Kamala Harris Pulls Well Ahead of Trump on Polymarket

This week in prediction markets:

U.S. Vice President Kamala Harris now has a seven-point lead over Donald Trump on Polymarket, after flipping the prediction market odds on Aug. 9 and tying the day before.

“Yes” shares for Harris were trading at 52 cents on the crypto-based betting platform Monday during U.S. morning hours, meaning the market sees a 52% chance she will win the presidency. Each share pays out $1 in USDC, a stablecoin, or cryptocurrency that trades at par with the U.S. dollar, if the prediction comes true, and zero if not. Trump shares were changing hands at 45 cents.

Less than a month ago, before President Joe Biden dropped out of the race and endorsed Harris, Trump’s odds were as high as 72%.

Current levels put Harris at a premium when compared to polling aggregates. Real Clear Polling’s average of polls has Harris with a 0.5 percentage point lead. Some mainstream polls still have Trump ahead: A poll sponsored by Harvard conducted at the end of July gives Trump a four percentage point lead, while CNBC’s latest poll that surveyed up until Aug. 4 has Trump with a two-point lead, and Republican-leaning Rasmussen Reports gives Trump a five-point advantage.

Since her ascension to the top of the Democratic ticket, Harris has received glowing coverage in the mainstream press and celebrity endorsements, while Trump’s recent public performances have reportedly rattled his advisors.

On the other hand, the idiosyncrasies of trading may overstate the magnitude of the reversal of fortune on Polymarket.

One active political bettor who goes by the handle CSPTrading.eth said he’s observed panic selling and groupthink in election markets.

“Someone who bought Trump at 70 and keeps seeing the price fall [has been] dumping to get out,” he said in an interview with CoinDesk. “It’s also a bit of a self-fulfilling prophecy because market makers will buy directionally to hedge their exposure to the price movements and then dump into each other.”

CSPTrading.eth points to some of the state-level markets (which party will carry Alaska or Indiana, for example) where one trader will dump a “rapidly depreciating asset” and others follow suit.

“They don’t want to be holding the bag,” he said.

There are arbitrage opportunities on Polymarket’s various balance-of-power contracts, where traders can buy exposure to a Trump presidency for less than what it costs in the main contract, CSPTrading.eth said.

For instance, a contract that asks bettors to put a stake on the possibility of Republicans taking the Presidency and Senate, but the Democrats winning the House – with some models showing the Republicans having a 78% chance of winning the Senate – is currently trading for 12 cents (or a 12% chance) while a contract predicting a Republican sweep is at 28%.

Together these would be trading at 40 cents, compared to the 46 cents Trump is trading for in the main contract.

Elon Musk and Donald Trump are set to take to X’s digital airwaves Monday night U.S. time, and there’s a good chance the former President will mention crypto.

Bettors are giving a 60% chance that Trump will do so during the interview, which seems like a safe bet considering that his appearance at the recent BTC 2024 conference was a sold-out affair with thunderous applause.

There’s a catch: Trump will need to say “crypto” specifically. Not bitcoin, not Ethereum, but rather the umbrella term itself. Any usage of the word, including plural or possessive forms, will count toward the market’s resolution, the rules read. Compound words including “crypto” referring to decentralized currencies will also count.

Bettors are also giving “MAGA,” “censor” or “censorship,” “Tesla,” and “illegal Immigrant” a high chance of being said.

Meanwhile, the former President also returned to X after a nearly year-long hiatus of tweeting. His last post on Musk’s social media site was from Aug. 24, 2023 and featured his mugshot.

Two contracts asked bettors about Trump’s return to X, with one asking whether he would tweet again before the election and another asking if he would do so by the end of September.

The former had nearly $150,000 in volume with a user going by the handle “BarronTrump” holding the most shares on the yes side, winning over $9,000. This trader has also taken a no position currently valued at $5,600 on a contract about Kamala Harris winning the election.

A user holding the second-largest position on the Trump-Twitter question made off with $8,500. This trader also holds a sizable “yes” stake on the Harris side of the election question, currently worth over $35,000.

On the smaller contract, which asked if Trump would return to X by Sept. 30, only two bettors held the “yes” side with $34 between them. On the “no” side, the largest user lost just over $5,100 – and only seems interested in betting on this contract and one about whether the Binance exchange will list the token of a self-described “crypto casino,” Rollbit.

When OpenAI unveiled its text-to-video generative artificial intelligence model Sora earlier this year, the world was amazed. The demo promised that someday video as realistic as a high-end Playstation game could be generated in real-time.

Missing from that initial demo was a launch date.

Bettors on Kalshi, a U.S.-regulated platform where bets are settled in dollars, are giving it a 38% chance of Sora being launched before 2025, down from 75% when the contract was initiated in May.

One of the problems OpenAI may be running into is the intense computational cost of generating real-time video. As CoinDesk reported earlier this year, research group Fractorial Funds estimates that 720,000 high-end Nvidia H100 GPUs would be required to serve the TikTok and YouTube creative community with AI-generated video.

This amount of silicon would come at a cost of $21.6 billion, stats from the report show, and the total count of cards would be more than Meta, Microsoft, Google, Amazon, Oracle and Tencent – some of Nvidia’s largest customers – have collectively.

Bettors may also be factoring in that AI’s massive demand for electricity could be straining the grid, and testing the political will to build more data centers.

Research from SemiAnalysis shows that AI’s demand for server capacity is expected to double, with global data center power needs jumping from 49 gigawatts in 2023 to 96 gw by 2026. This projected steep jump in power demands has meant that even Texas, which has historically embraced electricity-intensive bitcoin mining and data centers, is souring on the idea of building out more capacity.

One hedge to all of this: China’s ByteDance recently released its own version of a text-to-video generator, only available to users in China. There are limits to the amount of video each user can generate, but it does show that this whole thing isn’t impossible.