All-In's 2026 Predictions

Why does it matter? Because the assumptions most investors, founders, and policymakers carry into 2026 are about to get stress-tested simultaneously.

The All-In crew sat down to make their 2026 predictions — and the throughline isn't any single call. It's that the macro environment is shifting fast enough that old frameworks are liabilities. GDP could print 5-6%. The AI coding breakthrough just crossed a new threshold. California may be actively destroying its own tax base. And the M&A playbook just got rewritten overnight.

• California's proposed 5% wealth tax has a lethal quirk for founders: super-voting shares could make the effective rate 25%+, with no mechanism to pay it without bankrupting your own company
• The IP-license deal structure has replaced traditional M&A in AI — antitrust makes real acquisitions a 3-year slog, so expect hundreds of billions in licensing transactions in 2026
• Chamath's wildcard: SpaceX doesn't IPO — it reverse-merges into Tesla, consolidating Elon's empire into one cap table
• Enterprise SaaS faces structural collapse as AI destroys the 90% of revenue that comes from maintenance and migration, not licensing

California's wealth tax could bankrupt the founders it taxes — and half the tax base may already be gone

Chamath put a number on what's already happened: friends who have explicitly left California represent "about half a trillion of net worth, which I think is very bad for the long-term budget of California." That's before the ballot initiative even qualifies.

The mechanism is worse than most people realize. Sacks walked through the super-voting stock provision: the tax doesn't use liquid market value for shares with outsized voting rights — it multiplies your voting percentage by the full market cap of the company. Larry and Sergey each control roughly 26% of Google's votes; Google is worth ~$4.5 trillion. Their actual net worth is approximately $100 billion each. Under this formula, each would be deemed worth roughly $1 trillion — turning a 5% annual tax into something closer to a 25% levy on actual net worth. And to generate that cash, they'd need to sell more than double, because they'd owe capital gains on the sale itself.

For private company founders, it's even more brutal. If you're "stuck in success with a bunch of illiquid stock and have no way to pay 5% of that value, you're going to bankrupt your own company." If the company then goes to zero the following year, you still owe the prior year's bill.

Polymarket currently prices the initiative making the ballot at 69% — spiked from 45% after progressive politicians amplified it. Sacks and Freeberg both believe it probably doesn't pass if it gets there, but the damage may be irreversible either way: even a failed 2026 initiative signals a 2028 version is coming, which is what pushed Sacks to finally leave. The rush for the exits, he argues, will happen the moment signatures are certified — not after election day.

The agentic coding breakthrough is the ChatGPT moment for software — and it's happening right now

Something shifted in the last few weeks of 2025. Sacks was careful not to over-index on hype, but his read is unambiguous: "there seems to have been another level of quality achieved just in the last month or so."

This isn't just better autocomplete. The new tools have file system access, can take direct actions on your computer, and combine coding capability with general tool use into something that starts to look like an autonomous agent. The leap, Sacks argues, is comparable to late 2022 going into 2023 — when chatbots went from a curiosity to infrastructure. "This trend feels to me like chatbots did at the end of 2022 going into 23 where it's like people were really hyped about it but then continue to play out in the next year."

The implication isn't just productivity. If adoption follows the LLM curve, enterprise software budgets, developer headcount assumptions, and SaaS valuations all need simultaneous repricing. The window to position ahead of that repricing — as infrastructure, not accessory — is now, before the adoption curve becomes consensus.

Traditional M&A is dead in AI — the IP-license deal is the new acquisition, and it will generate hundreds of billions in 2026

Three years of antitrust review killed the old playbook. Chamath's framing was direct: "if any of those companies tried to buy, let's just say Anthropic, I think it's three years of antitrust minimum" — worse than Microsoft/Activision, which took nearly two years and required navigating the EU, UK, and China simultaneously.

So deals are happening differently. Google/Character AI, Microsoft/Inflection, Nvidia/Grok, and the Scale AI deal Chamath was involved in all follow the same structure: IP license plus talent, effective immediately. "The next day somebody like Sundep can be working for Jensen, which is what Jensen wants. They want the talent" — and the IP. No waiting for regulators.

The structure is still being refined. Chamath noted that the third or fourth iteration was already significantly better than the first. "By the time that you're into the middle part of next year and you've done 15 or 20 of these things, I think the lawyers that work on these things and the accountants will just be" dialed in. Tax treatment isn't ideal, but execution speed is transformational.

Jason floated the $50B+ threshold — a Mag-7 going for XAI, Mistral, Perplexity, or Anthropic. Chamath agreed that amount is coming, just almost certainly not as a traditional acquisition. Model this as the new M&A standard, not the exception.

SpaceX won't IPO independently — Chamath's contrarian call is a Tesla reverse merger that rewrites Elon's cap table

This is the wildcard that stopped the conversation cold. Chamath's prediction: "I don't think SpaceX will IPO. I think that it will reverse merge into Tesla and I think Elon will use it as a moment to consolidate control and power of his two seminal assets into one cap table."

Musk has talked publicly about a holding company structure for years. The logic is straightforward: a reverse merger lets Tesla shareholders get direct exposure to SpaceX's satellite and launch economics without paying a private-market premium, while Musk gains unified governance across his two most valuable assets. Neuralink, the Boring Company — all of it could eventually roll in.

For public markets, this would be the single biggest re-rating event in years. Tesla's current valuation already contains some implicit SpaceX optionality; a formal merger would force every model to price it explicitly. Worth tracking as a genuine tail risk, not a fantasy scenario.

Jevons paradox says AI creates more knowledge worker jobs than it destroys — and the historical data backs Sacks up

The entire AI-unemployment narrative may be running backward. Sacks invoked Jevons paradox — formally articulated in Erin Levy's post — to make the case: "as the cost of a resource goes down, the aggregate demand for it actually increases because you discover more and more use cases."

The radiology example is the sharpest rebuttal to the job-loss story. AI was supposed to replace radiologists. Instead, the number of radiologists is increasing — because cheaper, more accessible scans mean more people want them, and you still need a doctor to prompt, interpret, and validate the AI's output. More efficiency, more volume, net more jobs.

The same logic applies to code. Software generation was historically constrained by the scarcity and cost of engineers. As that cost collapses, the amount of software the economy wants to generate will expand massively — creating new categories, new use cases, new demand for people who can work alongside the tools. "We're going to look back and see that the job loss narrative was not only wrong but we actually got job gains."

Jason pushed back with ground-level data — a roundtable of 50 CEOs suggested junior hiring is down partly due to cultural and capability gaps in recent graduates, not purely AI. Both things, he acknowledged, are probably true. The smarter read: companies are cutting the bottom third of tasks, then rehiring people who can use the tools. The winners are AI-augmented workers, not AI-resistant ones.

Copper goes parabolic — a 70% global supply shortfall by 2040 makes it the most structurally scarce material on earth

By 2040, the world will be short approximately 70% of required copper supply "at current course and speed." That's Chamath's number, and it's not a cyclical forecast — it's a structural deficit measured in decades.

The demand drivers are converging from every direction: data centers, semiconductor fabrication, weapons systems, EV infrastructure, grid buildout. Copper is "at least as it stands today the most useful, cheap, amendable, conductive material that we have" — and there's no near-term substitute. The Trump-era national security frame amplifies the problem: unilateralism means domestic supply chains become strategic assets, which means government policy will accelerate demand even faster than the private market would on its own.

"The asset that is set up to go absolutely parabolic is copper." Chamath picked a basket of critical metals as his best-performing asset for 2026, with copper as the anchor. Deficits of this magnitude take 15-20 years to close through new mine development — meaning the structural long thesis has a long runway regardless of what happens in any single year.

Enterprise SaaS loses its profit engine in 2026 — AI agents are doing the migration and maintenance work that represents 90% of the sector's revenue

ServiceNow fell 30%. Workday dropped 18%. Dropbox down 23%. All of this while the S&P gained 17%. Chamath's thesis from last year is now visible in the stock charts, and he's doubling down.

The $3-4 trillion enterprise SaaS economy breaks into three buckets: initial licensing (5-10% of revenue), maintenance, and migration. "Those last two buckets represent 90% of all the dollars in revenue that's generated in software." AI agents can now do that work. The incumbents' entire growth model — land the client, expand as they add headcount — collapses simultaneously from both ends: fewer seats being added, and the recurring services revenue getting automated away.

Chamath is speaking from direct experience: "Our entire 8090s business has basically migrated to disrupting maintenance and migration patterns. I cannot describe how much opportunity there is." The work is unglamorous — moving legacy data, maintaining bulky $300M licenses — but it's where all the money lives, and AI is eating it. This isn't a slow secular shift; the stock charts are already reflecting an accelerating disruption.

Iran's fall might destabilize the Middle East more, not less — Freeberg's contrarian case for intra-Gulf conflict

Almost everyone assumes Iran's removal is unambiguously good for regional stability. Freeberg disagrees, and the logic is uncomfortable.

The premise he accepts: the Ayatollahs are likely out in 2026. Demographics, economic crisis, street pressure — the transition seems underway. But the assumption that Iran was purely a destabilizing force may be exactly wrong. Freeberg's contrarian read: "the contrarian point may in fact be that Iran has been a stabilizing force in that region" — and removing it opens a power vacuum that UAE, Saudi Arabia, and Qatar will compete to fill.

The conflict vector he's watching is the Palestinian question. As a two-state solution emerges, someone has to govern, fund, and politically own the outcome. Jordan, Egypt, Saudi — each has different interests and resources. "There's going to be this battle for who's going to take care of the Palestinians as the two-state solution emerges," with resource allocation disputes layered on top.

For investors pricing Middle East risk purely around Iran's internal transition, this is a blind spot. An intra-Gulf conflict scenario — even a cold one — could spike oil prices and defense spending regardless of what happens inside Iran's borders.

The 2026 prediction that ties everything together: the US economy is a coiled spring, and most people's models don't know it yet

Five to six percent GDP growth would match China's command-economy peak — achieved under democracy and capitalism. Chamath's lower bound is 5%, upper bound 6%. Sacks is at 5%. Freeberg at 4.6%. The Atlanta Fed's Q4 2025 GDP estimate just jumped to 5.4% in a single revision.

The bet isn't just on tax cuts or deregulation. It's that closing the border reset non-farm payroll baselines, AI productivity gains are compounding, accelerated depreciation is triggering a capital equipment supercycle, and rate cuts are still coming. Every variable is pointing the same direction at the same time.

If they're right, every asset class, political cycle, and employment model built on the assumption of 2-3% growth needs to be rebuilt from scratch.

---

Topics: California wealth tax, venture capital, AI coding tools, enterprise SaaS disruption, copper commodities, SpaceX Tesla merger, Jevons paradox, M&A deal structures, 2026 predictions, Middle East geopolitics, US GDP growth, IPO market