cash may go to investors, but gives the team jobs and the reputational line that it was acquired: “status but not the money.”
The new “acqui-fire” reverses that bargain: a large company buys selected researchers and engineers, while the corporate entity remains funded and intact—“the money but not the status.” Balaji groups Scale, Character, Inflection, Adept, Covariant and Windsurf as related, though not identical, examples.
Steven treats this as a familiar regulatory cycle, not unprecedented rule-breaking. NOW accounts routed checking balances through interest-bearing savings, while MCI’s Friends & Family pricing turned deregulated long-distance service into a referral network. Like intentionally fouling with seven seconds left in basketball, an unintended loophole eventually becomes ordinary strategy.

9. Windsurf revealed that status is real transaction consideration

In Balaji’s account, approximately 40 Windsurf employees went to Google, roughly 200 remained, and Google left more than $100 million in the company—money intended to approximate what stakeholders would receive through a liquidation waterfall. Most of the remaining employees had been hired within months as salespeople whom Google, with its own sales organization, did not want.
Because the arrangement could not be described as an acquisition, those negotiating it could not clearly explain it. The visible story became that the founders abandoned everyone and “broke the social contract”; Balaji’s alternative interpretation is that regulatory constraints forced a structure whose cash component survived but whose face-saving acquisition status disappeared.
The remaining organization then rationally sought that status through Cognition. But Balaji flags an integration problem rather than condemning either side: a roughly 60-person company taking on about 200 people is challenging under his size rule, while Cognition’s promise to take everyone may make later restructuring especially difficult.
His proposed fix is a contractual “non-keyman” or designated survivor: an executive paid to remain, distribute funds and conduct an orderly shutdown after an acqui-fire. The joke is Kiefer Sutherland inheriting the presidency after everyone else is “raptured,” but the governance point is serious—future financing documents must define whether decapitation counts as an exit.

10. AI makes seemingly irrational talent deals strategically coherent

Steven says “with near certainty” that AI is a platform shift, even though its winners and relationship to mobile and cloud remain unknown. The nearest-term competitive energy therefore concentrates in tooling, much as DOS and BASIC consolidated a fragmented PC operating-system market after Microsoft and IBM invested irrationally more than independent tool vendors could justify.
Tooling may never be the largest standalone business, but every platform contender needs it, producing many alternatives and strange-looking valuations. Balaji adds that AI enables “more with less,” stratifying companies around highly productive top talent and increasing the value of selective talent extraction.
Balaji warns that large companies may acquire “a taste for the One Ring.” A decapitation can be faster than M&A—closer to a giant purchase order, with money “left in the company”—and successful incumbents may repeat it. Steven expects regulators to scrutinize these deals and eventually change what structures are permitted.

11. Legal and physical constraints could reverse US AI progress

Balaji rejects the slogan that today’s AI is “the worst it’ll ever be.” Napster, he argues, was initially the best version of itself before copyright litigation degraded it; Google Books similarly went from extraordinary access to limited snippets. AI might also worsen if lawsuits, political hostility and energy scarcity collectively constrain training and deployment.
China’s Kimi, Qwen and DeepSeek are, in Balaji’s description, good and rapidly distributed models. He carefully calls them “open weights,” not fully open source, because the complete source and training machinery are unavailable. Restrictions might begin reasonably with hosted DeepSeek sending data to China, then expand to prohibitions on using Chinese open-weight models.
His adverse scenario is cumulative: American companies face copyright suits, data centers hit spare-energy limits, Chinese weights remain available, and US rules prohibit their use. AI also does “middle-to-middle,” not end to end, but that still reaches lawyers, doctors, teachers, artists and journalists—building a political axis of “futurism and primitivism.”
Erik closes by saying the US innovation trajectory faces arrows from technology, regulation, business practices, immigration and research funding. Steven says China’s release of open-weight models is aimed at a rigid American model that is cloud-hosted by a few large players and closed-source. Erik frames China’s strategy as “commoditizing our strength,” while Steven compares it to Google releasing Google Docs for free. Balaji adds that Microsoft ultimately relied on enterprise distribution lock-in, and that copyright and intellectual property also helped create Microsoft, Intel and Apple.

12. Predictable markets and jurisdictional competition are the exits

Asked what Lina Khan should have done, Steven answers that regulators cannot credibly predict the rare successful transaction when M&A “will almost certainly fail.” Markets are better placed to price that uncertainty; imposing investment conditions around an imagined winner is “rent control on investing.”
Balaji’s answer is more macro: technology should stop reacting case by case and write its ideal laws in advance. Use AI for a first draft, adapt model legislation for all 50 states and 190 sovereign countries, rank jurisdictions by receptiveness, then build a sales organization to approach their politicians.
The offer would pair legislation with commitment: 10 CEOs, founders or investors—or a broader group representing stated revenue or AUM—would promise investment if a jurisdiction passes the bill. “Small states are the friends of little tech” because they cannot take growth for granted and may welcome activity that larger governments obstruct.
His imagined “Elon Salvador” would let builders operate at “the speed of physics, not permits,” while retaining enduring prohibitions on assault and murder rather than indiscriminately deleting law. The final inversion is to treat the US federal government as a monopoly and build jurisdictional competition around it: 96% of the world is non-American, and jurisdictional choice exists within the US as well.

Jake Sullivan on Navigating AI Uncertainty and Managing Competition with China GPT-5 Arrives, and We Try the New Alexa+